Forge-hosted projects packaging automation

Projects published on a “forge” can be packaged using the forgemeta macro.

What is a “forge” for the forgemeta macro?
Any software publishing website, permitting the download of source code archives on normalized URLs, that can be deduced from a project root URL on the website plus version/commit/tag/scm/extension values, is a “forge”.

Links :

• FPC ticket
• fedora-rpm-macros RFE with the macro file
• packaging@lists.fedoraproject.org discussion
• pagure.io RFE to make pagure.io compatible with forgemeta
• GitLab CE RFE to remove GitLab quirks
• rpm RFE to get rid of the remaining %setup special cases

Usage

A. The packager declares upstream-dependent metadata :

• forgeurl: the project URL on the target forge
• version: the project release to package, if non nil (as Version: xxx)
• commit: the commit hash to package, if any
• tag: the project tag to package, if any

B. The packager calls the forgemeta macro to compute rpm-oriented metadata:

• forgesource: the corresponding source file URL, that can be used as Source:
• setupargs: the arguments that %setup will need
• archiveurl: the archive URL, without renaming
• archivename: the corresponding archive name (often used by setupargs)
• archiveext: the corresponding archive extension
• shortcommit: a commit hash reduction
• dist: auto-adjusted when packaging a tag or commit

C. (Optionnal) The packager calls the forgemetacheck macro to display the values of computed variables D. The packager uses the computed metadata as needed.

Packaging examples

Packaging a release (except for GitLab)

%global forgeurl    https://github.com/foo/bar
Version:            8.18.1
 
%forgemeta

Name:    foobar
Release: 1%{?dist}
…
URL:    %{forgeurl}
Source: %{forgesource}
…
%prep
%setup  %{?setupargs}

Packaging a release (GitLab)

GitLab is inconvenient

• Any download from GitLab requires knowledge of the corresponding commit.
• Release download depends on a Git tag, which is free-form. If your release tag value does not match Version (for example, if it is v%{version}), you also need to declare tag before calling forgemeta.

%global forgeurl    https://gitlab.example.com/foo/bar
%global commit      2a810629566a2d0f0d4107df244e8828b9f7bd5c
Version:            8.18.1
 
%forgemeta

Name:    foobar
Release: 1%{?dist}
…
URL:    %{forgeurl}
Source: %{forgesource}
…
%prep
%setup  %{?setupargs}

Packaging a specific commit

%global forgeurl    https://code.googlesource.com/foobar
%global commit      2a810629566a2d0f0d4107df244e8828b9f7bd5c

%forgemeta

Name:    foobar
Version: 0
Release: 0.1%{?dist}
…
URL:    %{forgeurl}
Source: %{forgesource}
…
%prep
%setup  %{?setupargs}

Packaging a release accessed through a tag

%global forgeurl    https://github.com/foo/bar
Version:            0.10.0
%global tag         %{version}
 
%forgemeta

Name:    foobar
Release: 1%{?dist}
…
URL:    %{forgeurl}
Source: %{forgesource}
…
%prep
%setup  %{?setupargs}

Packaging a pre-release commit

On Version declaration order
It is generally safe to declare Version before calling %forgemeta even for pre or post-releases. Except for GitLab. Due to GitLab's use of commit for release downloads, you MUST declare Version after calling %forgemeta when the packaged commit does not match the release commit.

%global forgeurl     https://gitlab.example.com/foo/bar/
%global commit       51637bc0960002b811e1c0c7be8671cf9a1cc5be
 
%forgemeta

Name:    foobar
Version: 3.2.6
Release: 0.1%{?dist}
…
URL:    %{forgeurl}
Source: %{forgesource}
…
%prep
%setup %{?setupargs}

Post-release packaging is similar. See also Packaging:Versioning#Release_and_post-release_versions.

%forgemetacheck output examples

On variable resolution
forgemetacheck resolves the values of rpm variables at one point of the spec file. Since most of those variables are themselves composed of other variables, it is possible for those values to change if one of the variables used in the composition is overridden by the packager after the forgemetacheck call.

forgemetacheck emits rpm warnings. It should be safe to call from any point of the spec file.

Forge-specific packaging variables
  forgeurl:        “https://github.com/satori/uuid”
  forgesource:     “https://github.com/satori/uuid/archive/v1.0.0.tar.gz#/go-uuid-1.0.0.tar.gz”
  shortcommit:     “”
Generic variables
  setupargs:       “-n go-uuid-1.0.0”
  archivename:     “go-uuid-1.0.0”
  archiveext:      “tar.gz”
  archiveurl:      “https://github.com/satori/uuid/archive/v1.0.0.tar.gz”
  scm:             “”
  tag:             “”
  commit:          “”
  dist:            “.fc28” (snapshot date is computed once the corresponding source file is available)

Forge-specific packaging variables
  forgeurl:        “https://github.com/apache/thrift”
  forgesource:     “https://github.com/apache/thrift/archive/0.10.0.tar.gz#/thrift-0.10.0.tar.gz”
  shortcommit:     “”
Generic variables
  setupargs:       “-n thrift-0.10.0”
  archivename:     “thrift-0.10.0”
  archiveext:      “tar.gz”
  archiveurl:      “https://github.com/apache/thrift/archive/0.10.0.tar.gz”
  scm:             “git”
  tag:             “0.10.0”
  commit:          “”
  dist:            “.fc28” (snapshot date is computed once the corresponding source file is available)

Forge-specific packaging variables
  forgeurl:        “https://code.googlesource.com/google-api-go-client”
  forgesource:     “https://code.googlesource.com/google-api-go-client/+archive/3a1d936b7575b82197a1fea0632218dd07b1e65c.tar.gz#/google-api-go-client-3a1d936b7575b82197a1fea0632218dd07b1e65c.tar.gz”
  shortcommit:     “”
Generic variables
  setupargs:       “-c”
  archivename:     “google-api-go-client-3a1d936b7575b82197a1fea0632218dd07b1e65c”
  archiveext:      “tar.gz”
  archiveurl:      “https://code.googlesource.com/google-api-go-client/+archive/3a1d936b7575b82197a1fea0632218dd07b1e65c.tar.gz”
  scm:             “git”
  tag:             “”
  commit:          “3a1d936b7575b82197a1fea0632218dd07b1e65c”
  dist:            “.git3a1d936.fc28” (snapshot date is computed once the corresponding source file is available)

Extending the macro

If the project you're packaging is published on a software publishing service forgemeta has no knowledge of, don't be sad, that's pretty easy to fix.

• note down the archive URLs you want to generate for versions, tags and commits
• locate the latest version of the forgemeta macro (it should be installed in /usr/lib/rpm/macros.d/macros.forge-srpm by fedora-rpm-macros)
• copy the definition block closest to your needs after the other definition blocks. It should look like :

 if (forge == "xxx") then
 …
 end

• change the "xxx" value
• adapt the initial Lua pattern test to guard against URLs you don't know how to handle:

  if not ( string.match(forgeurl, "'yyy") ) then
     error("xxx urls must match “zzz”!")
  end

Don't be afraid of Lua patterns!
While the Lua pattern syntax is a bit unusual, it is quite simple. You can launch an interactive interpreter to experiment in with the lua command. Most of the times the existing forgemeta macro will already contain something similar to what you need to accomplish.

• scrap the variables you need from forgeurl using:

  local myvariable = string.match(forgeurl, "abc")

• use them to define archivename and archiveurl in the following if tag/commit/version block
• install the result in /usr/lib/rpm/macros.d/
• test with forgemetacheck and rpmbuild -bs myspecfile.spec
• submit the enhancement for inclusion in fedora-rpm-macros once it works satisfactorily.

Benefits and limitations

• Benefits:

1. Fedora packagers no longer need to know about the URL structure of well-known forges.
2. Spec files are simpler, less error-prone, easier to maintain and audit.
3. Forge URLs can be defined and fixed in a single place, without waiting for guidelines to percolate.
4. The macros are mostly written in Lua, making them more verbose but easier to adjust and extend.
5. The macros are written in Lua and can perform error handling.
6. It is very easy to switch from commit to tag to version (or any combination of those).
7. spectool just works© (in Fedora).
8. scm snapshot date is exact and does not rely on a variable which may or may not have been updated.
9. Almost all the computations are done in optional rpm variables that can be ignored by the packager if he does not need or does not like the result.
10. Fedora forge know-how can be capitalized over time.
11. The macro normalizes common spec constructs such as archivename.

• Limitations:

1. The macro is not intended to be used in spec files that package multiple forge archives (it could probably be extended, is the added flexibility worth the complexity?).
2. The macro needs to be updated when a forge changes its structure (but better changing one place than lots of spec files).
3. New forges need to be added to the macro before it can be used with them.
4. dist munging can not be easily reverted once forgemeta has been called. The alternative would be to 1. change fedora-release to define dist as %{?snapid}%{?distroid} and set distroid to the value dist is set today, and 2. let other macros such as forgemeta define snapid later. Cleaner but a lot more invasive and more difficult to propagate downstream.
5. dist commit date computation relies on source files with the correct modification time (incorrect source file modification time → incorrect dist date).
6. spectool may not work in older Fedora derivatives (but just use forgemetacheck and copy the source URL form its output).
7. The macro highlights some historical rpm design mistakes: bad separation of upstream and package metadata, special magic variables. This is why version declaration occurs in an unusual (for rpm) order and requires a specific syntax.

Testing

Just drop in /usr/lib/rpm/macros.d/ the file proposed here for inclusion in fedora-rpm-macros, and play with the result. The file name must be prefixed with “macros.”.