// file      : doc/packaging.cli
// license   : MIT; see accompanying LICENSE file

"\name=build2-packaging-guide"
"\subject=toolchain"
"\title=Packaging Guide"

// NOTES
//
// - Maximum <pre> line is 70 characters.
//
// - In guideline titles (do/don't) omit a/the.
//

// @@ Close the issue in WISHLIST.

"
\h0#preface|Preface|

This document provides guidelines for converting third-party C/C++ projects to
the \c{build2} build system and making them available as packages from
\l{https://cppget.org cppget.org}, the \c{build2} community's central package
repository. For additional information, including documentation for individual
\c{build2} toolchain components, man pages, HOWTOs, etc., refer to the project
\l{https://build2.org/doc.xhtml Documentation} page.

\N|This document is a work in progress and is incomplete.|


\h1#intro|Introduction|

@@ Assume read through toolchain introduction and build system introduction.
Also, ideally, have some experience using \c{build2} in your own projects.

The aim of this guide is to ease the convertion of third-party C/C++ projects
to the \c{build2} build system and publishing them to the
\l{https://cppget.org cppget.org} package repository by codifying the best
practices and techniques. By following the presented guidelines you also make
it easier for others to review your work and help with ongoing maintenance.

The primary focus of this guide are existing C/C++ projects that use a
different build system and that are maintained by a third-party, which we will
refer to as \i{upstream}. Unless upstream is willing to incorporate support
for \c{build2} directly into their repository, such projects are normally
packaged for \c{build2} in a separate \c{git} repository under the
\l{https://github.com/build2-packaging github.com/build2-packaging}
organization. Note, however, that many of the presented guidelines are also
applicable when converting your own projects (that is, where you are the
upstream) as well as projects that use languages other than C or C++.

Most C/C++ packages that are published to \l{https://cppget.org cppget.org}
are either libraries or executables (projects that provide both are normally
split into several packages) with libraries being in the strong majority.
Libraries are also generally more difficult to build correctly. As a result,
this guide uses libraries as a baseline. In most cases, a library-specific
step is easily distinguished as such and can be skipped when dealing with
executables. And in cases where a more nuanced change is required, a note will
be provided.

At the high-level, packaging a third-party project involves the following
steps:

\ol|

\li|Create the \c{git} repository and import upstream source code.|

\li|Generate \c{buildfile} templates that match upstream layout.|

\li|Tweak the generated \c{buildfiles} to match upstream build.|

\li|Test using the \l{https://ci.cppget.org \c{build2} CI service}.|

\li|Publish the package to \l{https://cppget.org cppget.org}.|

|

Once this process is completed and the package is published, new releases
normally require a small amount of work provided there are no drastic changes
in the upstream layout or build. The sequence of steps for a new release would
typical look like this:

\ol|

\li|Add new and/or remove old upstream source code, if any.|

\li|Tweak \c{buildfiles} to match changes to upstream build, if any.|

\li|Test using the \l{https://ci.cppget.org \c{build2} CI service}.|

\li|Publish the package to \l{https://cppget.org cppget.org}.|

|

While packaging a simple library or executable is relatively straightforward,
the C and C++ languages and their ecosystem is famous for a large amount
varience in the platforms, compilers, and build systems used. This leads to
what appears to be an endless list of special considerations that are
applicable in certain, more complex cases.

As result, the presented guidelines are divided into four chapters: The
\l{#core Core Guidelines} cover steps that are applicable to all or most
packaging efforts. As mentioned earlier, these steps will assume packaging a
library but they should be easy to adapt to executables. This chapter is
followed by \l{#dont-do What Not to Do} which covers the common packaging
mistakes and omissions. These are unfortunately relatively common because
experience with other build systems often does not translate directly to
\c{build2} and some techniques (such as header-only libraries) are
discouraged. The last two chapters are \l{#howto HOWTO} and \l{#faq FAQ} which
cover the above-mentioned long list of special considerations that are only
applicable in certain cases as well as answer frequent packaging-related
questions, respectively.

@@ Purpose of notes to provide rationale.

Besides the presented guidelines you may also find the existing packages found
in \l{https://github.com/build2-packaging github.com/build2-packaging} a good
source of example material. The repositories pinned to the front page are the
recommended starting point.

\h#intro-term|Terminology|

upstream
upstream repository
project
package (third-party project)
package \c{git} repository
multi-package repository


\h1#core|Core Guidelines|

\h#core-repo|Setup the package repository|

This section covers the creation of the package \c{git} repository and
the importation of the upstream source code.

\h2#core-repo-exists|Check if package repository already exists|

Before deciding to package a third-party project you have presumably checked
on \l{https://cppget.org cppget.org} if someone has already packaged it. There
are several other places that make sense to check as well:

\ul|

\li|\l{https://queue.cppget.org queue.cppget.org} contains packages that
have been submitted but not yet published.|

\li|\l{https://queue.stage.build2.org queue.stage.build2.org} contains
packages that have been submitted but can only be published after the next
release of the \c{build2} toolchain (see \l{#faq-publish-stage Where to
publish if package requires staged toolchain?} for background).|

\li|\l{https://github.com/build2-packaging github.com/build2-packaging}
contains all the third-party package repositories. Someone could already be
working on the package but haven't they finished it.|

\li|\l{https://github.com/build2-packaging/WISHLIST/issues
github.com/build2-packaging/WISHLIST} contains as issues projects that people
wish were packaged. These may contain offers to collaborate or announcements
of ongoing work.||


In all these cases you should be able to locate the package \c{git} repository
and/or connect with others in order to collaborate on the packaging work. If
the existing effort looks abandoned (for example, there hasn't been any
progress for a while and the existing maintainer doesn't respond) and you
would like to take over the package,
\l{https://build2.org/community.xhtml#help get in touch}.


\h2#core-repo-name|Use upstream repository name as package repository name|

It is almost always best to use the upstream repository name as the package
repository name. If there is no upstream repository (for example, because the
project doesn't use a version control system), the name used in the source
archive distribution would be the natural fallback.

\N|See \l{#core-package-name Decide on the package name} for the complete
picture on choosing names.|

\h2#core-repo-create|Create package repository in personal workspace|

For a third-party project, the end result that we are aiming for is a package
repository under the \l{https://github.com/build2-packaging
github.com/build2-packaging} organization.

\N|We require all the third-party projects that are published to
\l{https://cppget.org cppget.org} to be under the
\l{https://github.com/build2-packaging github.com/build2-packaging}
organization in order to ensure some continuity in case the original
maintainer loose interest, etc. You will still be the owner of the repository
and by hosting your packaging efforts under this organization (as opposed to,
say, your personal workspace) you make it easier for others to discover your
work and to contribute to the package maintenance.

Note that this requirement does not apply to your own projects (that is, where
you are the upstream) and where the \c{build2} support is normally part of the
upstream repository.

Finally, a note on the use of \c{git} and GitHub: if for some reason you are
unable to use either, \l{https://build2.org/community.xhtml#help get in touch}
to discuss alternatives.|

However, the recommended approach is to start with a repository in your
personal workspace and then, when it is ready or in a reasonably stable shape,
transfer it to \l{https://github.com/build2-packaging
github.com/build2-packaging}. This gives you the freedom to make destructive
changes to the repository (including deleting it and strating over) during the
initial packaging work. It also removes the pressure to perform: you can give
it a try and if things turn out more difficult than you expected, you can
just drop the repository.

\N|For repositories under \l{https://github.com/build2-packaging
github.com/build2-packaging} the \c{master}/\c{main} branch is protected: it
cannot be deleted and its commit history cannot be overwritten with a forced
push.|

\N|While you can use any name for a repository under the personal workspace,
under \l{https://github.com/build2-packaging github.com/build2-packaging} it
should follow the \l{core-repo-name Use upstream repository name as package
repository name} guideline. In particular, there should be no prefixes like
\c{build2-} or suffixes like \c{-package}. If the repository under your
personal workspace does not follow this guideline, you should rename it before
transferring it to the \l{https://github.com/build2-packaging
github.com/build2-packaging} organization.|

There is one potenential problem with this approach: it is possible that
several people start working on the same third-party project without being
aware of each other's efforts. If the project you are packaging is relatively
small and you don't expect it to take more than a day or two, then this is
probably not worth worrying about. For bigger projects, however, it makes
sense to announce your work by creating (or updating) the corresponding issue
in \l{https://github.com/build2-packaging/WISHLIST
github.com/build2-packaging/WISHLIST}.

To put it all together, the recommended sequence of actions for this step:

\ol|

\li|Create a new empty repository under your personal workspace from GitHub
    UI. Don't automatically add any files (\c{README}, \c{LICENSE}, etc).|

\li|Set the repository description in GitHub UI to the \c{build2 package
for <name>} line, where \c{<name>} is the project name.|

\li|Clone the repository to your machine.||

\N|Since this is your personal repository, you can do the initial work
directly in \c{master}/\c{main} or in a separate branch, it's up to you.|

As a running example, let's assume we want to package a library called \c{foo}
whose upstream repository is at \c{https://github.com/<upstream>/foo.git}. We
have created its package repository at
\c{https://github.com/<personal>/foo.git} (with the \c{build2 package for foo}
description) and can now clone it:

\
$ git clone https://github.com/<personal>/foo.git
\


\h2#core-repo-init|Initialize package repository with \c{bdep new --type empty}|

Change to the root directory of the package repository that you have clonned
on the previous step and run (continuing with our \c{foo} example):

\
$ cd foo
$ bdep new --type empty
$ tree .
./
├── .gitattributes
├── .gitignore
├── README.md
└── repositories.manifest
\

This command creates a number of files in the root of the repository:

\dl|

\li|\n\c{README.md}\n

This is the project \c{README}. We will discuss the recommended content for
this file later.|

\li|\n\c{repositories.manifest}\n

This file specifies the repositories from which this project will obtain its
dependencies (see \l{intro#guide-add-remove-deps Adding and Removing
Dependencies}). If the project you are packaging has no dependencies, then you
can safely remove this file (it's easy to add later if this changes). And for
projects that do have dependecies we will discuss the appropriate changes to
this file later.|


\li|\n\c{.gitattributes} and \c{.gitignore}\n

These are the \c{git} infrastrucutre files for the repository. You shouldn't
normally need to change anything in them at this stage (see the comments
inside for details).||

Next add and commit these files:

\
$ cd foo/ # Change to the package repository root.
$ git add .
$ git status
$ git commit -m \"Initialize repository\"
\

\N|In these guidelines we will be using the package repository setup that is
capable of having multiple packages. This is recommended even for upstream
projects that only provides a single package because it gives us the
flexibility of adding new packages at a later stage without having to perform
a major restructuring of our repository.

Note also that upstream providing multiple package is not the only reason we
may end up having multiple \c{build2} packages. Another common reason is
factoring tests into a separate package due to a dependency on a testing
framework
(see \l{https://github.com/build2/HOWTO/blob/master/entries/handle-tests-with-extra-dependencies.md
How do I handle tests that have extra dependencies?} for background and
details). While upstream adding new packages may not be very common, upstream
deciding to use a testing framework is a lot more plausible.

The only notable drawback of using a multi-package setup with a single package
is the extra subdirectory for the package and a few extra files (such as
\c{packages.manifest} that lists the packages) in the root of the repository.
If you are certain that the project that you are converting is unlikely to
have multiple packages (for example, because you are the upstream) or need
extra dependencies for its tests (a reasonable assumption for a C project),
then you could instead go with the single-package repository where the
repository root is the package root. See \l{bdep-new(1)} for details on how to
initialize such a repository. In this guide, however, we will continue to
assume a multi-package repository setup.|


\h2#core-repo-submodule|Add upstream repository as \c{git} submodule|

If the third-party project is available from a \c{git} repository, then the
recommended approach is to use the \c{git} submodule mechanism to make the
upstream source code available inside the package repository, customarily in a
subdirectory called \c{upstream/}.

\N|While \c{git} submodules receive much criticism, in our case we use them
exactly as indended: to select and track specific (release) commits of an
external project. As a result, there is nothing tricky about their use for our
purpose and all the relevant commands will be provided and explained, in case
you are not familiar with this \c{git} mechanism.|

Given the upstream repository URL, to add it as a submodule, run the following
command from the package repository root (continuing with our \c{foo} example):

\
$ git submodule add https://github.com/<upstream>/foo.git upstream
\

\N|You should prefer \c{https://} over \c{git://} for the upstream repository
URL since the \c{git://} protocol may not be accessible from all networks.
Naturally, never use a URL that requires authentication, for example, SSH.|

Besides the repository URL, you also need the commit of the upstream release
which you will be packaging. It is common practice to tag releases so the
upstream tags would be the first place to check. Failed that, you can always
use the commit id.

Assuming the upstream release tag you are interested in is called \c{vX.Y.Z},
to update the \c{upstream} submodule to point to this release commit, run the
following command:

\
$ cd upstream
$ git checkout vX.Y.Z
$ cd ..
\

Then add and commit these changes:

\
$ cd foo/ # Change to the package repository root.
$ git add .
$ git status
$ git commit -m \"Add upstream submodule\"
\

Now we have all the upstream source code for the release that we are
interested in available in the \c{upstream/} subdirectory of our repository.

The plan is to then use symbolic links (symlinks) to non-invasively overlay
the \c{build2} files (\c{buildfile}, \c{manifest}, etc) with the upstream
source code, if necessary adjusting upstream structure to split it into
multiple packages and/or to better align with the source/output layouts
recommended by \c{build2} (see \l{https://build2.org/article/symlinks.xhtml
Using Symlinks in \c{build2} Projects} for background and rationale). But
before we can start adding symlinks to the upstream source (and other files
like \c{README}, \c{LICENSE}, etc), we want to generate the \c{buildfile}
templates that match the upstream source code layout. This is the subject of
the next section.

\N|While on UNIX-like operating systems symlinks are in widespread use, on
Windows it's a niche feature that unfortunately could be cumbersome to use
(see \l{https://build2.org/article/symlinks.xhtml#windows Symlinks and
Windows} for details). However, the flexibility afforded by symlinks when
packaging third-party projects is unmatched by any other mechanism and we
therefore use them despite potentially sub-optimal experience on Windows.|


\h#core-package|Create package and generate \c{buildfile} templates|

This section covers the addition of the package to the repository we have
prepared in the previous steps and the generation of the \c{buildfile}
templates that match the upstream source code layout.


\h2#core-package-name|Decide on the package name|

While choosing the package repository name was pretty straightforward, things
get less clear cut when it comes to the package name.

\N|If you need a refresher on the distinction between projects and packages,
see \l{#intro-term Terminology}.|

Picking a name for a package that provides an executable is still relatively
straightforward: you should use the upstream name (which is usually the same
as the upstream project name) unless there is a good reason to deviate. One
recommended place to check before deciding on a name is the
\l{https://packages.debian.org Debian package repository}. If their package
name differs from upstream, then there is likely a good reason for that and
it is worth trying to understand what it is.

\N|Tip: when trying to find the corresponding Debain package, search for the
executable file name in the package contents if you cannot fine the package by
its upstream name. Also consider searching in the \c{unstable} distribution in
addition to \c{testing} for newer packages.|

Picking a name for a package that provides a library is where things can get
more complicated. While all the recommendation that have been listed for
executables apply equally to libraries, there are additional considerations.

In \c{build2} we recommend (but not require) that new library projects use a
name that starts with \c{lib} in order to easily distinguish them from
executables and avoid any clashes, potential in the future (see
\l{intro#proj-struct Canonical Project Structure} for details). To illustrate
the problem, consider the \c{zstd} project which provides a library and an
executable. In upstream repository both are part of the same codebase that
doesn't try to separate them into packages so that, for example, library could
be used without downloading and building the executable. In \c{build2},
however, we do need to split them into two separate packages and both packages
cannot be called \c{zstd}. So we call them \c{zstd} and \c{libzstd}.

\N|If you are familiar with the Debian package naming policy, you will
undoubtedly recognize the approach. In Debian all the library packages (with
very few exceptions) start with the \c{lib} prefix. So when searching for an
upstream name in the \l{https://packages.debian.org Debian package repository}
make sure to prefix it with \c{lib} (unless it already starts with this
prefix, of course).|

This brings the question of what to do about third-party libraries: should we
add the \c{lib} prefix to the package name if it's not already there?
Unfortunately, there is no clear cut answer and whichever decision you make,
there will be drawbacks. Specifically, if you add the \c{lib} prefix, the main
drawback is that the package name now deviates from upstream name and if the
project maintainer ever decides to add \c{build2} support the upstream
repository, there could be substantial friction. On the other handle, if you
don't add the \c{lib} prefix, then you will always run the risk of a future
clash with an executable name. And, as was illustrated with the \c{zstd}
example, a late addition of an executable won't necessarily cause any issues
to upstream. As a result, we don't have a hard requirement for the \c{lib}
prefix unless there is already an executable that would cause the clash (this
applies even if it's not being packaged yet or is provided by an unrelated
project). If you don't have a strong preference, we recommend that you add the
\c{lib} prefix (unless it is already there). In particular, this will free you
from having to check for any potential clashes. See
\l{https://github.com/build2/HOWTO/blob/master/entries/name-packages-in-project.md
How should I name packages when packaging third-party projects?} for
additional background and details.

To build some intuition for choosing package names, let's consider several
real examples. We start with executables:

\
  upstream  |   upstream    |   Debian   | build2 package|   build2
project name|executable name|package name|repository name|package name
------------+---------------+------------+---------------+------------
byacc        byacc           byacc        byacc           byacc
sqlite       sqlite3         sqlite3      sqlite          sqlite3
vim          xxd             xxd          xxd             xxd
OpenBSD      m4              -            openbsd-m4      openbsd-m4
qtbase 5     moc             qtbase5-\    Qt5             Qt5Moc
                             dev-tools
qtbase 6     moc             qt6-base-\   Qt6             Qt6Moc
                             dev-tools
\

The examples are arranged from the most straightforward naming to the
least. The last two examples show that sometimes, after carefully considering
upstream naming, you nevertheless have no choice but to ignore it and forge
your own path.

Next let's look at library examples. Notice that some use the same \c{build2}
package repository name as the executables above. That means they are part of
the same multi-package repository.

\
  upstream  |  upstream     |   Debian   | build2 package|   build2
project name|library name   |package name|repository name|package name
------------+---------------+------------+---------------+------------
libevent     libevent        libevent     libevent        libevent
brotli       brotli          libbrotli    brotli          libbrotli
zlib         zlib            zlib         zlib            libz
sqlite       libsqlite3      libsqlite3   sqlite          libsqlite3
libsig\      libsigc++       libsigc++    libsig\         libsigc++
cplusplus                                 cplusplus
qtbase 5     QtCore          qtbase5-dev  Qt5             libQt5Core
qtbase 6     QtCore          qt6-base-dev Qt6             libQt6Core
\

If an upstream project is just a single library, then the project name is
normally the same as the library name (but there are exceptions, like
\c{libsigcplusplus} in the above table). However, when looking at upstream
repository that contains multiple components (libraries and/or executables,
like \c{qtcore} in the above example), it may not be immediately obvious what
the upstream's library names are. In such cases, the corresponding Debian
packages can really help clarify the situation. Failed that, look into the
existing build system. In particular, if it generates the \c{pkg-config} file,
then the name of this file is usually the upstream library name.

\N|Looking at the names of the library binaries is less helpful because on
UNIX-like systems they must start with the \c{lib} prefix. And on Windows the
names of library binaries often embed extra information (static/import,
debug/release, etc) and may not correspond directly to the library name.|

And, speaking of multiple components, if you realize the upstream project
provides multiple libraries and/or executables, then you need to decide
whether to split them into seperate \c{build2} packages and if so, how. Here,
again, the corresponding Debian packages can be a good strating point. Note,
however, that in this case we often deviate from their split, especially when
it comes to libraries. For example, \c{libevent} shown in the above table
provides several libraries (\c{libevent-core}, \c{libevent-extra}, etc) and in
Debian it is actually split into several binary packages along these lines. In
\c{build2}, however, there is a single package that provides all these
libraries with everything except \c{libevent-core} being optional. An example
which shows the decision made in a different direction would be the Boost
libraries: in Debian all the header-only Boost libraries are bundled into a
single package while in \c{build2} they are all seperate packages.

The overall criteria here can be stated as follows: if a small family of
libraries provide complimentary functionality (like \c{libevent}), then we put
them all into a single package, usually making the additional functionality
optional. However, if the libraries are independent (like Boost) or provide
alternative rather than complimentary functionality (for example, like
different backends in \c{imgui}), then we make them separate packages. Note
that we never bundle an executable and a (public) library in a single package.

Note also that while it's a good idea to decide on the package split and all
the package names upfront to avoid suprises later, you don't have to actually
provide all the packages right away. For example, if upstream provides a
library and an executable (like \c{zstd}), you can start with the library and
the executable package can be added later (potentially by someone else).

Admittedly, the recommendation in this section are all a bit fuzzy and one can
choose different names or different package splits that could all seem
reasonable. If you are unsure how to split the upstream project or what names
to use, \l{https://build2.org/community.xhtml#help get in touch} to discuss
the alternatives. It can be quite painful to change these things after you
have completed the remaining packaging steps.

Continuing with our \c{foo} example, we will follow the recommendation and
call the library package \c{libfoo}.


\h2#core-package-struct|Decide on the package source code layout|

Another aspect we need to decide on is the source code layout inside the
package. Here we want to stay as close to upstream layout as possible unless
there are valid reasons to deviate. This has the best chance of giving us a
build without any compile errors since the header inclusion in the project can
be sensitive to this layout. This also makes it easier for upstream to adopt
the \c{build2} build.

Sometimes, however, there are good reasons for deviating from upstream,
especially in cases where upstream is clearly following bad practices, for
example including generically-named public headers without the library name as
a subdirectory prefix. If you do decide to change the layout, it's usually
less disruptive (to the build) to rearrange things at the outer levels than at
the inner. For example, it should normally be possible to move/rename the
top-level \c{tests/} directory or to place the library source files into a
subdirectory.

Our overall plan for the package is to create the initial layout and
\c{buildfile} templates automatically using \l{bdep-new(1)} in the
\c{--package} mode, then tweak \c{buildfiles} if necessary, and finally
\"fill\" the package with upstream source code using symlinks.

The main rationale for using \l{bdep-new(1)} instead of doing everything by
hand is that there are many nuances in getting the build right and
auto-generated \c{buildfiles} had years of refinement and fine-tuning. The
familiar structure also makes it easier for others to understand your build,
for example while reviewing your package submission.

The \l{bdep-new(1)} command supports a wide variety of
\l{bdep-new.xhtml#src-layout source layouts}. While it may take a bit of time
to understand the customization points necessary to achieve the desired layout
for your first package, this will pay off in spades when you work on
converting subsequent packages.

And so the focus of the following several steps is to iteratively discover the
\l{bdep-new(1)} command line that best approximates the upstream layout. The
recommended procedure is as follows:

\ol|

\li|\nStudy the upstream source layout and existing build system.|

\li|\nCraft and execute the \l{bdep-new(1)} command line necessary to achieve
the upstream layout.|

\li|\nStudy the auto-generated \c{buildfiles} for things that don't fit and
need to change. But don't rush to start manually editing the result. First get
an overview of the required changes and then check if it's possible to achieve
these changes automatically using one of \l{bdep-new(1)} sub-options. If
that's the case, delete the package subdirectory, and restart from step #2.||

This and the following two sections discuss each of these steps in more detail
and also look at some examples.

The first step above is to study the upstream project in order to understand
where the various parts are (headers, sources, etc.) and how they are built.
Things that can help here include:

\ul|

\li|Read through the existing build system definitions.|

\li|Try to build the project using the existing build system.|

\li|Try to install the project using the existing build system.|

\li|Look into the Debian package contents to see if there are any differences
    with regards to the installation locations.||

For libraries, the first key pieces of information we need to find is how the
public headers are included and where they are installed. The two common
\i{good} practices is to either include the public headers with a library name
as a subdirectory, for example, \c{#include\ <foo/util.h>}, or to include the
library name into each public header name, for example, \c{#include\
<foo_util.h>} or \c{#include\ <foo.h>} (in the last example the header name is
the library name itself, which is also fairly common). Unfortunately, there is
also a fairly common \i{bad} practice: having generically named headers (such
as \c{util.h}) included without the library name as a subdirectory.

\N|The reason this is a bad practice is that libraries that have such headers
cannot coexist, neither in the same build nor when installed. See
\l{intro#proj-struct Canonical Project Structure} for background and details.
See \l{#howto-bad-inclusion-practice How do I deal with bad header inclusion
practice} if you encounter such a case.|

Where should we look to get this information? While the library source files
sound like a natural place, oftentimes they include own headers with the
\c{\"\"} style inclusion, either because the headers are in the same directory
or because the library build arranges for them to be found this way with
additional header search paths. As a result, a better place to look could be
library's examples and/or tests. Some libraries also describe which headers
they provide and how to include them in their documentation.

The way public headers are included normally determines where they are
installed. If they are included with a subdirectory, then they are normally
installed into the same subdirectory in, say, \c{/usr/include/}. Continuing
with the above example, a header that is included as \c{<foo/util.h>} would
normally be installed as \c{/usr/include/foo/util.h}. On the other hand, if
the library name is part of the header name, then the headers are usually (but
not always) installed directly into, say, \c{/usr/include/}, for example as
\c{/usr/include/foo_util.h}.

\N|While these are the commonly used installation schemes, there are
deviations. In particular, in both cases upstream may choose to add an
additional subdirectory when installing (so the above examples we instead end
up with, say, \c{/usr/include/foo_v1/foo/util.h} and
\c{/usr/include/foo_v1/sub/foo_util.h}). See
\l{#howto-extra-header-install-subdir How do I handle extra header
installation subdirectory} if you encounter such a case.|

The inclusion scheme would normally be recreated in the upstream source code
layout. In particular, if upstream includes public headers with a subdirectory
prefix, then this subdirectory would normally also be present in the upstream
layout so that such a header can be included form the upstream codebase
directly. As an example, let's say we determined that public headers of
\c{libfoo} are included with the \c{foo/} subdirectory, such as
\c{<foo/util.hpp>}. One of the typical upstream layouts for such a library
would look like this:

\
$ tree upstream/
upstream/
├── include/
│   └── foo/
│       └── util.hpp
└── src/
    ├── priv.hpp
    └── util.cpp
\

Notice how the \c{util.hpp} header is in the \c{foo/} subdirectory rather
than in \c{include/} directly.

The second key pieces of information we need to find is whether and, if so,
how the public headers and sources are split. For instance, in the above
example, we can see that public headers go into \c{include/} while sources and
private headers go into \c{src/}. But they could also be combined in the same
directory, for example, as in the following layout:

\
upstream/
└── foo/
    ├── priv.hpp
    ├── util.cpp
    └── util.hpp
\

\N|In multi-package projects, for example, those that provide both a library
and an executable, you would also want to understand how the sources are split
between the packages.|

If the headers and sources are split into different directories, then the
source directory may or may not have the inclusion subdirectory, similar to
the header directory. In the above split layout the \c{src/} directory doesn't
contain the inclusion subdirectory (\c{foo/}) while the following layout does:

\
upstream/
├── include/
│   └── foo/
│       └── util.hpp
└── src/
    └── foo/
        ├── priv.hpp
        └── util.cpp
\

With the understanding of these key properties of upstream layout you should
be in a good position to start crafting the \l{bdep-new(1)} command line that
recreates it.

\N|The \c{bdep-new} documentation uses a slightly more general terminology
compared to what we used in the previous section in order to also be
applicable to projects that use modules instead of headers.

Specifically, the inclusion subdirectory (\c{foo/}) is called \i{source
subdirectory} while the header directory (\c{include/}) and source directory
(\c{src/}) are called \i{header prefix} and \i{source prefix}, respectively.|


\h2#core-package-craft-cmd|Craft \c{bdep new} command line to create package|

The recommened procedure for this step is to read through the \c{bdep-new}'s
\l{bdep-new.xhtml#src-layout SOURCE LAYOUT} section (which contains a large
number of examples) while experimenting with various options in an attempt to
create the desired layout. If the layout you've got isn't quite right yet,
simply remove the package directory along with the \c{packages.manifest} file
and try again.

Let's illustrate this approach on the original example of the split layout:

\
upstream/
├── include/
│   └── foo/
│       └── util.hpp
└── src/
    ├── priv.hpp
    └── util.cpp
\

We know it's split, so let's start with that and see what we get. Remember,
our \c{foo} package repository that we have clonned and initialized earlier
looks like this:

\
$ tree foo/
foo/
├── .gitattributes
├── .gitignore
├── README.md
└── repositories.manifest
\

Now we create the \c{libfoo} package inside:

\
$ cd foo
$ bdep new --package --lang c++ --type lib,split libfoo
$ tree libfoo/
libfoo/
├── include/
│   └── libfoo/
│       └── foo.hxx
└── src/
    └── libfoo/
        └── foo.cxx
\

The outer structure looks right, but inside \c{include/} and \c{src/} things
are a bit off. Specifically, the source subdirectory should be \c{foo/}, not
\c{libfoo/}, there shouldn't be one inside \c{src/}, and the file extensions
don't match upstream. All this can be easily tweaked, however:

\
$ rm -r libfoo/ packages.manifest
$ bdep new --package \
  --lang c++,cpp   \
  --type lib,split,subdir=foo,no-subdir-source \
  libfoo
$ tree libfoo/
libfoo/
├── include/
│   └── foo/
│       └── foo.hpp
└── src/
    └── foo.cpp
\

The other \c{bdep-new} sub-options (see the \l{bdep-new(1)} man page for the
complete list) that you will likely want to use when packaging a third-party
project include:

\dl|

\li|\n\cb{no-version}

Omit the auto-generated version header. Usually upstream will provided its own
equivalent to this functionality.

\N|Note that even if upstream doesn't provide any version information, it's
not a good idea to try to rectify this by providing your own version header
since upstream may add it in a future version and you may end up with a
conflict. Instead, work with the project maintainer to rectify this in
upstream.||

\li|\n\cb{no-symexport}\n\cb{auto-symexport}

The \c{no-symexport} sub-option suppresses the generation of the DLL symbol
exporting header. This is an appropriate option if upstream provides its
own symbol exporting arrangements.

The \c{auto-symexport} sub-option enables automatic DLL symbol exporting
support (see \l{b##cc-auto-symexport Automatic DLL Symbol Exporting} for
background). This is an appropriate option if upstream relies on similar
support in the existing build system. It is also recommended that you give
this functionality a try even if upstream does not support building
shared libraries on Windows.|

\li|\n\cb{binless}

Create a header-only library. See \l{#dont-header-only Don't make
library header-only if it can be compiled} and
\l{https://github.com/build2/HOWTO/blob/master/entries/make-header-only-library.md
How do I make a header-only C/C++ library?}|

\li|\n\cb{buildfile-in-prefix}

Place header/source \c{buildfiles} into the header/source prefix directory
instead of source subdirectory. To illustrate the difference, compare these
two auto-generated layouts paying attention to the location of \c{buildfiles}:

\
$ bdep new ... --type lib,split,subdir=foo libfoo
$ tree libfoo/
libfoo/
├── include/
│   └── foo/
│       ├── buildfile
│       └── foo.hpp
└── src/
    └── foo/
        ├── buildfile
        └── foo.cpp
\

\
$ bdep new ... --type lib,split,subdir=foo,buildfile-in-prefix libfoo
$ tree libfoo/
libfoo/
├── include/
│   ├── foo/
│   │   └── foo.hpp
│   └── buildfile
└── src/
    ├── foo/
    │   └── foo.cpp
    └── buildfile
\

Note that this sub-option only makes sense if we have the header and/or source
prefixes (\c{include/} and \c{src/} in our case) as well as the source
subdirectory (\c{foo/} in our case).

Why would we want to do this? The main reason is to be able to symlink the
entire upstream directories rather than individual files. In the first
listing, the generated \c{buildfiles} are inside the \c{foo/} subdirectories
which mean we cannot just symlink \c{foo/} from upstream.

With a large number of files to symlink, this can be such a strong motivation
that it may make sense to invent a source subdirectory in the source prefix
even if upstream doesn't have one. See \l{#dont-main-target-root-buildfile
Don't build your main targets in root \c{buldfile}} for details on this
technique.

Another reason we may want to move \c{buildfiles} to prefix is to be able to
handle upstream projects that have multiple source subdirectories. While this
situation is not very common in the header prefix, it can be enountered in the
source prefix of more complex projects, where upstream wishes to organize the
source files into components.||

Continuing with our \c{libfoo} example, assuming upstream provides own symbol
exporting, the final \c{bdep-new} command line would be:

\
$ bdep new --package \
  --lang c++,cpp   \
  --type lib,split,subdir=foo,no-subdir-source,no-version,no-symexport \
  libfoo
\


\h2#core-package-review|Review and test auto-genetated \c{buildfile} templates|

Let's get a more complete view of what got generated by the final \c{bdep-new}
command line from the previous section:

\
$ tree libfoo/
libfoo/
├── build/
│   └── ...
├── include/
│   └── foo/
│       ├── buildfile
│       └── foo.hpp
├── src/
│   ├── buildfile
│   └── foo.cpp
├── tests/
│   ├── build/
│   │   └── ...
│   ├── basics/
│   │   ├── buildfile
│   │   └── driver.cpp
│   └── buildfile
├── buildfile
├── manifest
└── README.md
\

Once the overall layout looks right, the next step is to take a closer look at
the generated \c{buildfiles} to make sure that overall they match the upstrem
build. Of particular interest are the header and source directory
\c{buildfiles} (\c{libfoo/include/foo/buildifle} and \c{libfoo/src/buildifle}
in the above listing) which define how the library is built and installed.

Here we are focusing on the macro-level differences that are easier to change
by tweaking the \c{bdep-new} command line rather than manually. For example,
if we look at the generated source directory \c{buildfile} and realize it
builds a \i{binful} library (that is, a library that includes source files and
therefore produces library binaries) while the upsteam library is header-only,
it is much easier to fix this by re-running \c{bdep-new} with the \c{binless}
sub-option than by changing the \c{buildfiles} manually.

\N|Don't be tempted to start making manual changes at this stage even if you
cannot see anything else that can be fixed with a \c{bdep-new} re-run. This
is still a dry-run and we will recreate the package one more time in the
following section before starting manual adjustments.|

Besides examining the generated \c{buildfiles}, it's also a good idea to
build, test, and install the generated package to make sure everything ends up
where you expected and matches upstream where necessary. In particular, make
sure public headers are installed into the same location as upstream.

\N|The \c{bdep-new}-generated library is a simple \"Hello, World!\" example
that can nevertheless be built, tested, and installed. The idea here is to
verify it matches upstream using the generated source files before replacing
them with the upstream source file symlinks.|

Note that at this stage its easiest to build, test, and install in source
directly sidestepping the \c{bdep} initialization of the package (which you
would have to de-initalize before you can re-run \c{bdep-new}). Continue
with the above example, the recommended sequence of commands would be:

\
$ cd libfoo
$ b update
$ b test
$ b install config.install.root=/tmp/install
$ b clean
\

Let's also briefly discuss other subdirectories and files found in the
\c{bdep-new}-generated \c{libfoo} package.

The \c{build/} subdirectory is the standard \c{build2} place for project-wide
build system information (see \l{b#intro-proj-struct Project Structure} for
details). We will look closer at its contents in the following sections.

In the root directory of our package we find the root \c{buildfile} and
package \c{manifest}. We will be tweaking both in the following steps. There
is also \c{README.md} which we will replace with the upstream symlink.

The \c{tests/} subdirectory is the standard \c{build2} tests subproject (see
\l{b#intro-operations-test Testing} for details). While you can suppress its
generation with the \c{no-tests} \c{bdep-new} sub-option, we recommend that
you keep it and use it as a starting point for porting upstream tests or, if
upstream doesn't provide any, for a basic \"smoke test\" (@@ ref HOWTO).

\N|You can easily add/remove/rename this \c{tests/} subproject. The only place
where it is mentioned explicitly and where you will need to make changes is
the root \c{buildfile}. In pacticular, if upstream provides examples that you
wish to port, it is recommended that you use a copy of the generated
\c{tests/} subproject as a starting point (not forgeting to add the
corresponding entry in the root \c{buildfile}).|


\h2#core-package-create|Create final package|

If you are satisfied with the \c{bdep-new} command line and there are no more
automatic adjustments you can squeeze out of it, then it's time to re-run
\c{bdep-new} one last time to create the final package.

\N|While redoing this step later will require more effort, especially if
you've made manual modifications to \c{buildfile} and \c{manifest}, nothing is
set in stone and it can be done again by simply removing the package directory
and removing (or editing, if you have multiple packages and only want to redo
some of them) \c{packages.manifest} and starting over.|

This time, however, we will do things a bit differently in order to take
advantage of some additional automation offered by \c{bdep-new}. If the
package directory already exists and contains certain files, \c{bdep-new} can
take this into account when generating the root \c{buildfile} and package
\c{manifest}. In particular, it will try to guess the license from the
\c{LICENSE} file and extract the summary from \c{README.md} and use this
information in \c{manifest}.

\N|If the file names or formats used by upstream don't match those recognized
by \c{bdep-new} or if an attempt to extra the information is unsuccessful,
then for now simply omit the corresponding files from the package directory
and add them later manually.

Specifically, for \c{README}, \c{bdep-new} only recognizes \c{README.md}.

For license files, \c{bdep-new} recognizes \c{LICENSE}, \c{LICENSE.txt}
\c{LICENSE.md}, \c{COPYING}, and \c{UNLICENSE}.


@@ TODO: PACKAGE-README.md and README-PACKAGE.md (and below)

|

Continuing with our \c{libfoo} example and assuming upstream provides the
\c{README.md} and \c{LICENSE} files, we first manually create the package
directory, then add the symlinks, and finally run \c{bdep-new} (notice that we
have omitted the package name from the \c{bdep-new} command line since we are
running from inside the package directory):

\
$ cd foo/ # Change to the package repository root.

$ rm -r libfoo/ packages.manifest
$ mkdir libfoo/

$ cd libfoo/ # Change to the package root.
$ ln -s ../upstream/README.md ./
$ ln -s ../upstream/LICENSE   ./

$ bdep new --package                            \
  --lang c++,cpp   \
  --type lib,split,subdir=foo,no-subdir-source,no-version,no-symexport
\

If auto-detection succeeds, then you should see the \c{summary} and
\c{license} values automatically populated in \c{manifest} and the symlinked
files listed in the root \c{buildfile}.


\h2#core-package-adjust-version|Adjust package version|

While adjusting the \c{bdep-new}-generated code is the subject of the
following sections, one tweak that we want to make right away is to change the
package version in the \c{manifest} file.

In this guide we will assume the upstream package uses semver (semantic
version) or semver-like (that is, has three version components) and will rely
on the \i{continuous versioning} feature of \c{build2} to make sure that each
commit in our package repository has a distinct version (see
\l{intro#guide-versioning-releasing Versioning and Release Management} for
background).

\N|If upstream does not use semver, then see
\l{https://github.com/build2/HOWTO/blob/master/entries/handle-projects-which-dont-use-semver.md
How do I handle projects that don't use semantic versioning?} and
\l{https://github.com/build2/HOWTO/blob/master/entries/handle-projects-which-dont-use-version.md
How do I handle projects that don't use versions at all?} for available
options. If you decide to use the non-semver upstream version as is, then you
will have to forgo \i{continuous versioning} as well as the use of
\l{bdep-release(1)} for release management. The rest of the guide, however,
will still apply. In particular, you will still be able to use \l{bdep-ci(1)}
and \l{bdep-publish(1)} with a bit of extra effort.|

The overall plan to implement continous versioning is to start with a
pre-release snapshot of the upsream version, keep it like that while we are
adjusting the \c{bdep-new}-generated package and committing our changes (at
which point we get distinct snapshot versions), and finally, when the package
is ready to publish, change to the final upstream version with the help of
\l{bdep-release(1)}. Specifically, if the upstream version is
\c{\i{X}.\i{Y}.\i{Z}}, then we start with the \c{\i{X}.\i{Y}.\i{Z}-a.0.z}
pre-release snapshot.

Let's see how this works for our \c{libfoo} example. Say, the upstream version
that we are packaging is \c{2.1.0}. This means we start with \c{2.1.0-a.0.z}.

\N|Naturally, the upstream version that we are using should correspond to the
commit of the \c{upstream} submodule we have added on the
\l{#core-repo-submodule Add upstream repository as \c{git} submodule} step.|

Next we edit the \c{manifest} file in the \c{libfoo} package and change the
\c{version} value to read:

\
version: 2.1.0-a.0.z
\

Let's also commit this initial state of the package for easier rollbacks:

\
$ cd foo/ # Change to the package repository root.
$ git add .
$ git status
$ git commit -m \"Initialize package\"
\


\h#core-fill|Fill package with source code and add dependencies|

With the package skeleton ready, the next steps are to fill it with upstream
source code, add dependencies, and make any necessary manual adjustments to
the generated \c{buildfiles}, \c{manifest}, etc. If we do this all at once,
however, it can be hard to pin-point the cause of build failures. For example,
if we convert both the library and its tests right away and something doesn't
work, it can be hard to determine whether the mistake is in the library or in
the tests.  As a result, we are going to split this work into a sequence or
smaller steps that incrementally replace the \c{bdep-new}-generated code with
upstream while allowing us to test each change individually. We will also
commit the changes on each step for easy rollbacks. Specifically, the overall
plan is as follows:

\ol|

\li|Initialize (\c{bdep-init}) the package in one or more build configurations.|

\li|Add dependencies, if any.|

\li|Fill the library with upstream source code.|

\li|Adjust project-wide and source subdirectory \c{buildfiles}.|

\li|Make a smoke test for the library.|

\li|Replace the smoke test with upstream tests.|

\li|Tweak root \c{buildfile} and \c{manifest}.|

\li|Test the result using the CI service. @@ Actually doing it as ready.|

|

The first three steps are the subject of this section with the following
sections covering the rest of the plan.

\N|As you become more experienced with packaging third-party projects for
\c{build2} it may make sense to start combining or omitting some steps,
especially for simpler libraries. For example, if you see that a library
comes with a simple test that shouldn't cause any complications, then you
could omit the smoke test.|


\h2#core-fill-init|Initialize package in build configurations|

Before we start making any changes to the \c{bdep-new}-generated files, let's
initialize the package in at least one build configuration so that we are able
to build and test our changes (see \l{intro#guide Getting Started Guide} for
background on \c{bdep}-based development workflow). Continuing with our
\c{libfoo} example from the earlier steps:

\
$ cd foo/ # Change to the package repository root.
$ bdep init -C ../foo-gcc @gcc cc config.cxx=g++
\

Let's build and test the \c{bdep-new}-generated package to make sure
everything is in order:

\
$ bdep update
$ bdep test
$ bdep clean
\

You can create additional configurations, for example, if you have access to
several compilers. For instance, to create a build configuration for Clang:

\
$ bdep init -C ../foo-clang @clang cc config.cxx=clang++
\

If you would like to perform a certain operation on all the build
configurations, pass the \c{-a|--all} flag to \c{bdep}:

\
$ bdep update -a
$ bdep test -a
$ bdep clean -a
\

Let's also verify that the resulting package repository is clean (doesn't have
any uncommitted or untracked files):

\
$ git status
\

\h2#core-fill-depend|Add dependencies|

If the upstream project has any dependencies, now is a good time to specify
them so that when we attempt to build upstream source code, they are already
present.

Identifiying whether the upstream project has dependencies is not always easy.
The natural first places to check are the documentation and the existing build
system. Sometimes projects also bundle their dependencies with the project
source code (also called vendoring). So it makes sense to look around the
upstream repository for anything that looks like bundled dependencies.
Normally we would need to \"unbundle\" such dependencies when converting
to \c{build2} by instead specifying a dependency on an external package.

\N|While there are several reasons we insist on unbundling of dependencies,
the main one is that bundling can cause multiple, potentially conflicting
copied of the same dependency to exist in the build. This can cause subtle
build failures that are hard to understand and to track down.|

One particularly common case to check for is bundling of the testing
framework, such as Catch2, by C++ projects. If you have identified that the
upstream tests depend on a testing framework (whether bundled or not), see
\l{https://github.com/build2/HOWTO/blob/master/entries/handle-tests-with-extra-dependencies.md
How do I handle tests that have extra dependencies?}

If you have concluded that the upstream project doesn't have any dependencies,
then you can remove \c{repositories.manifest} from the package repository root
(uness you have already done so), commit this change, and skip the rest of
this section.

And if you are still reading, then we assume you have a list of dependencies
you need to add, preferably with their minimum required versions. If you could
not identify the minimum required version for a dependency, then you can
fallback to the latest available version, as will be described in a moment.

With the list of dependencies in hand, the next step is to determine whether
they are already available as \c{build2} packages. For that, head over to
\l{https://cppget.org cppget.org} and seach for each dependency.

If you are unable to find a package for a dependency, then it means it hasn't
been packaged for \c{build2} yet. Check the places mentioned in the
\l{#core-repo-exists Check if package repository already exists} step to see
if perhaps someone is already working on the package. If not and the
dependency is not optional, then the only way forward is to first package
the dependency.

If you do find a package for a dependency, then note the section of the
repository (\c{stable}, \c{testing}, etc; see \l{intro#guide-repositories
Package Repositories} for background) from which the minimum required version
of the package is available. If you were unable to identify the minimum
required version, then note the latest version available from the \c{stable}
section.

Given the list of repository sections, edit the \c{repositories.manifest} file
in the package repository root and uncomment the entry for \c{cppget.org}:

\
:
role: prerequisite
location: https://pkg.cppget.org/1/stable
#trust: ...
\

Next, replace \c{stable} at the end of the \c{location} value with the least
stable section from your list. For example, if your list contains \c{stable},
\c{testing}, and \c{beta}, then you need \c{beta} (the sections form a
hierarchy so that \c{beta} includes \c{testing} which in turn inclues
\c{stable}).

\N|If you wish, you can also uncomment the \c{trust} value and replace \c{...}
with the \l{https://cppget.org/?about repostitory fingerprint}. This way you
won't be prompted to confirm the repository authenticity on first fetch. See
\l{intro#guide-add-remove-deps Adding and Removing Dependencies} for details.|

Once this is done, edit \c{manifest} in package root and add the \c{depends}
value for each dependency. See \l{intro#guide-add-remove-deps Adding and
Removing Dependencies} for background. In particular, here you will use the
minimum required version (or the latest available) to form a version
contraint. Which constaint operator to use will depend on the dependency's
versioning policies. If the dependency uses semver, then a \c{^}-based
constraint is a sensible default.

As an example, let's say our \c{libfoo} depends on \c{libz}, \c{libasio}, and
\c{libsqlite3}. To specify these dependencies we would add the following
entries to its \c{manifest}:

\
depends: libz ^1.2.0
depends: libasio ^1.28.0
depends: libsqlite3 ^3.39.4
\

With all the dependencies specified, now let's synchronize the state of the
build configurations with our changes by running \l{bdep-sync(1)} from the
package repository root:

\
$ bdep sync -a
\

\N|If you have any build-time dependencies (see
\l{intro#guide-build-time-linked Build-Time Dependencies and Linked
Configurations} for background), then you will get a warning about the
corresponding \c{config.import.*} variable being unused and therefore
dropped. This is because we haven't yet added the corresponding \c{import}
directives to our \c{buildfiles}. For now you can ignore this warning and we
will fix it later, when we adjust the generated \c{buildfiles}.|

This command should first fetch the metadata for the repository we specified
in \c{repositories.manifest} and then fetch, unpack and configure each
dependency that we specified in \c{manifest}. We can examine the resulting
state, including the version of each dependency, with \l{bdep-status(1)}:

\
$ bdep status -ai
\

The last step for this section is to commit our changes:

\
$ cd foo/ # Change to the package repository root.
$ git add .
$ git status
$ git commit -m \"Add dependencies\"
\


\h2#core-fill-source|Fill with upstream source code|

Now we are ready to begin replacing the \c{bdep-new}-generated files with
upstream source code symlinks and we start with library's header and source
files. Continuing with our \c{libfoo} example, this is what we currently have
(notice that \c{LICENSE} and \c{README.md} are already symlinks to upstream):

\
$ cd foo/ # Change to the package repository root.

$ tree libfoo/
libfoo/
├── build/
│   └── ...
├── include/
│   └── foo/
│       ├── buildfile
│       └── foo.hpp
├── src/
│   ├── buildfile
│   └── foo.cpp
├── tests/
│   └── ...
├── LICENSE   -> ../upstream/LICENSE
├── README.md -> ../upstream/README.md
├── buildfile
└── manifest
\

Now we replace generated \c{include/foo/foo.hpp} with library's real headers
and \c{src/foo.cpp} with its real source files:

\
$ cd libfoo/ # Change to the package root.

$ cd include/foo/
$ rm foo.hpp
$ ln -s ../../../upstream/include/foo/*.hpp ./
$ cd -

$ cd src
$ rm foo.cpp
$ ln -s ../../upstream/src/*.cpp ./
$ cd -

$ tree libfoo/
libfoo/
├── build/
│   └── ...
├── include/
│   └── foo/
│       ├── buildfile
│       ├── core.hpp -> ../../../upstream/include/foo/core.hpp
│       └── util.hpp -> ../../../upstream/include/foo/util.hpp
├── src/
│   ├── buildfile
│   ├── core.cpp -> ../../upstream/src/core.cpp
│   └── util.cpp -> ../../upstream/src/util.cpp
├── tests/
│   └── ...
└── ...
\

Note that the wildcards used above may not be enough in all situations and
it's a good idea to manually examine the relevant upstream directories and
make sure nothing is missing. Specifically, look out for:

\ul|

\li|Header/sources with other extensions, for example, C, Objective-C, etc.|

\li|Other files that may be need, for example, \c{.def}, \c{config.h.in}, etc.|

\li|Subdirectories that contain more header/source files.||

If upstream contains subdirectories with addition header/source files, then
you can symlink entire subdirectories instead of doing it file by file. For
example, let's say \c{libfoo}'s upstream source directory contains the
\c{impl/} subdirectory with additional source files:

\
$ cd src
$ ln -s ../../upstream/impl ./
$ cd -

$ tree libfoo/
libfoo/
├── build/
│   └── ...
├── include/
│   └── ...
├── src/
│   ├── impl/ -> ../../upstream/src/impl/
│   │   ├── bar.cpp
│   │   └── baz.cpp
│   ├── buildfile
│   ├── core.cpp -> ../../upstream/src/core.cpp
│   └── util.cpp -> ../../upstream/src/util.cpp
├── tests/
│   └── ...
└── ...
\

Wouldn't it be nice if we could symlink the entire top-level subdirectories
(\c{include/foo/} and \c{src/}) in our case instead of symlinking individual
files? As discussed in \l{#core-package-craft-cmd Craft \c{bdep new} command
line to create package}, we can but we will need to change the package layout.
Specifically, we will need to move the \c{buildfiles} out of the source
subdirectories with the help of the \c{buildfile-in-prefix} sub-option of
\c{bdep-new}. In the above case, we will need to invent a source subdirectory
in \c{src/}. Whether this is a worthwhile change largely depends on how many
files you have to symlink individually. If it's just a handful, then it's
probably not worth the complication, especially if you have to invent source
subdirectories. On the other hand, if you are looking at symlinking hundreds
of files, changing the layout makes perfect sense.

\N|One minor drawback of symlinking entire directories is that you cannot
easily patch individual upstream files (see \l{#howto-patch-upstream-source
How do I patch upstream source code}).

You will also need to explicitly list such directories as symlinks in
\c{.gitattributes} if you want your package to be usable from the \c{git}
repository on Windows. See
\l{https://build2.org/article/symlinks.xhtml#windows Symlinks and Windows} for
details.|

We won't be able to test this change yet because to make things build will
most likely also need to tweak the generated \c{buildfiles}, which is the
subject of the next section. However, it still makes sense to commit our
changes to make rollbacks easier:

\
$ cd foo/ # Change to the package repository root.
$ git add .
$ git status
$ git commit -m \"Add upstream source symlinks\"
\


\h#core-adjust-build|Adjust project-wide and source \c{buildfiles}|

With source code and dependencies added, the next step is to adjust the
regenerated \c{buildfiles} that build the library. This involves two places:
the project-wide build system files in \c{build/} and the source subdirectory
\c{buildfiles} (in \c{include/} and \c{src/} for our \c{libfoo} example).


\h2#core-adjust-build-wide|Review project-wide build system files in \c{build/}|

We start with reviwing the files in the \c{build/} subdirectory of our
package, where you will find three files: \c{bootstrap.build}, \c{root.build},
and \c{export.build}. To recap, the first two contain the project-wide build
system setup (see \l{b#intro-proj-struct Project Structure} for details) while
the last is an export stub that facilitates the importation of targets from
our package (see \l{b#intro-import Target Importation} for details).

Normally you don't need to change anything in \c{bootstrap.build} \- all it
does is specify the build system project name and load a standard set of core
build system modules. Likewise, \c{export.build} is good as generated unless
you need to do something special, like exporting targets from different
subdirectories of your package.

While \c{root.build} is also often good as is, situations where you may
need to tweak it are not uncommon and include:

\ul|

\li|Loading additional build system module.

    For example, if your package makes use of Objective-C/C++ (see \l{b#c-objc
    Objective-C Compilation} and \l{b#cxx-objcxx Objective-C++ Compilation})
    or Assembler (see \l{b#c-as-cpp Assembler with C Preprocessor
    Compilation}), then \c{root.build} would be the natural place to load the
    correponding modules.

    \N|If your package uses a mixture of C and C++, then it's recommended to
    set this up using the \c{--lang} sub-option of \c{bdep-new} rather
    than manually. For example:

    \
    $ bdep new --lang c++,c ...
    \

    ||

\li|Specifying package configuration variable.

    If upstream provides the ability to configure their code, for example to
    enable optional features, then you may want to translate this to
    \c{build2} configuration variables, which must be specified in
    \c{root.build} (see \l{b#proj-config Project Configuration} for background
    and details).

    Note that you don't need to add all the configuration variables right
    away. Instead, you could first handle the \"core\" functionality which
    doesn't require any configuration and then add the configuration variables
    one by one while also making the corresponding changes in \c{buildfiles}.

    \N|One type of configuration that you should normally not expose when
    packaging for \c{build2} is support for both header-only and compiled
    modes. See \l{b#dont-header-only Don't make library header-only if it
    can be compiled}.|||

Also, in C++ projects, if you don't have any inline or template files, then
you can drop the assignment of the file extension for the \c{ixx} and \c{txx}
target types, respectively.

If you have added any configuration variables and would like to use
non-default values for some of them in your build, then you will need to
reconfigure the package. For example, let' say we have added the
\c{config.libfoo.debug} variable to our \c{libfoo} package which enables
additional debugging facilities in the library. This is how we can reconfigure
all our builds to enable this functionality:

\
$ bdep sync -a config.libfoo.debug=true
\

If you have made any changes, commit them (similar to the previous step, we
cannot test things just yet):

\
$ cd foo/ # Change to the package repository root.
$ git add .
$ git status
$ git commit -m \"Adjust project-wide build system files\"
\


\h2#core-adjust-build-src|Adjust source subdirectory \c{buildfiles}|

The next step we need to perform before we can build our library is to adjust
its \c{buildfiles}. These \c{buildfiles} are found in the source subdirectory
or, if we used the \c{buildfile-in-prefix} sub-option, in the prefix
directory. There will be two \c{buildfiles} if we use the split layout
(\c{split} sub-option) or a single \c{buildfile} in the combined layout. The
single \c{buildfile} in the combined layout contains essentially the same
definitions as the split \c{buildfiles} but combined into one and with some
minor simplifications that this allows. So here we will assume the split
layout and will continue with our \c{libfoo} from the previous sections. To
recap, here is the layout we've got with the \c{buildfiles} of interest found
in \c{include/foo/} and in \c{src/}:

\
libfoo/
├── build/
│   └── ...
├── include/
│   └── foo/
│       ├── buildfile
│       ├── core.hpp -> ../../../upstream/include/foo/core.hpp
│       └── util.hpp -> ../../../upstream/include/foo/util.hpp
├── src/
│   ├── buildfile
│   ├── core.cpp -> ../../upstream/src/core.cpp
│   └── util.cpp -> ../../upstream/src/util.cpp
├── tests/
│   └── ...
└── ...
\


\h2#core-adjust-build-src-header|Adjust header \c{buildfile}|

The \c{buildfile} in \c{include/foo/} is pretty simple:

\N|The \c{buildfile} in your package may look slightly different depending on
the exact \c{bdep-new} sub-options used. However, all the relevant definitions
discussed below should still be easily recognizable.|

\
pub_hdrs = {hxx ixx txx}{**}

./: $pub_hdrs

# Install into the foo/ subdirectory of, say, /usr/include/
# recreating subdirectories.
#
{hxx ixx txx}{*}:
{
  install         = include/foo/
  install.subdirs = true
}
\

Normally the only change that you would make to this \c{buildfile} is to
adjust the installation location of headers (see \l{b#intro-operations-install
Installing} for background). In particular, if our headers were included
without the \c{<foo/...>} prefix but instead contained the library name in
their names (for example, \c{foo_util.hpp}), then the installation setup would
instead look like this:

\
# Install directly into say, /usr/include/ recreating subdirectories.
#
{hxx ixx txx}{*}:
{
  install         = include/
  install.subdirs = true
}
\

If the library doesn't have any headers in subdirectories, you can drop
the \c{install.subdirs} variable:

\
# Install into the foo/ subdirectory of, say, /usr/include/.
#
{hxx ixx txx}{*}: install = include/foo/
\

\N|In the combined layout, the installation-related definitions are at
the end of the combined \c{buildfile}.|

See also \l{#howto-extra-header-install-subdir How do I handle extra header
installation subdirectory}.


\h2#core-adjust-build-src-source|Adjust source \c{buildfile}: overview|

Next is the \c{buildfile} in \c{src/}:

\N|Again, the \c{buildfile} in your package may look slightly different
depending on the exact \c{bdep-new} sub-options used. However, all the
relevant definitions discussed below should still be easily recognizable.

For a binless (header-only) library, this \c{buildfile} will contain only a
small subset of the definitions shown below. See
\l{https://github.com/build2/HOWTO/blob/master/entries/make-header-only-library.md
How do I make a header-only C/C++ library?} for additional considerations
when packaging header-only libraries.|

\
intf_libs = # Interface dependencies.
impl_libs = # Implementation dependencies.
#import xxxx_libs += libhello%lib{hello}

# Public headers.
#
pub = [dir_path] ../include/foo/

include $pub

pub_hdrs = $($pub/ pub_hdrs)

lib{foo}: $pub/{$pub_hdrs}

# Private headers and sources as well as dependencies.
#
lib{foo}: {hxx ixx txx cxx}{**} $impl_libs $intf_libs

# Build options.
#
out_pfx_inc = [dir_path] $out_root/include/
src_pfx_inc = [dir_path] $src_root/include/
out_pfx_src = [dir_path] $out_root/src/
src_pfx_src = [dir_path] $src_root/src/

cxx.poptions =+ \"-I$out_pfx_src\" \"-I$src_pfx_src\" \
                \"-I$out_pfx_inc\" \"-I$src_pfx_inc\"

#{hbmia obja}{*}: cxx.poptions += -DFOO_STATIC_BUILD
#{hbmis objs}{*}: cxx.poptions += -DFOO_SHARED_BUILD

# Export options.
#
lib{foo}:
{
  cxx.export.poptions = \"-I$out_pfx_inc\" \"-I$src_pfx_inc\"
  cxx.export.libs = $intf_libs
}

#liba{foo}: cxx.export.poptions += -DFOO_STATIC
#libs{foo}: cxx.export.poptions += -DFOO_SHARED

# For pre-releases use the complete version to make sure they cannot
# be used in place of another pre-release or the final version. See
# the version module for details on the version.* variable values.
#
if $version.pre_release
  lib{foo}: bin.lib.version = \"-$version.project_id\"
else
  lib{foo}: bin.lib.version = \"-$version.major.$version.minor\"

# Don't install private headers.
#
{hxx ixx txx}{*}: install = false
\


\h2#core-adjust-build-src-source-clean|Adjust source \c{buildfile}: cleanup|

As a first step, let's remove all the definitions that we don't need in our
library. The two common pieces of functionality that are often not needed
are support for auto-generated headers (such as \c{config.h} generated from
\c{config.h.in}) and dependencies on other libraries.

If you don't have any auto-generated headers, then remove all the assignments
and expansions of the \c{out_pfx_inc} and \c{out_pfx_src} variables. Here
is what the relevant lines in the above \c{buildfile} should look like after
this change:

\
# Build options.
#
src_pfx_inc = [dir_path] $src_root/include/
src_pfx_src = [dir_path] $src_root/src/

cxx.poptions =+ \"-I$src_pfx_src\" \"-I$src_pfx_inc\"

# Export options.
#
lib{foo}:
{
  cxx.export.poptions = \"-I$src_pfx_inc\"
}
\

\N|If you do have auto-genetated headers, then in the split layout you can
remove \c{out_pfx_inc} if you only have private auto-generated headers and
\c{out_pfx_src} if you only have public.|

\N|In the combined layout the single \c{buildfile} does not set the
\c{*_pfx_*} variables. Instead it uses the \c{src_root} and \c{out_root}
variables directly. For example:

\
# Build options.
#
cxx.poptions =+ \"-I$out_root\" \"-I$src_root\"

# Export options.
#
lib{foo}:
{
  cxx.export.poptions = \"-I$out_root\" \"-I$src_root\"
}
\

To remove support for auto-generated headers in the combined \c{buildfile},
simply remove the corresponding \c{out_root} expansions:

\
# Build options.
#
cxx.poptions =+ \"-I$src_root\"

# Export options.
#
lib{foo}:
{
  cxx.export.poptions = \"-I$src_root\"
}
\

If you only have private auto-genertated headers, then only remove the
expansion from \c{cxx.export.poptions}.|

If you don't have any dependencies, then remove all the assignments and
expansions of the \c{intf_libs} and \c{intf_libs} variables. That is,
the following lines in the original \c{buildfile}:

\
intf_libs = # Interface dependencies.
impl_libs = # Implementation dependencies.
#import xxxx_libs += libhello%lib{hello}

# Private headers and sources as well as dependencies.
#
lib{foo}: {hxx ixx txx cxx}{**} $impl_libs $intf_libs

# Export options.
#
lib{foo}:
{
  cxx.export.poptions = \"-I$out_pfx_inc\" \"-I$src_pfx_inc\"
  cxx.export.libs = $intf_libs
}
\

Become just these:

\
# Private headers and sources as well as dependencies.
#
lib{foo}: {hxx ixx txx cxx}{**}

# Export options.
#
lib{foo}:
{
  cxx.export.poptions = \"-I$out_pfx_inc\" \"-I$src_pfx_inc\"
}
\


\h2#core-adjust-build-src-source-dep|Adjust source \c{buildfile}: dependencies|

If you do have have dependencies, then let's handle them now.

\N|Here we will assume dependencies on other libraries, which is the common
case. If you have dependencies on executables, for example, source code
generators, see \l{intro#guide-build-time-linked Build-Time Dependencies and
Linked Configurations} on how to handle that. In this case you will also need
to reconfigure your package after adding the corresponding \c{import}
directives in order to re-acquire the previously dropped \c{config.import.*}
values. Make sure to also pass any configuration variables you specified in
\l{#core-adjust-build-wide Review project-wide build system files in
\c{build/}}. For example:

\
$ bdep sync -a --disfigure config.libfoo.debug=true
\

|

For each library that your package depends on (and which you have added
to \c{manifest} on the \l{#core-fill-depend Add dependencies} step),
you need to first determine whether it's an interface of implementation
dependency and then import it either into \c{intf_libs} or \c{impl_libs}
variable, respectively.

See \l{b#intro-lib Library Exportation and Versioning} for background on the
interface vs implementation distinction. But as a quick rule of thumb, if your
library includes a header from the dependency library in one of its public
headers, then it's an interface dependency. Otherwise, it's an implementation
dependency.

Continuing with our \c{libfoo} example, as we have established in
\l{#core-fill-depend Add dependencies}, it depends on \c{libasio}, \c{libz},
and \c{libsqlite3} and let's say we've determined that \c{libasio} is an
interface dependency because it's included from \c{include/foo/core.hpp} while
the other two are implementation dependencies because they are only included
from \c{src/}. Here is how we would change our \c{buildfile} to import them:

\
intf_libs = # Interface dependencies.
impl_libs = # Implementation dependencies.
import intf_libs += libasio%lib{asio}
import impl_libs += libz%lib{z}
import impl_libs += libsqlite3%lib{sqlite3}
\

And you can tidy this a bit further if you would like:

\
import intf_libs = libasio%lib{asio}
import impl_libs = libz%lib{z}
import impl_libs += libsqlite3%lib{sqlite3}
\

\N|If you don't have any implementation or interface dependencies, you can
remove the assignment and all the expansion of the corresponding \c{*_libs}
variable.|

Note also that system libraries like \c{-lm}, \c{-ldl} on UNIX or
\c{advapi32.lib}, \c{ws2_32.lib} on Windows should not be imported. Instead,
they should be listed in the \c{c.libs} or \c{cxx.libs} variables. See
\l{https://github.com/build2/HOWTO/blob/master/entries/link-system-library.md
How do I link a system library} for details.


\h2#core-adjust-build-src-source-pub|Adjust source \c{buildfile}: public headers|

With the unnecessary parts of the \c{buildfile} cleaned up and dependenies
handled, let's discuss the common changes to the remaining definitions, going
from top to bottom. We start with the public headers block:

\
# Public headers.
#
pub = [dir_path] ../include/foo/

include $pub

pub_hdrs = $($pub/ pub_hdrs)

lib{foo}: $pub/{$pub_hdrs}
\

This block gets hold of the list of public headers and makes them
prerequisites of the library. Normally you shouldn't need to make any changes
here. If you need to exclude some headers, it should be done in the
\c{buildfile} in the \c{inlcude/} directory.

\N|In the combined layout the single \c{buildfile} does not have such code.
Instead, all headers are covered by the wildcard pattern in following block.|


\h2#core-adjust-build-src-source-src|Adjust source \c{buildfile}: sources, private headers|

The next block deals with sources, private headers, and dependencies, if any:

\
# Private headers and sources as well as dependencies.
#
lib{foo}: {hxx ixx txx cxx}{**} $impl_libs $intf_libs
\

By default it will list all the relevant files as prerequisites of the
library, starting from the directory of the \c{buildfile} and including all
the subdirectories, recursively (see \l{b#name-patterns Name Patterns} for
background on wildcard patterns).

If your C++ package doesn't have any inline or template files, then you can
remove the \c{ixx} and \c{txx} target types, respectively (which is parallel
to the change made in \c{root.build}; see \l{#core-adjust-build-wide Review
project-wide build system files in \c{build/}}). For example:

\
# Private headers and sources as well as dependencies.
#
lib{foo}: {hxx cxx}{**} $impl_libs $intf_libs
\

The other common change to this block is the exclusion of certain files or
making them conditionally included. As an example, let's say in our \c{libfoo}
the source subdirectory contains a bunch of \c{*-test.cpp} files which are
unit tests and should not be listed as prerequisites of a library. Here is how
we can exclude them:

\
# Private headers and sources as well as dependencies.
#
lib{foo}: {hxx cxx}{** -**-test} $impl_libs $intf_libs
\

Let's also assume our \c{libfoo} contains \c{impl-win32.cpp} and
\c{impl-posix.cpp} which provide alternative implementations of the same
functionality for Windows and POSIX and should only be included as
prerequisites on the respective platforms. Here is how we can handle that:

\
# Private headers and sources as well as dependencies.
#
lib{foo}: {hxx cxx}{** -impl-win32 -impl-posix -**-test}
lib{foo}: cxx{impl-win32}: include = ($cxx.target.class == 'windows')
lib{foo}: cxx{impl-posix}: include = ($cxx.target.class != 'windows')
lib{foo}: $impl_libs $intf_libs
\

There are two nuances in the above example worth highlighting. Firstly, we
have to exclude the files from the wildcard pattern before we can
conditionally include them. Secondly, we have to always link libraries
last. In particual, the following is a shorter but an incorrect version of the
above:

\
lib{foo}: {hxx cxx}{** -impl-win32 -impl-posix -**-test} \
          $impl_libs $intf_libs
lib{foo}: cxx{impl-win32}: include = ($cxx.target.class == 'windows')
lib{foo}: cxx{impl-posix}: include = ($cxx.target.class != 'windows')
\

\N|You may also be tempted to use the \c{if} directive instead of the
\c{include} variable for conditional prerequisites. For example:

\
if ($cxx.target.class == 'windows')
  lib{foo}: cxx{impl-win32}
else
  lib{foo}: cxx{impl-posix}
\

This would also be incorrect. For background and details, see
\l{https://github.com/build2/HOWTO/blob/master/entries/keep-build-graph-config-independent.md
How do I keep the build graph configuration-independent?}|


\h2#core-adjust-build-src-source-opt|Adjust source \c{buildfile}: build and export options|

The next two blocks are the build and export options, which we will discuss
together:

\
# Build options.
#
out_pfx_inc = [dir_path] $out_root/include/
src_pfx_inc = [dir_path] $src_root/include/
out_pfx_src = [dir_path] $out_root/src/
src_pfx_src = [dir_path] $src_root/src/

cxx.poptions =+ \"-I$out_pfx_src\" \"-I$src_pfx_src\" \
                \"-I$out_pfx_inc\" \"-I$src_pfx_inc\"

#{hbmia obja}{*}: cxx.poptions += -DFOO_STATIC_BUILD
#{hbmis objs}{*}: cxx.poptions += -DFOO_SHARED_BUILD

# Export options.
#
lib{foo}:
{
  cxx.export.poptions = \"-I$out_pfx_inc\" \"-I$src_pfx_inc\"
  cxx.export.libs = $intf_libs
}

#liba{foo}: cxx.export.poptions += -DFOO_STATIC
#libs{foo}: cxx.export.poptions += -DFOO_SHARED
\

The build options are in effect when the library itself is being build and the
exported options are propagated to the library consumers (see \l{b#intro-lib
Library Exportation and Versioning} for background on exported options). For
now we will ignore the commented out lines that add \c{-DFOO_STATIC*} and
\c{-DFOO_SHARED*} macros \- they are for symbol exporting and we will discuss
this topic separately.

If the library you are packaging only uses portable APIs, then chances are you
won't need to change anything here. On the other hand, if it does anything
platform-specific, then you will most likely need to add some options here.

As discussed in the \l{b#intro-dirs-scopes Output Directories and Scopes}
section of the build system introduction, there is a number of variables that
are used to specify compilation and linking options, such as \c{*.poptions}
(\c{cxx.poptions} in the above example), \c{*.coptions}, etc. The below table
shows all of them with their rough \c{make} equivalents in the third column:

\
*.poptions   preprocess        CPPFLAGS
*.coptions   compile           CFLAGS/CXXFLAGS
*.loptions   link              LDFLAGS
*.aoptions   archive           ARFLAGS
*.libs       system libraries  LIBS/LDLIBS
\

The recommended approach here is to study the upstream build system definition
and copy custom compile/link options to the appropriate \c{build2} variables.
Note, however, that doing it thoughtlessly/faithfully by copying all the
options may not always be a good idea. See
\l{https://github.com/build2/HOWTO/blob/master/entries/compile-options-in-buildfile.md
Which C/C++ compile/link options are OK to specify in a project's buildfile?}
for the guidelines.

Also, oftentimes, such custom options must only be specified for certain
target platforms or when using a certain compiler. While \c{build2} provides a
large amount of information to identiy the build configuration as well as more
advanced \c{buildfile} language mechanism (such as \l{b#intro-switch Pattern
Matching (\c{switch})} to make sense of it, this is a large topic for which we
refer you to \l{b The \c{build2} Build System} manual. Additionally,
\l{https://github.com/build2-packaging github.com/build2-packaging} now
contains a large number of packages that you can study and search for
examples.

Let's also consider a representative example based on our \c{libfoo} to get a
sense of what this normally looks like as well as to highlight a few nuances.
Let's assume our \c{libfoo} requires either the \c{FOO_POSIX} or \c{FOO_WIN32}
macro to be defined during the build in order to identify the target
platform. Additionaly, extra features can be enabled by defining
\c{FOO_EXTRAS} both during the build and for consumption (so this macro must
also be exported). Next, this library requires the \c{-fno-strict-aliasing}
compile option for the GCC-class compilers (GCC, Clang, etc). Finally, we need
to link \c{pthread} on POSIX and \c{ws2_32.lib} on Windows. This is how we
would work all this into the above fragment:

\
# Build options.
#
out_pfx_inc = [dir_path] $out_root/include/
src_pfx_inc = [dir_path] $src_root/include/
out_pfx_src = [dir_path] $out_root/src/
src_pfx_src = [dir_path] $src_root/src/

cxx.poptions =+ \"-I$out_pfx_src\" \"-I$src_pfx_src\" \
                \"-I$out_pfx_inc\" \"-I$src_pfx_inc\"

cxx.poptions += -DFOO_EXTRAS

if ($cxx.target.class == 'windows')
  cxx.poptions += -DFOO_WIN32
else
  cxx.poptions += -DFOO_POSIX

#{hbmia obja}{*}: cxx.poptions += -DFOO_STATIC_BUILD
#{hbmis objs}{*}: cxx.poptions += -DFOO_SHARED_BUILD

if ($cxx.class == 'gcc')
  cxx.coptions += -fno-strict-aliasing

switch $cxx.target.class, $cxx.target.system
{
  case 'windows', 'mingw32'
    cxx.libs += -lws2_32
  case 'windows'
    cxx.libs += ws2_32.lib
  default
    cxx.libs += -pthread
}

# Export options.
#
lib{foo}:
{
  cxx.export.poptions = \"-I$out_pfx_inc\" \"-I$src_pfx_inc\" -DFOO_EXTRAS
  cxx.export.libs = $intf_libs
}

#liba{foo}: cxx.export.poptions += -DFOO_STATIC
#libs{foo}: cxx.export.poptions += -DFOO_SHARED
\

There are a few nuances in the above code worth keeping in mind. Firstly,
notice that we append (rather than assign) to all the non-export variables
(\c{*.poptions}, \c{*.coptions}, \c{*.libs}). This is because they may already
contain some values specified by the user with their \c{config.*.*}
counterparts. On the other hand, the \c{*.export.*} variables are assigned.

Secondly, the order in which we append to the variables is important for the
value to accumulate correctly. You want to fist append all the scope-level
values, then target type/pattern-specific, and finally any target-specific;
that is, from more general to more specific (see \l{b#intro-lang Buildfile
Language} for background). To illustrate this point, let's say in our
\c{libfoo}, the \c{FOO_POSIX} or \c{FOO_WIN32} macro are only necessary when
compiling \c{util.cpp}. Below would be the correct order of assigning to
\c{cxx.poptions}:

\
cxx.poptions =+ \"-I$out_pfx_src\" \"-I$src_pfx_src\" \
                \"-I$out_pfx_inc\" \"-I$src_pfx_inc\"

cxx.poptions += -DFOO_EXTRAS

#{hbmia obja}{*}: cxx.poptions += -DFOO_STATIC_BUILD
#{hbmis objs}{*}: cxx.poptions += -DFOO_SHARED_BUILD

if ($cxx.target.class == 'windows')
  {obja objs}{util}: cxx.poptions += -DFOO_WIN32
else
  {obja objs}{util}: cxx.poptions += -DFOO_POSIX
\

\N|Not that target-specific \c{*.poptions} and \c{*.coptions} must be
specified on the object file targets while \c{*.loptions} and \c{*.libs} \- on
the library or executable targets.|


\h2#core-adjust-build-src-source-sym|Adjust source \c{buildfile}: symbol exporting|

Let's now turn to a special sub-topic of the build and export options that
relates to the shared library symbol exporting. To recap, a shared library on
Windows must explicitly specify the symbols (functions and global data) that
it wishes to make accessible to its users. This can be achieved in three
different way: The library can explicitly mark in its source code the names
whose symbols should be exported. Alternatively, the library can profide a
\c{.def} file to the linker that lists the symbols to be exported. Finally,
the library can request automatic exporting of all symbols, which is the
default semantics on non-Windows platforms. Note that the last two approaches
only work for exporting functions, not data, unless extra steps are taken by
the library users. Let's discuss each of these approaches in the reverse
order, that is, starting with the automatic symbol exporting.

The automatic symbol exporting is implemented in \c{build2} by generating a
\c{.def} file that exports all the relevant symbols. It requires a few
additional definitions in our \c{buildfile} as described in
\l{b#cc-auto-symexport Automatic DLL Symbol Exporting}. You can automacially
generate the necessary setup with the \c{auto-symexport} \c{bdep-new}
sub-option.

Using a custom \c{.def} file to export symbols is fairly straightforward:
simply list it as a prerequsite of the library and it will be automatically
passed to the linker. For example:

\
# Private headers and sources as well as dependencies.
#
lib{foo}: {hxx cxx}{**} $impl_libs $intf_libs def{foo}
\

The last approach is to explicitly specify in the source code which symbols
must be exported by marking the corresponding declarations with
\c{__declspec(dllexport)} during the library build and
\c{__declspec(dllimport)} during the library use. This is commonly achieved
with a macro, customarily called \c{*_EXPORT} or \c{*_API}, which is defined
to one of the above specifiers based on whether static or shared library is
being build or being consumed, which, in turn, is also normally signalled with
a few more macros, such as \c{*_BUILD_DLL} and \c{*_USE_STATIC}.

In \c{build2} you can explicitly signal any of the four situations by
uncommending and adjusting the following four lines in the build and export
options blocks:

\
# Build options.
#

...

#{hbmia obja}{*}: cxx.poptions += -DFOO_STATIC_BUILD
#{hbmis objs}{*}: cxx.poptions += -DFOO_SHARED_BUILD

# Export options.
#

...

#liba{foo}: cxx.export.poptions += -DFOO_STATIC
#libs{foo}: cxx.export.poptions += -DFOO_SHARED
\

As an example, let's assume our \c{libfoo} defines in one of its headers the
\c{FOO_EXPORT} macro based on the \c{FOO_BUILD_DLL} (shared library is being
build) and \c{FOO_USE_STATIC} (static library is being used) macros that it
expects to be appropriately defined by the build system. This is how we would
modify the above fragment to handle this setup:

\
# Build options.
#

...

{hbmis objs}{*}: cxx.poptions += -DFOO_BUILD_DLL

# Export options.
#

...

liba{foo}: cxx.export.poptions += -DFOO_USE_STATIC
\


\h2#core-adjust-build-src-source-ver|Adjust source \c{buildfile}: shared library version|

The final few lines in the above \c{buildfile} deal with shared library binary
(ABI) versioning:

\
# For pre-releases use the complete version to make sure they cannot
# be used in place of another pre-release or the final version. See
# the version module for details on the version.* variable values.
#
if $version.pre_release
  lib{foo}: bin.lib.version = \"-$version.project_id\"
else
  lib{foo}: bin.lib.version = \"-$version.major.$version.minor\"
\

The \c{bdep-new}-generated setup arranges for the platform-independent
versioning where the package's major and minor version components are embedded
into the shared library binary name (and \c{soname}) under the assumption
that only patch versions are ABI-compatible.

The two situation where you would want to change this are when the above
assumption does not hold and/or when the upstream provides platform-specific
shared library versions which you would like to re-create in your \c{build2}
build. See \l{b#intro-lib Library Exportation and Versioning} for background
and details.


\h2#core-adjust-build-src-source-ext|Adjust source \c{buildfile}: extra requirements|

The changes discussed so far should be sufficient to handle a typical library
that is written in C and/or C++ and is able to handle platform differences
with the preprocessor and compile/link options. However, sooner or later you
will run into a more complex library that may use additional languages,
require more elaborate platform detection, or use additional functionality,
such as support for source code generators. The below list provides pointers
to resources that cover the more commonly encountered additional requirements.

\ul|

\li|\l{b#module-in The \c{in} build system module}

Use to process \c{config.h.in} (or other \c{.in} files) that don't require
Autoconf-style platform probing (\c{HAVE_*} options).|

\li|\l{https://github.com/build2/libbuild2-autoconf The \c{autoconf} build
    system module}

Use to process \c{config.h.in} (or their CMake/Meson variants) that require
Autoconf-style platform probing (\c{HAVE_*} options).|

\li|\l{b#c-objc Objective-C Compilation} and
    \l{b#cxx-objcxx Objective-C++ Compilation}

Use to compile Objective-C (\c{.m}) or Objective-C++ (\c{.mm}) source files.|

\li|\l{b#c-as-cpp Assembler with C Preprocessor Compilation}

Use to compile Assembler with C Preprocessor (\c{.S}) source files.|

\li|\l{b#intro-unit-test Implementing Unit Testing}

Use if upstream has tests (normally unit tests) in the source subdirectory.|

\li|\l{intro#guide-build-time-linked Build-Time Dependencies and Linked
    Configurations}

Use if upstream relies on source code generators, such as
\l{https://cppget.org/reflex \c{lex}} and \l{https://cppget.org/byacc
\c{yacc}}.|

\li|\l{https://github.com/build2/HOWTO/ The \c{build2} HOWTO}

See the \c{build2} HOWTO article collection for more unusual requirements.||


\h2#core-adjust-build-test|Test library build|

At this point our library should be ready to build, at least in theory. While
we cannot build and test the entire package before adjusting the generated
\c{tests/} subproject (the subject of the next step), we can try to build just
the library and, if it has any unit tests in the source subdirectory, even run
some tests.

\N|Is the library is header only, there won't be anything to build unless
there unit tests. Still you may want to continue with this exercise to detect
any syntactic mistakes in the \c{buildfiles}, etc.|

To build only a specific subdirectory of our package we use the build system
directly (continuing with our \c{libfoo} example):

\
$ cd libfoo/src/ # Change to the source subdirectory.
$ b update
\

If there are any issues, try to fix them and then build again. Once the
library builds and if you have unit tests, you can try to run them:

\
$ b test
\

Once the library builds, it makes sense to commit our changes for easier
rollbacks:

\
$ cd foo/ # Change to the package repository root.
$ git add .
$ git status
$ git commit -m \"Adjust source subdirectory buildfiles\"
\


\h#core-test-smoke|Make smoke test|

With the library build sorted, we need tests to make sure it is actually
functional. As \l{#core-fill discussed earlier}, it is recommended to start
with a simple smoke test, make sure that works, and then replace it with
upstream tests. However, if upstream tests look simple enough, you can skip
the smoke test. For example, if upstream has all its tests in a single source
file and its build doesn't look too complicated, then you can just use that
source file in place of the smoke test.

\N|If upstream has no tests, then the smoke test will have to stay. A library
can only be published if it has at least one test.

It is also recommended to have the smoke test if upstream tests are in a
separate package. See
\l{https://github.com/build2/HOWTO/blob/master/entries/handle-tests-with-extra-dependencies.md
How do I handle tests that have extra dependencies?} for background and
details.|

To recap, the \c{bdep-new}-generated \c{tests/} subdirectory looks like this
(continuing with our \c{libfoo} example):

\
libfoo/
├── ...
└── tests/
    ├── build/
    │   ├── bootstrap.build
    │   └── root.build
    ├── basics/
    │   ├── driver.cpp
    │   └── buildfile
    └── buildfile
\

The \c{tests/} subdirectory is a build system subproject, meaning that it can
be built independently, for example, to test the installed version of the
library (see \l{b#intro-operations-test Testing} for background). In
particular, this means it has the \c{build/} subdirectory with project-wide
build system files the same as our library. Then there is the \c{basics/}
subdirectory which contains the generated test and which is what we will be
turning into a smoke test. The subproject root \c{buildfile} rarely needs
changing.


\h2#core-test-smoke-build-wide|Review project-wide build system files in \c{tests/build/}|

Review the generated \c{bootstrap.build} and \c{root.build} (there will be no
\c{export.build}) similar to \l{#core-adjust-build-wide Review project-wide
build system files in \c{build/}}.

Here the only change you would normally make is in \c{root.build} to drop the
assignment of extensions for target types that are not used in tests.


\h2#core-test-smoke-adjust|Convert generated test to library smoke test|

The \c{basics/} subdirectory contains the \c{driver.cpp} source file that
implements the test and \c{buildfile} that builds it. You can rename both the
test directory (\c{basics/}) and the source file \c{driver.cpp}, for example,
if you are going with the upstream tests directly. You can also add more tests
by simply copying \c{basics/}.

The purpose of a smoke test is to make sure the library's public headers can
be included (including in the installed case, no pun intended), it can be
linked, and its basic functionality works.

To achieve this, we modify \c{driver.cpp} to include the library's main
headers and call a few functions. For example, if the library has the
init/deinit type of functions, those are good candidates to call. If the
library is not header-only, make sure that the smoke test calls at least one
non-inline/template function to test symbol exporting.

\N|Make sure that your test includes the library's public headers the same way
as would be used by the library users.|

Continuing with our \c{libfoo} example, this is what its smoke test might look
like:

\
#include <foo/core.hpp>
#include <foo/util.hpp>

#undef NDEBUG
#include <cassert>

int main ()
{
  foo::context* c (foo::init (0 /* flags */));
  assert (c != nullptr);
  foo::deinit (c);
}
\

\N|The C/C++ \c{assert()} macro is often adequate for simple tests and does
not require extra dependencies. But see
\l{https://github.com/build2/HOWTO/blob/master/entries/use-assert-in-tests.md
How do I correctly use C/C++ assert() in tests?}|

The test \c{buildfile} is pretty simple:

\
import libs = libfoo%lib{foo}

exe{driver}: {hxx ixx txx cxx}{**} $libs testscript{**}
\

If you have adjusted the library target name (\c{lib{foo\}}) in the source
subdirectory \c{buildfile}, then you will need to make the corresponding
change in the \c{import} directive here. You may also want to tidy it up by
removing unused prerequisite types. For example:

\
import libs = libfoo%lib{foo}

exe{driver}: {hxx cxx}{**} $libs
\


\h2#core-test-smoke-locally|Test locally|

With the smoke test ready, we can finally do some end-to-end testing of our
library build. We will start with doing some local testing to catch basic
mistakes and then do the full CI to detect any platform/compiler-specific
issues.

First let's run the test in the default build configuration by invoking
the build system directly:

\
$ cd libfoo/tests/ # Change to the tests/ subproject.
$ b test
\

If there are any issues (compile/link errors, test failures), try to address
them and re-run the test.

Once the default configuration builds and passes the tests, you can do the
same for all the build configurations, in case you have \l{#core-fill-init
initialized} your library in several:

\
$ bdep test -a
\

\h2#core-test-smoke-locally-install|Test locally: installation|

Once this works, let's test the installed version of the library. In
particular, this makes sure that the public headers are installed in a way
that is compatible with how they are included by our test (and would be
included by the users of our library). To test this we first install
the library into some temporary directory:

\
$ cd libfoo/ # Change to the package root.
$ b install config.install.root=/tmp/install
\

Next we build just the \c{tests/} subproject arranging for it to find
the installed library:

\
$ cd libfoo/ # Change to the package root.
$ b test: tests/@/tmp/libfoo-tests-out/ \
  config.cc.loptions=-L/tmp/install/lib \
  config.bin.rpath=/tmp/install/lib
\

\N|The equivalent MSVC command line would be:

\
> b install config.install.root=c:\tmp\install

> b test: tests\@c:\tmp\libfoo-tests-out\^
  config.cc.loptions=/LIBPATH:c:\tmp/\install\lib
\

|

\N|It is a good idea to look over the installed files and make sure there is
nothing unexpected, for example, missing or extraneous files.|

Once done testing the installed case, let's clean things up:

\
$ rm -r /tmp/install /tmp/libfoo-tests-out
\

\h2#core-test-smoke-locally-dist|Test locally: distribution|

Another special case worth testing is the preparation of the source
distribution (see \l{b#intro-operations-dist Distributing} for
background). This, in particular, is how your package will be turned into the
source archive for publishing to \l{https://cppget.org cppget.org}. Here we
are primarily looking for missing files. As a bonus, this will also allow us
to test the in source build. First we distribute our package to some temporary
directory:

\
$ cd libfoo/ # Change to the package root.
$ b dist config.dist.root=/tmp/dist config.dist.uncommitted=true
\

The result will be in the \c{/tmp/dist/libfoo-<version>/} directory which
should resemble our \c{libfoo/} package but without files like \c{.gitignore}.
Next we build and test the distribution in source:

\
$ cd /tmp/dist/libfoo-<version>/
$ b configure config.cxx=g++
$ b update
$ b test
\

\N|If your package has dependencies that you import in your \c{buildfile},
then the above \c{configure} operation will most likely fail because such
dependencies cannot be found (it may succeed if they are available as
system-installed). The error message will suggest specifying the location of
each dependency with \c{config.import.*} variable. You can fix this by setting
each such \c{config.import.*} to the location of the build configuration
\l{#core-fill-init created by \c{bdep}} which should contain all the necessary
dependecies. Simply re-run the \c{configure} operation until you have
discovered and specified all the necessary \c{config.import.*} variables,
for example:

\
$ b configure config.cxx=g++ \
  config.import.libz=.../foo-gcc \
  config.import.libasio=.../foo-gcc \
  config.import.libsqlite3=.../foo-gcc
\

|

\N|It is a good idea to look over the distributed files and make sure there is
nothing missing or extraneous.|

Once done testing the distribution, let's clean things up:

\
$ rm -r /tmp/dist
\


\h2#core-test-smoke-ci|Commit and test with CI|

With local testing complete, let's commit our changes and submit a remote CI
job to test our library on all the major platforms and with all the major
compilers:

\
$ cd foo/ # Change to the package repository root.
$ git add .
$ git status
$ git commit -m \"Add smoke test\"
$ git push

$ bdep ci
\

The result of the \l{bdep-ci(1)} command is a link where you can see the
status of the builds. If any fail, view the logs to determine the cause, try
to fix it, commit your fix, and CI again.

\N|It is possible that upstream does not support some platforms or compilers.
For example, it's common for smaller projects not to bother with supporting
\"secondary\" compilers, such as MinGW GCC on Windows or Homebrew GCC on Mac
OS.

If upstream expressly does not support some platform or compiler, it's probably
not worth spending time and energy trying to support it in the package. Most
likely it will require changes to upstream source code and that is best done
upstream rather than in the package (see \l{#dont-fix-upstream Don't try to
fix upstream issues in the package} for background). In this case you would
want to exclude these platforms/compilers from the CI builds using the
\l{bpkg#manifest-package-builds \c{builds} package \c{manifest} value}.

The other common cause of a failed build is a newer version of a compiler or
platform that breaks upstream. In this case there are three options: Ideally
you would want to fix this upstream and have a new version released. Failed
that, you may want to patch the upstream code to fix the issues, especially if
this is one of the major platforms and/or primary compilers (see
\l{#howto-patch-upstream-source How do I patch upstream source code} for
details). Finally, you can just leave the build failing with the expectation
that it will be fixed in the next upstream version. Note that in this case you
should not exclude the failing build from CI.|


\h#core-test-upstream|Replace smoke test with upstream tests|

With the smoke test working we can now proceed with replacing it with the
upstream tests.


\h2#core-test-upstream-understand|Understand how upstream tests work|

While there are some commonalities in how C/C++ libraries are normally built,
when it comes to tests there is unfortunately little common ground in how they
are arranged, built, and executed. As a result, the first step in dealing with
upstream tests is to study the existing build system and try to understand how
they work. To get you started, below are some of the questions you would
likely need answered before you can proceed:

\ul|

\li|\b{Are upstream tests unit tests or integration tests?}

While the distinction is often fuzzy, for our purposes the key differentiator
between unit and integration tests is which API they use: integration tests
only use the library's public API while unit tests need access to the
implementation details.

Normally (but not always), unit tests will reside next to the library source
code since they need access to more than just the library binary (individual
object files, utility libraries, etc). While integration tests are normally
(but again not always) placed into a seperate subdirectory, usually called
\c{tests} or \c{test}.

If the library has unit tests, then refer to \l{b#intro-unit-test Implementing
Unit Testing} for background on how to hanle them in \c{build2}.

If the library has integration tests, then use them to to replace (or
complement) the smoke test.

If the library has unit tests but no integration tests, then it's recommended
to keep the smoke test since that's the only way the library will be tested
via its public API.|


\li|\b{Do upstream tests use an external testing framework?}

Oftentimes a C++ library will use an external testing framework to implement
tests. Popular choices include \l{https://cppget.org/catch2 \c{catch2}},
\l{https://cppget.org/gtest \c{gtest}}, \l{https://cppget.org/doctest
\c{doctest}}, and \l{https://cppget.org/libboost-test \c{libboost-test}}.

If a library uses such an external testing framework, then it is recommended
to factor tests into a separate package in order to avoid making the library
package depend on the testing framework (which is only required during
testing). See
\l{https://github.com/build2/HOWTO/blob/master/entries/handle-tests-with-extra-dependencies.md
How do I handle tests that have extra dependencies?} for details.

\N|Sometimes you will find that upstream bundles the source code of the
testing framework with their tests. This is especially common with
\c{catch2}. If that's the case, it is strongly recommended that you
\"unbundle\" it by making it a proper external dependency.||


\li|\b{Are upstream tests in a single or multiple executables?}

It's not unusual for libraries to have a single test executable that runs all
the test cases. This is especially common if a C++ testing framework is used.
In this case it is natural to replace the contents of the smoke test with the
upstream source code, potentially renaming the test subdirectory (\c{basics/})
to better match upstream naming.

If upstream has multiple test executables, then they could all be in single
test subdirectory (potentially reusing some common bits) or spread over
multiple subdirectories. In both cases it's a good idea to follow the upstream
structure unless you have good reasons to deviate. In the former case (all
executables in the same subdirectory), you can re-purpose the smoke test
subdirectory. In the latter case (each executable in a separate subdirectory)
you can make copies of the smoke test subdirectory.|


\li|\b{Are upstream tests well behaved?}

Unfortunately it's not uncommon for upstream tests not to behave well, such as
write diagnostics to \c{stdout} instead of \c{stderr}, create temporary files
without cleaning them up, or assume presence of input files in the current
working directory. For details on how to deal with such situations see
\l{https://github.com/build2/HOWTO/blob/master/entries/sanitize-test-execution.md
How do I sanitize the execution of my tests?}||


\h2#core-test-upstream-convert|Convert smoke test to upstream tests|

Once you have a good grasp of how upstream tests work, convert or replace the
smoke test with the upstream tests. If upstream has multiple test executables,
you may want to deal with one test at a time, making sure that it passes
before moving to the next one.

It's normally a good idea to use the smoke test \c{buildfile} as a starting
point for upstream tests. To recap, the smoke test \c{buildfile} for our
\c{libfoo} example ended up looking like this:

\
import libs = libfoo%lib{foo}

exe{driver}: {hxx cxx}{**} $libs
\

At a minimum you will most likely need to change the name of the executable to
match upstream. If you need to build multiple executables in the same
directory, then it's probably best to get rid of the name pattern for the
source files and specify the prerequisite names explicitly, for example:

\
import libs = libfoo%lib{foo}

./: exe{test1}: cxx{test1} $libs
./: exe{test2}: cxx{test2} $libs
\

If you have a large number of such test executables, then a \c{for}-loop might
be a more scalable option:

\
import libs = libfoo%lib{foo}

for src: cxx{test*}
  ./: exe{$name($src)}: $src $libs
\


\h2#core-test-upstream-locally|Test locally|

With the upstream tests ready, we re-do the same end-to-end testing as we did
with the smoke test:

\l{#core-test-smoke-locally Test locally}\n
\l{#core-test-smoke-locally-install Test locally: installation}\n
\l{#core-test-smoke-locally-dist Test locally: distribution}\n


\h2#core-test-upstream-ci|Commit and test with CI|

With local testing complete, we commit our changes and submit a remote CI
job. This step is similar to what \l{#core-test-smoke-ci we did for the smoke
test} but this time we are using the upstream tests:

\
$ cd foo/ # Change to the package repository root.
$ git add .
$ git status
$ git commit -m \"Add upstream tests\"
$ git push

$ bdep ci
\


\h#core-examples-banchmarks|Add upstream examples, benchmarks, if any|

If the upstream project provides examples and/or benchmarks and you wish to
add them to the \c{build2} build (which is not strictly necessary for the
\c{build2} package to be usable), then now is a good time to do that.

As was mentioned in \l{#core-package-review Review and test auto-genetated
\c{buildfile} templates}, the recommended approach is to copy the \c{tests/}
subproject (potentially from the commit history before the smoke test was
replaced with the upstream tests) and use that as a starting point for
examples and/or benchmarks. Just do not forgeting to add the corresponding
entry in the root \c{buildfile}.

Once that is done, follow the same steps as in \l{#core-test-upstream Replace
smoke test with upstream tests} to add upstream examples/benchmarks and test
the result.


\h#core-root|Adjust root \c{buildfile} and \c{manifest}|

The last few files that we need to review and potentially adjust are
the root \c{buildfile} and package \c{manifest}.


\h2#core-root-buildfile|Adjust root \c{buildfile}|

The main function of the root \c{buildfile} is to pull all the subdirectories
that need building plus list targets that are usually found in the root
directory of a project, typically \c{README.md}, \c{LICENSE}, etc. This is
what the generated root \c{buildfile} looks like for our \c{libfoo} project
assuming we have symlinked \c{README.md} and \c{LICENSE} from upstream on the
\l{#core-package-create Create final package} step:

@@ PACKAGE-README.md?

\
./: {*/ -build/} doc{README.md} legal{LICENSE} manifest

# Don't install tests.
#
tests/: install = false
\

If the upstream project provides any other documentation (change log, news,
etc) or legal files (list of authorts, code of conduct, etc), then you may
want to symlink and list them as the \c{doc{\}} and \c{legal{\}}
prerequisites, respectively.

\N|One file you don't need listing is \c{INSTALL} (or equivalent) which
normally contains the installation instructions for the upstream build
system. In the \c{build2} package the \c{PACKAGE-README.md} file serves this
purpose.|


\h2#core-root-buildfile-doc|Adjust root \c{buildfile}: other subdirectories|

If the upstream project has other subdirectories that makes sense to include
into the \c{build2} package, then now is good time to take care of that.  The
most common such case will be extra documentation (besides the root
\c{README}), typically in a subdirectory called \c{doc/}, \c{docs/}, or
\c{documentation/}.

The typical procedure for handling such subdirectories will be to symlink the
relevant files (or the entire subdirectory) and then list the files as
prerequisites. For this last step, there are two options: we can list the
files directly in the root \c{buildfile} or we can create a seperate
\c{buildfile} in the subdirectory.

Let's examine both approaches using our \c{libfoo} as an example. Assume that
upstream \c{libfoo} contains the \c{docs/} subdirectory with additional
\c{*.md} files that document its API. It would make sense to include them into
the \c{build2} package.

Listing the subdirectory files directly in the root \c{buildfile} works best
for simple case, where you have a bunch of static files that don't require any
customizations, such as to their installation location. In this case we can
symlink the entire \c{docs/} subdirectory:

\
$ cd libfoo/ # Change to the package root.
$ ln -s ../upstream/docs ./
\

The adjustment to the root \c{buildfile} are pretty straightforward: we
exclude the \c{docs/} subdirectory (since it has no \c{buildfile}) and list
the \c{*.md} files as prerequisites using the \c{doc{\}} target type (which,
in particular, makes sure they are installed into the appropriate location):

\
./: {*/ -build/ -docs/} \
    doc{README.md} docs/doc{*.md} \
    legal{LICENSE} manifest
\

The alternative approach (create a seperate \c{buildfile}) is a good choice if
things are more complicated then that. Let's say we need to adjust the
installation location of the files in \c{docs/} because there is another
\c{README.md} that would conflict with the root one when installed into the
same location. This time we cannot symlink the top-level \c{docs/}
subdirectory (because we need to place a \c{buildfile} there). The two options
here is to either symlink the individual files or introduce another
subdirectory level inside \c{docs/} (which is the same approach as discussed
in \l{#dont-main-target-root-buildfile Don't build your main targets in root
\c{buldfile}}). Let's illustrate both sub-cases.

Symlinking individual files works best when you don't expect the set of files
to change often. For example, if \c{docs/} contains a man page and its HTML
rendering, then it's unlikely this set will change. On the other hand, if
\c{docs/} contains a manual split into an \c{.md} file per chapter, then there
is good chance this set of files will fluctuate between releases.

Continuing with our \c{libfoo} example, this is how we symlink the individual
\c{*.md} files in \c{docs/}:

\
$ cd libfoo/ # Change to the package root.
$ mkdir docs
$ cd docs/
$ ln -s ../../upstream/docs/*.md ./
\

Then write a new \c{buildfile} in \c{docs/}:

\
./: doc{*.md}

# Install the documentation in docs/ into the manual/ subdirectory of,
# say, /usr/share/doc/libfoo/ since we cannot install both its and root
# README.md into the same location.
#
doc{*.md}: install = doc/manual/
\

Note that we don't need to make any changes to the root \c{buildfile} since
this subdirectory will automatically get picked up by the \c{{*/\ -build/\}}
name pattern that we have there.

Let's now look at the alternative arrangement with another subdirectory level
inside \c{docs/}. Here we achieve the same result but in a slightly different
way. Specifically, we call the subdirectory \c{manual/} and install recreating
subdirectories (see \l{b#intro-operations-install Installing} for background):

\
$ cd libfoo/ # Change to the package root.
$ mkdir -p docs/manual
$ cd docs/manual/
$ ln -s ../../../upstream/docs/*.md ./
\

And the corresponding \c{buildfile} in \c{docs/}:

\
./: doc{**.md}

# Install the documentation in docs/ into, say, /usr/share/doc/libfoo/
# recreating subdirectories.
#
doc{*}:
{
  install = doc/
  install.subdirs = true
}
\


\h2#core-root-buildfile-commit|Adjust root \c{buildfile}: commit and test|

Once all the adjustments to the root \c{buildfile} are made, it makes sense to
test it locally (this time from the root of the package), commit our changes,
and test with CI:

\
$ cd libfoo/ # Change to the package root.
$ b test
$ bdep test -a
\

If you had to add any extra files to the root \c{buildfile} or add
\c{buildfile} in extra subdirectories, then it also makes sense to test
installation (\l{#core-test-smoke-locally-install Test locally: installation})
and preparation of the source distribution (\l{#core-test-smoke-locally-dist
Test locally: distribution}) and make sure the extra files end up in the right
places.

Then commit our changes and CI:

\
$ cd foo/ # Change to the package repository root.
$ git add .
$ git status
$ git commit -m \"Adjust root buildfile\"
$ git push

$ bdep ci
\


\h2#core-root-manifest|Adjust \c{manifest}|

The last file we need to look over is the package's \c{manifest}. Here is what
it typically looks like, using our \c{libfoo} as an example:

@@ TODO: regenerate with final version.

\
: 1
name: libfoo
version: 0.1.0-a.0.z
language: c++
project: foo
summary: C++ library implementing secure Foo protocol
license: MIT ; MIT License.
description-file: README.md
url: https://example.org/foo
email: boris@codesynthesis.com
#build-error-email: boris@codesynthesis.com
depends: * build2 >= 0.16.0
depends: * bpkg >= 0.16.0
#depends: libhello ^1.0.0
\

You can find the description of these and other package \c{manifest} values in
the \l{bpkg#manifest-package Package Manifest} section of the \l{bpkg The
\c{build2} Package Manager} manual.

In the above listing the values that we likely need to adjust are \c{summary}
and \c{license} (@@ see sections below), unless correctly auto-detected by
\c{bpkg-new} on the \l{#core-package-create Create final package} step, as
well as \c{url}, \c{email}, @@ TODO: update with final list (\c{package-*},
etc). Also make links.

You may also want to add the following value in certain cases:

\dl|

\li|\l{bpkg#manifest-package-changes \cb{changes-file}}

If you have added a news of change log file to the root \c{buildfile} (see
\l{#core-root-buildfile Adjust root buildfile}), then it also makes sense
to list it in the \c{manifest}. For example:

\
changes-file: NEWS
\

|

\li|\l{bpkg#manifest-package-topics \cb{topics}}

Package topics. For example:

\
topics: network protocol, network security
\

\N|If the upstream project is hosted on GitHub or similar, then you can
usually copy the topics from the upstream repository description.||

\li|\l{bpkg#manifest-package-doc-url \cb{doc-url}}\n
    \l{bpkg#manifest-package-src-url \cb{src-url}}

Documentation and source code URLs. For example:

\
doc-url: https://example.org/foo/doc/
src-url: https://github.com/.../foo
\

||


\h2#core-root-manifest-license|Adjust \c{manifest}: \c{license}|

For \c{license}, use the \l{https://spdx.org/licenses/ SPDX license ID} if at
all possible. If multiple licenses are involved, use the SPDX License
expression. See the
\l{https://build2.org/bpkg/doc/build2-package-manager-manual.xhtml#manifest-package-license
\c{license} manifest value} documentation for details, including the list of
the SPDX IDs for the commonly used licenses.


\h2#core-root-manifest-summary|Adjust \c{manifest}: \c{summary}|

For \c{summary} use a brief description of the functionality provided by the
package. Less than 70 characters is a good target to aim for. Don't capitalize
subsequent words unless proper nouns and omit the trailing dot. For example:

\
summary: Vim xxd hexdump utility
\

Omit weasel words such as \"modern\", \"simple\", \"fast\", \"small\", etc.,
since they don't convey anything specific. Omit \"header-only\" or
\"single-header\" for C/C++ libraries since, at least in the context of
\c{build2}, it does not imply any advantage.

If upstream does not offer a sensible summary, the following template is
recommended for libraries:

\
summary: <functionality> C library
summary: <functionality> C++ library
\

For example:

\
summary: Event notification C library
summary: Validating XML parsing and serialization C++ library
\

If the project consists of multiple packages it may be tempting to name each
package in terms of the overall project name, for example:

\
summary: libigl's core module
\

This doesn't give the user any clue about what functionality is provided
unless they find out what \c{libigl} is about. Better:

\
summary: Geometry processing C++ library, core module
\

If you follow the above pattern, then to produce a summary for external tests
or examples packages simply add \"tests\" or \"examples\" at the end,
for example:

\
summary: Event notification C library tests
summary: Geometry processing C++ library, core module examples
\


@@ Maybe we should start with pre-release of upstream version? If want
   strict versioning?

@@ Maybe re-CI at the ends?

@@ Next section: release and publish (see README in build2-packaging/).
@@ GH issue #?? has some notes.

========

@@ Add example of propagating config.libfoo.debug to macro on build options?

@@ Note on library metadata where talk about configuration. Also about
   autoconf.

@@ Use of the version module and non-semver versions? Links to HOWTO entries!

@@ The 'Don't write buildfiles by hand entry' is now mostly duplicate/redundant.

======================================================================


\h1#dont-do|What Not to Do|

@@ Reorder.

\h#dont-fix-upstream|Don't try to fix upstream issues in the package|

@@ TODO

- support officially unsupported platforms/compiler
- suppress warnings

Any deviations from upstream makes it more difficult to maintain. If your
package makes a large number of changes to upstream, releasing a new version
will require a lot of work.


\h#dont-from-scratch|Don't write \c{buildfiles} from scratch, use \c{bdep-new}|

Unless you have good reasons not to, create the initial project layout
automatically using \l{bdep-new(1)}, then tweak it if necessary and fill with
upstream source code.

The main rationale here is that there are many nuances in getting the build
right and auto-generated \c{buildfiles} had years of refinement and
fine-tuning. The familiar structure also makes it easier for others to
understand your build, for example while reviewing your package submission.

The \l{bdep-new(1)} command supports a wide variety of
\l{bdep-new.xhtml#src-layout source layouts}. While it may take a bit of time
to understand the customization points necessary to achieve the desired layout
for your first package, this will pay off in spades when you work on
converting subsequent packages. The recommended sequence of steps is
as follows:

\ol|

\li|Study the upstream source layout. We want to stay as close to upstream as
possible since this has the best chance of producing an issues-free result
(see \l{#dont-change-upstream Don't change upstream source code layout
unnecessarily} for details).|

\li|Craft and execute the \l{bdep-new(1)} command line necessary to achieve
the upstream layout.|

\li|Study the auto-generated \c{buildfiles} for things that don't fit and need
to change. But don't rush to start manually editing the result. First get an
overview of the required changes and then check if it's possible to achieve
these changes automatically using one of \l{bdep-new(1)} sub-options.

For example, if you see that the generated project assumes the wrong C++ file
extensions, these can be changed with \c{--lang|-l} sub-options.|

\li|Once you have squeezed as much as possible out of \l{bdep-new(1)}, it's
time for manual customizations. These would normally include:

\ul|

\li|Replace generated source code with upstream, normally as symlinks from the
    \c{upstream/} \c{git} submodule.|

\li|Tweak source subdirectory \c{buildfile} that builds the main target
    (library, executable).|

\li|Add tests and, if necessary, examples.|

\li|Tweak \c{manifest} (in particular the \c{version}, \c{summary}, and
    \c{license} values).|

\li|Fill in \c{README.md}.|||

|


\h#dont-change-upstream|Don't change upstream source code layout unnecessarily|

It's a good idea to stay as close to the upstream's source code layout as
possible. For background and rationale, see \l{#core-package-struct Decide on
the package source code layout}.


\h#dont-forget-update-manifest|Don't forget to update \c{manifest} values|

After \l{#dont-from-scratch generating the project template with \c{bdep-new}},
don't forget to update at least the key values in the generated \c{manifest}:
\l{#dont-forget-update-manifest-version \c{version}},
\l{#dont-forget-update-manifest-license \c{license}}, and
\l{#dont-forget-update-manifest-summary \c{summary}}.


\h2#dont-forget-update-manifest-version|Don't forget to update \c{manifest} value \c{version}|

For \c{version}, use the upstream version directly if it is semver (or
semver-like, that is, has three version components). Otherwise, see
\l{https://github.com/build2/HOWTO/blob/master/entries/handle-projects-which-dont-use-semver.md
How do I handle projects that don't use semantic versioning?} and
\l{https://github.com/build2/HOWTO/blob/master/entries/handle-projects-which-dont-use-version.md
How do I handle projects that don't use versions at all?}


\h2#dont-forget-update-manifest-license|Don't forget to update \c{manifest} value \c{license}|

\h2#dont-forget-update-manifest-summary|Don't forget to update \c{manifest} value \c{summary}|

\h#dont-header-only|Don't make library header-only if it can be compiled|

Some libraries offer two alternative modes: header-only and compiled. Unless
there are good reasons not to, a \c{build2} build of such a library should use
the compiled mode.

\N|Some libraries use the \i{precompiled} term to describe the non-header-only
mode. We don't recommend using this term in the \c{build2} build since it has
a strong association with precompiled headers and can therefore be
confusing. Instead, use the \i{compiled} term.|

The main rationale here is that a library would not be offering a compiled
mode if there were no benefits (usually faster compile times of library
consumers) and there is no reason not to take advantage of it in the
\c{build2} build.

There are, however, reasons why a compiled mode cannot be used, the most
common of which are:

\ul|

\li|The compiled mode is not well maintained/tested by upstream and therefore
offers inferior user experience.|

\li|The compiled mode does not work on some platforms, usually Windows due to
the lack of symbol export support (but see \l{b##cc-auto-symexport Automatic
DLL Symbol Exporting}).|

\li|Uses of the compiled version of the library requires changes to the
library consumers, for example, inclusion of different headers.|

|

If a compiled mode cannot be always used, then it may be tempting to support
both modes potentially making the mode user-configurable. Unless there are
strong reasons to, you should resist this temptation and, if the compiled
mode is not universally usable, then use the header-only mode everywhere.

The main rationale here is that variability adds complexity which makes the
result more prone to bugs, more difficult to use, and harder to review and
maintain. If you really want to have the compiled mode, then the right
way to do it is to work with upstream to fix any issues that prevent its
use in \c{build2}.

There are, however, reasons why supporting both mode may be needed, the most
common of which are:

\ul|

\li|The library is widely used in both modes but switching from one mode to
the other requires changes to the library consumers (for example, inclusion of
different headers). In this case only supporting one mode would mean not
supporting a large number of library consumers.|

\li|The library consists of a large number of independent components and its
common for applications to only use a small subset of them. On the other hand,
compiling all of them in the compiled mode takes a substantial amount of time.
(Note that this can also be addressed by making the presence of optional
components user-configurable.)|

|


\h#dont-main-target-root-buildfile|Don't build your main targets in root \c{buldfile}|

It may be tempting to have your main targets (libraries, executables) in the
root \c{buildfile}, especially if it allows you to symlink entire directories
from \c{upstream/} (which is not possible if you have to have a \c{buildfile}
inside). However, this is a bad idea except for the simplest projects.

Firstly, this quickly gets messy since you have to combine managing
\c{README}, \c{LICENSE}, etc., and subdirectories with you main target builds.
But, more importantly, this means that when you main target is imported (and
thus the \c{buildfile} that defines this target must be loaded), your entire
project will be loaded, including any \c{tests/} and \c{examples/} subproject,
which is wasteful.


If you want to continue symlinking entire directories from \c{upstream/} but
without moving everything to the root \c{buildfile}, the recommended approach
is to simply add another subdirectory level. Let's look at a few concrete
example to illustrate the technique (see \l{#core-package-struct Decide on the
package source code layout} for background on the terminology used).

Here is the directory structure of a package which uses a combined layout (no
header/source split) and where everything is in the root \c{buildfile}:

\
libigl-core/
├── igl/ -> upstream/igl/
├── tests/
└── buildfile                      # Defines lib{igl-core}.
\

And here is the alternative structure where we have added the extra
\c{libigl-core} subdirectory with its own \c{buildfile}:

\
libigl-core/
├── libigl-core/
│   ├── igl/ -> ../upstream/igl/
│   └── buildfile                 # Defines lib{igl-core}.
├── tests/
└── buildfile
\

Below is the \c{bdep-new} invocation that can be used to automatically
create this alternative structure (see \l{#core-package-craft-cmd Craft
\c{bdep\ new} command line to create package} for background and
\l{bdep-new(1)} for details):

\
$ bdep new \
 --type lib,prefix=libigl-core,subdir=igl,buildfile-in-prefix \
 libigl-core
\

Let's also look at an example of a split layout, which may require a slightly
different \c{bdep-new} sub-options to achieve the same result. Here is the
layout which matched upstream exactly:

\
$ bdep new --type lib,split,subdir=foo,no-subdir-source libfoo
$ tree libfoo
libfoo/
├── include/
│   └── foo/
│       ├── buildfile
│       └── ...
└── src/
    ├── buildfile
    └── ...
\

However, with this layout we will not be able to symlink the entire
\c{include/foo/} and \c{src/} subdirectories because there are \c{buildfiles}
inside (and which may tempt you to just move everything to the root
\c{buidfile}). To fix this we can move the \c{buildfiles} out of source
subdirectory \c{foo/} and into prefixes (\c{include/} and \c{src/}) using the
\c{buildfile-in-prefix} sub-option. And since \c{src/} doesn't have a source
subdirectory, we have to invent one:

\
$ bdep new --type lib,split,subdir=foo,buildfile-in-prefix libfoo
$ tree libfoo
libfoo/
├── include/
│   ├── foo/ -> ../upstream/include/foo/
│   └── buildfile
└── src/
    ├── foo/ -> ../upstream/src/
    └── buildfile
\


\h1#howto|Packaging HOWTO|

@@ howto make smoke test (and fix ref). Actually, we now have a step for
this.


\h#howto-debug-macro|How do I expose extra debug macros of a library|

Sometime libraries provide extra debugging facilities that are usually enabled
or disabled with a macro. For example, \c{libfoo} may provide the
\c{LIBFOO_DEBUG} macro that enables additional sanity checks, tracing, etc.
Normally such facilities are disable by default.

While it may seem like a good idea to detect a debug build and enable this
automatically, it is not: such facilities usually impose substantial overhead
and the presence of debug information does not mean that performance is not
important (people routinely make optimized builds with debug information).

As a result, the recommended approach is to expose this as a configuration
variable that the end-users of the library can use (see \l{b#proj-config
Project Configuration} for background). Continue with the \c{libfoo} example,
we can add \c{config.libfoo.debug} to its \c{build/root.build}:

\
# build/root.build

config [bool] config.libfoo.debug ?= false
\

And then define the \c{LIBFOO_DEBUG} macro based on that in the \c{buildfile}:

\
# src/buildfile

if $config.libfoo.debug
  cxx.poptions += -DLIBFOO_DEBUG
\

If the macro is also used in the library's interface (for example, in inline
or template functions), then we will also need to export it:

\
# src/buildfile

if $config.libfoo.debug
{
  cxx.poptions += -DLIBFOO_DEBUG
  lib{foo}: cxx.export.poptions += -DLIBFOO_DEBUG
}
\

\N|If the debug facility in question should be enabled by default even in the
optimized builds (in which case the macro usually has the \c{NO_DEBUG}
semantics), the other option is to hook it up to the standard \c{NDEBUG}
macro, for example, in the library's configuration header file.|

Such \c{.debug} configuration variables should primarily be meant for the
end-user to selectively enabled extra debugging support in certain libraries
of their build. However, if your project depends on a number of libraries with
such extra debuggin support and it generally makes sense to also enable this
support in dependencies if it is enabled in your project, then you may want to
propagate your \c{.debug} configuration value to the dependencies (see the
\l{bpkg#manifest-package-depends \c{depends} package \c{manifest} value} for
details on dependency configuration). You, however, should still allow the
user to override this decision on the per-dependency basis.

Continuing with the above example, let's say we have \c{libbar} with
\c{config.libbar.debug} that depends on \c{libfoo} and that wishes to by
default enable debugging in \c{libfoo} if it is enabled in \c{libbar}.
This is how we can correctly arrange for this in \c{libbar}'s \c{manifest}:

\
depends:
\\
libfoo ^1.2.3
{
  # We prefer to enable debug in libfoo if enabled in libbar
  # but accept if it's disabled (for example, by the user).
  #
  prefer
  {
    if $config.libbar.debug
      config.libfoo.debug = true
  }

  accept (true)
}
\\
\


\h#howto-patch-upstream-source|How do I patch upstream source code|

@@ TODO

\h#howto-bad-inclusion-practice|How do I deal with bad header inclusion practice|

This sections explains how to deal with libraries that include their public,
generically-named headers without the library name as directory prefix. Such
libraries cannot coexist, neither in the same build nor when installed. For
background and details, see \l{intro#proj-struct Canonical Project Structure}.

@@ TODO


\h#howto-extra-header-install-subdir|How do I handle extra header installation subdirectory|

This sections explains how to handle an additional header installation
subdirectory.

@@ TODO


\h#howto-no-extension-header|How do I handle headers without extensions|

If all the headers in a project have no extension, then you can simply
specify the empty \c{extension} value for the \c{hxx{\}} target type
in \c{build/root.build}:

\
hxx{*}: extension =
cxx{*}: extension = cpp
\

Note, however, that using wildcard patterns for such headers in your
\c{buildfile} is a bad idea since such a wildcard will most likely pick up
other files that also have no extension (such as \c{buildfile}, executables on
UNIX-like systems, etc). Instead, it's best to spell the names of such headers
explicitly. For example, instead of:

\
lib{hello}: {hxx cxx}{*}
\

Write:

\
lib{hello}: cxx{*} hxx{hello}
\

If only some headers in a project have no extension, then it's best to specify
the non-empty extension for the \c{extension} variable in \c{build/root.build}
(so that you can still use wildcard for headers with extensions) and spell out
the headers with no extension explicitly. Continuing with the above example,
if we have both the \c{hello.hpp} and \c{hello} headers, then we can handle
them like this:

\
hxx{*}: extension = hpp
cxx{*}: extension = cpp
\

\
lib{hello}: {hxx cxx}{*} hxx{hello.}
\

Notice the trailing dot in \c{hxx{hello.\}} \- this is the explicit \"no
extension\" specification. See \l{b#targets Targets and Target Types}
for details.


\h1#faq|Packaging FAQ|

\h#faq-alpha-stable|Why is my package in \c{alpha} rather than \c{stable}?|

If your package uses a semver version (or semver-like, that is, has three
version components) and the first component is zero (for example, \c{0.1.0}),
then, according to the semver specification, this is an alpha version and
\l{bdep-publish(1)} automatically published such a version to the \c{alpha}
section of the repository.

Sometimes, however, in a third-party package, while the version may look like
semver, upstream may not assign the zero first component any special
meaning. In such cases you can override the \c{bdep-publish} behavior with the
\c{--section} option, for example:

\
$ bdep publish --section=stable
\

Note that you should only do this if you are satisfied that by having the zero
first component upstream does not imply alpha quality.


\h#faq-publish-stage|Where to publish if package requires staged toolchain?|

If your package requires the \l{https://build2.org/community.xhtml#stage staged
toolchain}, for example, because it needs a feature or bugfix that is not yet
available in the released toolchain, then you won't be able to publish it to
\c{cppget.org}. Specifically, if your package has the accurate \c{build2}
version constraint and you attempt to publish it, you will get an error like
this:

\
error: package archive is not valid
  info: unable to satisfy constraint (build2 >= 0.17.0-) for package foo
  info: available build2 version is 0.16.0
\

There are three alternative ways to proceed in this situation:

\ol|

\li|Wait until the next release and then publish the package to
\c{cppget.org}.|

\li|If the requirement for the staged toolchain is \"minor\", that is, it
doesn't affect the common functionality of the package or only affects a small
subset of platforms/compilers, then you can lower the toolchain version
requirement and publish the package to \c{cppget.org}. For example, if
you require the staged toolchain because of a bugfix that only affects
one platform, it doesn't make sense to delay publishing the package
since it is perfectly usable on all the platforms in the meantime.|

\li|Publish it to \l{https://queue.stage.build2.org queue.stage.build2.org},
the staging package repository. This repository contain new packages that
require the staged toolchain to work and which will be automatically
moved to \c{cppget.org} once the staged version is released. The other
advantage of publishing to this repository (besides not having to remember
to manually publish the package once the staged version is released) is
that your package becomes available from an archive repository (which is
substantially faster than a \c{git} repository).

To publish to this repository, use the following \c{bdep-publish} command
line:

\
$ bdep publish --repository=https://stage.build2.org ...
\

||


"