aboutsummaryrefslogtreecommitdiff
path: root/bpkg/repository-types.cli
blob: 1692a134050001532ae9f83d676d351059418c26 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
// file      : bpkg/repository-types.cli
// license   : MIT; see accompanying LICENSE file

include <bpkg/common-options.hxx>;

"\section=1"
"\name=bpkg-repository-types"
"\summary=repository types, structure, and URLs"

"
\h|DESCRIPTION|

This help topic describes the repository types recognized by \cb{bpkg}, their
structure, and the format of their URLs. Currently three types of repositories
are supported: archive-based \cb{pkg}, directory-based \cb{dir}, and version
control-based \cb{git}.

The repository location may specify the repository type as part of the URL
scheme component in the \c{\i{type}\b{+}\i{protocol}} form. For example:

\
git+https://example.com/foo
dir+file:///tmp/repo
\

Note that the explicit specification is only needed when the correct type
cannot be guessed from the URL. See \l{bpkg-rep-add(1)} for details.


\h|PKG REPOSITORIES|

A \cb{pkg} repository is \i{archive}-based. That is, it contains a collection
of various packages/versions as archive files. For more information on the
structure of \cb{pkg} repositories refer to \l{bpkg The \cb{build2} Package
Manager Manual}. The \cb{pkg} repository location can be a local directory
path or an \cb{http(s)://} URL.


\h|DIR REPOSITORIES|

A \cb{dir} repository is \i{directory}-based. That is, it contains a
collection of various packages as directories but only a single version per
package can be present in such a repository. The \cb{dir} repository location
can be a local directory path or a \cb{file://} URL.

A \cb{dir} repository is expected to contain either the \cb{manifest} or
\cb{packages.manifest} file in the root directory of the repository. If it
only contains \cb{manifest}, then it is assumed to be a simple, single-package
repository with the \cb{manifest} file being its package manifest. Otherwise,
the \cb{packages.manifest} file should list the locations of available
packages as described in \l{bpkg#manifest-package-list-dir Package List
Manifest for \cb{dir} Repositories}.

A \cb{dir} repository may also contain the \cb{repositories.manifest} file in
the root directory of the repository. This file can be used to describe the
repository itself as well as specify its prerequisite and complement
repositories. See \l{bpkg#manifest-repository-list Repository List Manifest}
for details on the format and semantics of this file.


\h|GIT REPOSITORIES|

A \cb{git} repository is \i{version control}-based. That is, it normally
contains multiple versions of the same package (but can also contain several,
usually related, packages in the same repository).

A \cb{git} repository has the same structure and manifest files as the
\cb{dir} repository. See \l{bpkg#manifest-package-list-dir Package List
Manifest for \cb{dir} Repositories} and \l{bpkg#manifest-repository-list
Repository List Manifest} for details on their format and semantics.

Theoretically, a \cb{git} repository may contain as many package versions as
there are commits. Practically, however, we are normally only interested in a
small subset of them while fetching and processing the necessary information
for all of them could be prohibitively expensive.  As a result, by default,
only advertised tags in the \cb{refs/tags/v*} form where the part after \cb{v}
is also a valid \l{b#module-version standard version} are considered to be
sources of useful package versions. These commits normally correspond to
released versions and are called the \i{default set}. Note that only the
latest revision of each such version is considered.

Instead of the default set, it is possible to provide a custom set of
available versions by specifying one or more commit ids and/or references
and/or reference patterns in the repository URL fragment (see
\cb{git-ls-remote(1)} for details on advertised references). For example:

\
https://example.com/foo.git#v1.2.3
https://example.com/foo.git#master
https://example.com/foo.git#af234f56
https://example.com/foo.git#tags/releases/*
https://example.com/foo.git#HEAD,tags/v1.*.*,heads/feature-*
\

Furthermore, it is possible to expand (or narrow down) the default set using
the special \cb{##} fragment notation. For example:

\
https://example.com/foo.git##HEAD     - default set plus HEAD
https://example.com/foo.git##heads/*  - default set plus branches
https://example.com/foo.git##-v1.*    - default set minus v1.*
\

A \cb{git} repository URL fragment is a comma-separated list of reference
filters in the following form:

\c{[\i{refname}][\b{@}\i{commit}]}

Either \ci{refname}, \ci{commit}, or both must be specified. If both are
specified then \ci{refname} is only used to minimize the amount of data
fetched and \ci{commit} is expected to belong to its history. For example:

\
.../foo.git#master@48fba3625d65941bb85a39061bcf795d4949c778
\

The \ci{refname} part can be an abbreviated commit id or an advertised
reference or reference pattern under \cb{refs/}. While \ci{commit} must be the
complete, 40-characters SHA1 that need not be advertised. For convenience, a
40-characters filter that consists of only hexadecimal digits is assumed to be
\ci{commit} even if not prefixed with \cb{@}. In an unlikely event this
produces an incorrect result, the \cb{@}-form with omitted \ci{commit} can be
used. For example:

\
.../foo.git#48fba3625d65941bb85a39061bcf795d4949c778   (commit id)
.../foo.git#deadbeefdeadbeefdeadbeefdeadbeefdeadbeef@  (reference)
\

The \ci{refname} part can use the \cb{*} and \cb{?} wildcard pattern
characters with the standard semantics as well as the \cb{**} character
sequence which matches in subdirectories, recursively. For example:

\
.../foo.git#tags/v*    - tags/v1.2.3 but not tags/old/v0.1.0
.../foo.git#tags/v**   - tags/v1.2.3 and tags/old/v0.1.0
\

A relative \ci{refname} is searched for in \cb{refs/}, \cb{refs/tags/}, and
\cb{refs/heads/} as well as among symbolic references like \cb{HEAD}. To
anchor it to \cb{refs/} make it absolute, for example:

\
.../foo.git#tags/v*   - refs/tags/v1.2.3 but also refs/heads/tags/voo
.../foo.git#/tags/v*  - refs/tags/v1.2.3 only
\

While a \ci{refname} pattern is allowed not match any references, a
non-pattern that doesn't resolve to a reference is invalid.

If a \ci{refname} starts with minus (\cb{-}) then it is treated as an
exclusion filter \- any references that it matches are excluded from the set
included by the preceding filters (or the default set). For example:

\
.../foo.git#v*,-v1.*  - exclude v1.* from v*
.../foo.git##-v1.*    - exclude v1.* from default set
\

To support specifying literal leading minus, a \ci{refname} that starts with
plus (\cb{+}) is treated as an inclusion filter. For example:

\
.../foo.git#+x   - include  x
.../foo.git#+-x  - include -x
.../foo.git#++x  - include +x
\

Currently supported \cb{git} protocols are \cb{git://}, \cb{ssh://} (but not
\c{scp} pseudo-URL syntax), \cb{http://}, and \cb{https://} for remote
repositories and \cb{file://} for local repositories. While \cb{bpkg} tries to
minimize the amount of information (history) fetched, it is not always
possible for some protocols and/or server configurations, as discussed next.

A \cb{git} repository accessible via \cb{http(s)://} can use either \i{dumb}
or \i{smart} protocol (refer to the \cb{git} documentation for details). The
dumb protocol provides only limited support for fetch minimization and if this
protocol is used, then \cb{bpkg} has no choice but to download a substantial
amount of history.

The smart protocol allows fetching of minimal history for tags and
branches. Whether this is also possible for (all) commit ids depends on
whether the server is configured to allow fetching unadvertised commits. For
details, refer to the \cb{uploadpack.allowReachableSHA1InWant} and
\cb{uploadpack.allowAnySHA1InWant} \cb{git} configuration values.

The \cb{git://} and \cb{ssh://} protocols are similar to smart \cb{http://} in
that they support fetching minimal history for tags and branches and may or
may not support this for commit ids depending on the server configuration.
Note, however, that unlike for \cb{http(s)://}, for these protocols \cb{bpkg}
does not try to sense if fetching unadvertised commits is allowed and always
assumes that it is not. Also note that the sensed or assumed protocol
capabilities can be overridden for a \cb{git} repository URL prefix using the
\cb{--git-capabilities} option (\l{bpkg-common-options(1)}).

Based on this information, to achieve optimal results the recommended protocol
for remote repositories is smart \cb{https://}. Additionally, if you are
planning to refer to unadvertised commit ids, then also consider configuring
the server to allow fetching unadvertised commits.

The \cb{file://} protocol has the same fetch minimization support as
\cb{git://} and is therefore treated the same.
"