|Age||Commit message (Collapse)||Author||Files||Lines|
Sometimes a fix release may be required which only contains fixes to
parts not related to a shared library, for instance a fix in the build
system. Thus allow releasing without bumping LIBVERSION, but only
through env var setting for people who know what they are doing.
Before this commit, for library projects (containing LIBVERSION in some
Makefile), the entire commit list was not stored into the changelog, but
only a few lines from TODO-RELEASE files.
This is a bad approach for several reasons. First, because that file was
only aimed at containing API/ABI breaks, and not the full relevant
changeset (like bugfixes, new features, etc.). Second, because it relies
on every developer making API/ABI changes to remember to store the
change in there during commit break time.
Let's instead always store the entire commit list in changelog, and
let's use TODO-RELEASE only as a list of hints for the maintainer to
help him evaluate how LIBVERSION needs to be bumped for each library.
Other tools such as osmo-abi-check.git can be used to help with the
process of decission too.
Let's take the opportunity too to only commit stuff already added to the
staging area, as it proved easier to manage from my personal experinece
making latest releases.
It turns out git status doesn't return an error code in any of the
modified/unmodified cases. It's not clear anyway why we check
TODO-RELEASE when we actually care about the file containing the
If no line is provided, then the xargs line doesn't call dch (due to -r
param) and as a result no new version entry is created in the log, and
the old one is updated.
Otherwise, while updating osmo-trx 0.2.0->0.3.0, the version printed in
the changelog by gbp dch is 0.2.1
Make variable substitution calling bumpversion and other commands
introduce by osmo-release.mk has severely slowed down builds.
Inside the makefile we could use $(eval FOO:=$(shell...)) constructs
to have variable substitution only happen once the 'release' target
is executed. However, 'ifeq' and friends don't work with such late
constructs. Let's shift all release action into a helper shell script
that is called from the Makefile instead.
This way we get the best of both worlds: No performance impact during
normal builds, and the convenience of 'make release'.
Modified-by: Max <firstname.lastname@example.org>