---
title: Req doc - Dropping release and changelog from spec file
tags: Requirements Document, Fedora
---

# Getting rid of release and changelog fields in spec files

## Requirements

* Opt-in
* Opt-out
* Changelog generated must be editable
* Not all commits must appear in the changelog
* Changelog and release fields must be easily "guessable" before making a build
* Must not break current RPM/spec files
* Must not break local builds
* Changes should be minimal for packagers
* A build should be traceable to a specific commit at least as easily as now

## Approaches

### Changelog

#### Get the list of builds from the build system

The macro would start by getting the commit hash for each of the previous builds so the
corresponding changelog entries can have the associate <version>-<release>.
- Note: This only covers the <version>-<release> part of the changelog and needs to be combined
with another idea (cf below)
- (-) Heavy reliance on the build system

#### Changelog from last build

The macro starts by getting the old changelog from the last build and only appends to it
the new changes
- (-) This is breaking the "must be able to edit the changelog" requirement
- (-) Heavy reliance on the build system

#### Annotated git tags

Use annotated git tags to store the changelog information.
* (-) This is not getting rid of a changelog, it just moves it from the spec file into git
* (-) Editing the changelog would mean allowing to remove the git tags, thus leading to potential
issue in build reproducibility
* (+) The user can be prompted with a pre-filled changelog they would just need to review/adjust before
doing the tagging operation itself (helps filling the changelog, does not solve editing it)
* (?) How is the release field handled? (cf ideas below)

#### Towncrier-like

Towncrier auto-generates changelog from a collection of "news" files. Everytime someone contributes a change that is worth informing users about, they add a "news" file, which are then gathered together when a release is made.
We could have a similar approach, with a `news` folder in the git repo, and have packager contribute content to it as they make changes that would need to be notified to users.
* (-) This is not getting rid of a changelog, it moves it to a number of small files in git
* (+) Each entry being in a different file, there would not be any conflicts with pull-request and git merges.
* (-) For upstream project, towncrier works because when a release is made, the "news" folder is cleared, for a dist-git environment a solution would be needed to "map" news fragment to releases (sub-folder?)
* (+) Editing the changelog would mean editing the corresponding news fragment
* (-) Maintaining the news folder and fragment would be adding quite some overlap to the workflow
* (?) How is the release field handled? (cf ideas below)

#### Inclusive Magic Keywords

The commit message contains a commit message to tell the process to include its message in the
generate changelog.

The generated changelog can be tweaked using other magic keywords: #ignore <commit hash> or
#replace <commit hash> to either remove or replace a commit message from the generate changelog.

* (-) Fragile, parsing commit messages with regex is not nice
* (-) Requires changes to our packaging guidelines as not all commit message will be associable
to a release, so some entries will not contain the <version>-<release> in the first line.
* (+) Self-contained to git, ie: does not require any information from the build system
* (?) How to regenerate the changelog for the old releases?

#### Exclusive Magic Keywords

The commit message contains a commit message to tell the process to exclude a message in the
generated changelog.

The generated changelog can be tweaked using other magic keywords: #ignore <commit hash> or
#replace <commit hash> to either remove or replace a commit message from the generate changelog.

* (-) Fragile, parsing commit messages with regex is not nice
* (-) Requires changes to our packaging guidelines as not all commit message will be associable
to a release, so some entries will not contain the <version>-<release> in the first line.
* (+) Self-contained to git, ie: does not require any information from the build system
* (?) How to regenerate the release for the old entries?

#### External changelog

The changelog is stored outside of the spec file, in a different, dedicated, file.
* (+) Conflicts in the spec file would no longer occur in the %changelog section
* (-) Conflicts in PRs or within git would occur in this new file
* (-) This is not getting rid of a changelog, it just moves it to a separate file
* (+) No dependency on the build system

#### Exclusive Magic Keyword + external changelog

The current RPM changelog is removed from the current spec file and placed into another
file in dist-git.
The macro gets specified the name of the external changelog.
The macro generates the changelog up until the last commit touching the external changelog file.
A magic keyword can be used to ignore a specific commit (#ignore?).

Editing the changelog ends up doing:
fedpkg generate-changelog
Take the output and put it in the external changelog, tweak it as desired and commit your changes.
The macro will include this external changelog and ignore the entire history before that
commit.

* (+) The external changelog makes it easy for the macro to know when to stop browsing the history
* (+) Edit is fairly straight forward
* (-) Packagers must remember to include the old entries when they touch the external changelog file
* (?) How to get the release of the old entries?

### Release

- Get the release field using the number of tags since the last version change
- (-) Breaks if two builds are triggered from the same commit
-- Note: this is status quo compared to today
- (+) Does not rely on the build system

- Get the release field from the annotation of the git tag
- (-) requires manual work on behalf of the maintainer
- (-) Fragile: this means parsing using a regex the git annotation to extract the release
information
- (-) Breaks if two builds are triggered from the same commit
-- Note: this is status quo compared to today
- (+) Does not rely on the build system

- Compute the release field from the number of successful build since the last version change
- (+) Allows trigger two builds from the same commit without any manual change
- (+) Close the behavior maintainers have currently
- (-) Heavy reliance on the build system
- (-) Requires changes to the guidelines: not all changelog entries will be linked to a release
- (-) Cannot "guess" the next release value without information from the build system

- Compute the release field from the number of commits since the last version change: ``<version>-<commits_number>%{dist}``
- (-) Breaks if two builds are triggered from the same commit
-- Note: this is status quo compared to today
- (+) Does not rely on the build system
- (+) No changes to the guidelines: All changelog entries are linked to a release
- (+) Easy to reproduce locally
- (+) Easy to link a certain build with a specific commit
- (+) Easy to "guess" the next release value before triggering the build
- (-) Upgrade path may be problematic if Fn-1 is upgraded to the same version as Fn with more commits

- Compute the release field from the number of commits since the last version change and number
of successful builds for this ``<version>-<commits_number>.<number of builds for this commit>%{dist}``
- Note: the build number could be hidden for the first build of the release and only added if
a specific commit is built twice
- (+) Allows trigger two builds from the same commit without any manual change
- (-) Relies a little on the build system (can be closely simulated locally)
- (+) No changes to the guidelines: All changelog entries are linked to a release
- (-) A rebuild from the same commit not having an entry in the git history, will not have a
corresponding entry in the changelog
- (-) Not fully to reproducible locally (the number of builds would be missing)
- (+) Easy to link a certain build with a specific commit
- (+) Easy to "guess" the next release value before triggering the build
- (-) Upgrade path may be problematic if Fn-1 is upgraded to the same version as Fn with more commits

- Compute the release field from EVRs of builds that must be sorted lower (i.e. anything in Fedora
versions up to the current one) and EVRs from builds that should be higher, if possible at all
(anything in later Fedora versions)
- (+) Allows triggering two builds from the same commit without manual intervention
- (+) Emulates what many maintainers do manually today for most use cases
- (+) Can cater for pre-releases if done right
(e.g.: provide a macro for the whole release field, and one for <pkgrel> and <minorbump>
as per the versioning guidelines)
- (-) Needs the build system for information about builds in this and other Fedora versions
- (-) Not easy to reproduce locally because we don't have machine-consumable knowledge about
other builds in e.g. the dist-git repo
- (-) Covers new release but does not allow to generate old changelog entries with the release
information