build: keep full git version out of config.h (ccache-friendly) by nmbro · Pull Request #3759 · networkupstools/nut

nmbro · 2026-10-14T19:40:45Z

Follow-up to #3754: instead of an AppVeyor-only CI_AUTOGEN_STABLE_VERSION workaround in ci_build.sh, keep the volatile git-derived version out of include/config.h for everyone (CI, packagers/OBS, local builds).

Problem

include/config.h is included by every source file and carried the full version (e.g. 2.8.5.1466.13) in PACKAGE_VERSION, PACKAGE_STRING and VERSION. These change with every commit, so ccache direct mode missed for every object on each new commit or fresh checkout.

Debug builds with CCACHE_DEBUG=1 across two commits showed:

Only those 3 lines of config.h differ between commits. No C code uses them (only an #if 0 block and #undefs next to net-snmp headers).
Without -g3, ccache's preprocessed mode mostly recovered the hits (364/381 on Linux), at the cost of a preprocessor pass per file.
With -g3 (as on AppVeyor), gcc -E keeps all #define lines in its output, so preprocessed mode misses too, and every object was recompiled. AppVeyor stats before CI: speed up AppVeyor build, fix ccache misses across PRs #3754: preprocessed mode hit 15 times in 1071 lookups.

Change

configure.ac: AC_INIT gets the base SEMVER (e.g. 2.8.5), which only changes with releases, so that is what lands in config.h. Right after AC_INIT, the shell PACKAGE_VERSION/PACKAGE_STRING are set back to the full VER50 value, so all @PACKAGE_VERSION@ substitutions are unchanged (Makefiles, *.pc, packaging recipes, driver.list, tarball name, configure report).
AM_INIT_AUTOMAKE([... no-define]): drops the unused PACKAGE/VERSION macros; VERSION is reset to the full version for distdir.
Programs still report the full version via NUT_VERSION_MACRO in include/nut_version.h, which only common/common-nut_version.c and clients/upsclient.c include.
Second commit: removes the now redundant CI_AUTOGEN_STABLE_VERSION from ci_build.sh (back to its pre-CI: speed up AppVeyor build, fix ccache misses across PRs #3754 form) and appveyor.yml. That workaround still changed whenever master advanced, and it also shortened the substituted version (tarball, *.pc).
Docs: INSTALL.nut.adoc ("Repetitive build/test optimizations"), docs/developers.txt, NEWS.adoc, UPGRADING.adoc.

Testing

Clean build (git clean -fdx, autogen, configure, make with CC="ccache gcc"), empty commit, then the same again: config.h byte-identical; all objects direct hits except the 2 nut_version.h consumers (x2 for libtool).
- WSL Ubuntu 24.04, gcc 13, ccache 4.9: 377/381 hits, all direct.
- ubuntu-plucky container, gcc 14, ccache 4.10: 427/431 hits, all direct.
make distcheck-light, plus building from the resulting tarball without git (VERSION_DEFAULT path): config.h gets 2.8.5, Makefiles get the full version.
NUT_VERSION_FORCED set: the semver goes into config.h, the forced value everywhere else.
autoconf 2.69 on CentOS 7 and Ubuntu Trusty containers: same result (same pre-existing "AC_INIT: not a literal" warning).
make spellcheck, ASCII-only check on all changed text.
Not yet verified: AppVeyor ccache hits without CI_AUTOGEN_STABLE_VERSION. This PR's AppVeyor run should show it (after a warm-up build, since config.h content changes once).

Open question

./configure --version and the config.log header now show the base semver (2.8.5) instead of the full version. If that matters, the alternative is to keep AC_INIT full and strip the defines from confdefs.h before AC_OUTPUT, which is uglier and more fragile.

Checklist

Described the changes in the PR submission
Multiple commits, each with a functional change
Use of coding helper tools and AI disclosed: Claude Code was used for development assistance
Revised for NUT code style, ASCII-only text
Added a bullet point into NEWS.adoc, and UPGRADING.adoc
Passed make spellcheck
make distcheck-light passes

🤖 Generated with Claude Code

github-actions · 2026-10-14T19:43:30Z

A ZIP file with standard source tarball and another tarball with pre-built docs for commit b6ff843 is temporarily available: NUT-tarballs-PR-3759.zip.

jimklimov

Looks like a promising start, thanks! Added a few thoughts as I was reading through this PR, to challenge some assumptions. Feel free to prove me misled :D

Pass a new SEMVER_ACINIT query of gitlog2version.sh to AC_INIT instead of the bare SEMVER: it is the SEMVER for releases, and by the legacy convention a nominal "SEMVER.1" for any development iteration after that release, so the include/config.h macros still tell releases and development builds apart, but do not change between releases and keep build caches valid. It is built from the triplet, so a 4-component NUT_VERSION_DEFAULT (e.g. 2.8.5.1) does not get a second ".1" suffix. For pre-release tagged commits SEMVER is the upcoming release, so the VER50 triplet is used there (unless NUT_VERSION_FORCED_SEMVER is set). Checked with sh, dash and bash for git and NUT_VERSION_DEFAULT values (release, development, tarball and pre-release ones), and that config.h gets 2.8.5.1 while Makefiles keep the full 2.8.5.1503.2. Spellcheck passes with both current and old aspell-en 7.1. Suggested in PR networkupstools#3759 review. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: nmbro <nmbro@users.noreply.github.com>

user01010111 · 2026-10-15T02:54:59Z

Tested bb5de632 on native Linux using the source archive, NUT_VERSION_FORCED, GCC 16.2.1 and ccache 4.14 with -g3 -O2.

A clean rebuild after changing the full version produced 174 direct hits out of 178 cacheable calls. The four misses were exactly common-nut_version.c and upsclient.c, each built twice by libtool. config.h stayed byte-identical, and upsc -V reported the new full version. The cache behaviour matches the intention in this build.

One test expectation update turned up: make shellcheck-nde, with NDE pointed at the generated enumerator, gave 13 failures and 123 passes. report_debug() now includes SEMVER_ACINIT, but the existing version tests still expect the old diagnostic output.

Keeping the PR’s helper unchanged and updating only those 13 expected diagnostic strings in a temporary test copy gave 136 passes and no failures. Looks like those expectations need updating alongside the helper.

include/config.h is included by every source file, and carried the full git-derived version (e.g. 2.8.5.1466.13) in PACKAGE_VERSION, PACKAGE_STRING and VERSION macros. It changed with every commit, so ccache direct mode missed for all objects on each new commit or fresh CI checkout. Without -g3, preprocessed mode mostly recovered the hits (no code uses these macros), but with -g3 (e.g. AppVeyor) the preprocessed output keeps all #define lines, so every object was recompiled. Pass the base SEMVER (e.g. 2.8.5) to AC_INIT, so config.h only changes with releases, and set the shell PACKAGE_VERSION/PACKAGE_STRING right after it back to the full VER50 value, so all @PACKAGE_VERSION@ substitutions (Makefiles, *.pc files, packaging recipes, tarball names, configure reports) keep the full version. Use AM_INIT_AUTOMAKE no-define (PACKAGE and VERSION macros are unused) and reset VERSION for distdir. Programs still report the full version via NUT_VERSION_MACRO in include/nut_version.h, consumed by two files. Checked with CCACHE_DEBUG=1 builds across two commits (WSL gcc 13, ubuntu-plucky gcc 14): config.h byte-identical, all objects direct hits except upsclient.o and common-nut_version.o. Also checked a dist tarball build (VERSION_DEFAULT), NUT_VERSION_FORCED, and autoconf 2.69 on CentOS 7 and Ubuntu Trusty. Docs go into the "Repetitive build/test optimizations" chapter of INSTALL.nut.adoc. Follow-up to PR networkupstools#3754. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: nmbro <nmbro@users.noreply.github.com>

With the full git version kept out of include/config.h, there is no need for ci_build.sh to force a shortened version into autogen.sh to keep ccache hits across pull requests. The workaround was AppVeyor only, still changed whenever master advanced, and also shortened the substituted @PACKAGE_VERSION@ (tarball, *.pc). Restore the plain autogen.sh call and drop the setting from appveyor.yml and the docs. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: nmbro <nmbro@users.noreply.github.com>

Pass a new SEMVER_ACINIT query of gitlog2version.sh to AC_INIT instead of the bare SEMVER: it is the SEMVER for releases, and by the legacy convention a nominal "SEMVER.1" for any development iteration after that release, so the include/config.h macros still tell releases and development builds apart, but do not change between releases and keep build caches valid. It is built from the triplet, so a 4-component NUT_VERSION_DEFAULT (e.g. 2.8.5.1) does not get a second ".1" suffix. For pre-release tagged commits SEMVER is the upcoming release, so the VER50 triplet is used there (unless NUT_VERSION_FORCED_SEMVER is set). Checked with sh, dash and bash for git and NUT_VERSION_DEFAULT values (release, development, tarball and pre-release ones), and that config.h gets 2.8.5.1 while Makefiles keep the full 2.8.5.1503.2. Spellcheck passes with both current and old aspell-en 7.1. Update the gitlog2version test expectations in tests/nut-driver-enumerator-test.sh for the SEMVER_ACINIT value now included in the debug output, and add test cases for the new query (release, development, 4-component default and release candidate). Suggested in PR networkupstools#3759 review. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: nmbro <nmbro@users.noreply.github.com>

nmbro · 2026-10-15T09:40:39Z

Thanks, fixed in 4878552 (also added SEMVER_ACINIT test cases).

…SEMVER_ACINIT corner case of `X.Y.Z.0(.0...)` [networkupstools#3759] Signed-off-by: Jim Klimov <jimklimov+nut@gmail.com>

… not worth mentioning in these notes [networkupstools#3759] Signed-off-by: Jim Klimov <jimklimov+nut@gmail.com>

jimklimov

Thanks for both PRs!

Add a "GHA-07: ShellCheck" workflow for pull requests, which runs ShellCheck on the shell scripts changed by the PR. The existing scripts have many known findings (about 2900 in the tree, 28 of them errors), so checking the whole tree would always fail. Instead, the new tools/shellcheck-ci.pl helper checks both the base and the PR versions of the changed scripts, and reports only new findings, matched by file, ShellCheck code and source line text (so moved code does not count as new). It fails for new "warning" or "error" level findings; new "note" level (info and style) ones are shown as notices. In GitHub Actions, the findings are shown as annotations on the changed lines and listed in the job summary, with links to the lines and the ShellCheck wiki. It can be run locally against any base revision, including uncommitted changes. The helper is in Perl, like tools/check-source-nonascii.pl (Perl is already needed for autogen.sh in Git workspaces). It only uses core Perl 5 modules, and parses the "gcc" output format of shellcheck, so no JSON module is needed on older systems. The workflow uses only actions/checkout (no new third-party actions) and compares the PR merge commit with its first parent. A new .shellcheckrc file disables the checks for deliberate portability idioms of NUT scripts: backticks (SC2006), x"$var" comparisons (SC2268) and expr (SC2003). It is distributed like .editorconfig, to help the tarball users contribute too. Describe both in docs/qa-guide.adoc, replacing the note that the tool could be introduced at a later time. The .shellcheckrc backticks setting does for ShellCheck what the previous commit did for "make shellcheck" (issue networkupstools#1058). Checked with the "build: keep full git version out of config.h" series (PR networkupstools#3759): its first version is reported for one new SC2166 warning ([ p -a q ]) and the final one passes. Checked renamed, added, removed and non-shell files, a bad base revision, and the workflow with actionlint and zizmor. In a test PR on a fork, deliberate findings showed as annotations on the changed lines. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: nmbro <nmbro@users.noreply.github.com>

…SEMVER_ACINIT corner case of `X.Y.Z.0(.0...)` [networkupstools#3759] Signed-off-by: Jim Klimov <jimklimov+nut@gmail.com>

… not worth mentioning in these notes [networkupstools#3759] Signed-off-by: Jim Klimov <jimklimov+nut@gmail.com>

Add a "GHA-07: ShellCheck" workflow for pull requests, which runs ShellCheck on the shell scripts changed by the PR. The existing scripts have many known findings (about 2900 in the tree, 28 of them errors), so checking the whole tree would always fail. Instead, the new tools/shellcheck-ci.pl helper checks both the base and the PR versions of the changed scripts, and reports only new findings, matched by file, ShellCheck code and source line text (so moved code does not count as new). It fails for new "warning" or "error" level findings; new "note" level (info and style) ones are shown as notices. In GitHub Actions, the findings are shown as annotations on the changed lines and listed in the job summary, with links to the lines and the ShellCheck wiki. It can be run locally against any base revision, including uncommitted changes. The helper is in Perl, like tools/check-source-nonascii.pl (Perl is already needed for autogen.sh in Git workspaces). It only uses core Perl 5 modules, and parses the "gcc" output format of shellcheck, so no JSON module is needed on older systems. The workflow uses only actions/checkout (no new third-party actions) and compares the PR merge commit with its first parent. A new .shellcheckrc file disables the checks for deliberate portability idioms of NUT scripts: backticks (SC2006), x"$var" comparisons (SC2268) and expr (SC2003). It is distributed like .editorconfig, to help the tarball users contribute too. Describe both in docs/qa-guide.adoc, replacing the note that the tool could be introduced at a later time. The .shellcheckrc backticks setting does for ShellCheck what the previous commit did for "make shellcheck" (issue networkupstools#1058). Checked with the "build: keep full git version out of config.h" series (PR networkupstools#3759): its first version is reported for one new SC2166 warning ([ p -a q ]) and the final one passes. Checked renamed, added, removed and non-shell files, a bad base revision, and the workflow with actionlint and zizmor. In a test PR on a fork, deliberate findings showed as annotations on the changed lines. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: nmbro <nmbro@users.noreply.github.com>

AppVeyorBot · 2026-10-19T08:30:46Z

❌ Build nut 2.8.5.5807-master failed (commit 27de48059c by @jimklimov)

artifacts (if any)

AppVeyorBot · 2026-10-19T08:30:47Z

❌ Build nut 2.8.5.5807-master failed (commit 27de48059c by @jimklimov)

AppVeyorBot · 2026-10-19T13:09:29Z

❌ Build nut 2.8.5.5816-master failed (commit 27de48059c by @jimklimov)

artifacts (if any)

AppVeyorBot · 2026-10-19T13:09:30Z

❌ Build nut 2.8.5.5816-master failed (commit 27de48059c by @jimklimov)

nmbro · 2026-10-20T11:25:27Z

@jimklimov I'm trying to collect all of the red crosses, got to catch them all!

I have spotted some other improvements that can be done to AppVeyor build (again) (will open another PR with those)

but this is still the biggest immediate save, since it's adding unused ccache per PR

jimklimov · 2026-10-20T11:58:09Z

@nmbro : thanks! Note that PR #3786 aims to fix some flaky test failure causes that got into upstream and were discovered with recent faults of unrelated PRs, so assuming it passes okay, many of the recently-regular red crosses will dissipate. I expect to merge it then, roughly in 12 hours...

jimklimov reviewed Oct 14, 2026

View reviewed changes

Comment thread configure.ac

Comment thread configure.ac Outdated

Comment thread UPGRADING.adoc

jimklimov mentioned this pull request Oct 14, 2026

CI: speed up AppVeyor build, fix ccache misses across PRs #3754

Merged

nmbro and others added 3 commits October 15, 2026 10:27

nmbro force-pushed the build-stable-config-h branch from bb5de63 to 4878552 Compare October 15, 2026 09:40

nmbro marked this pull request as ready for review October 15, 2026 10:17

jimklimov added enhancement CI Entries related to continuous integration infrastructure (here CI = tools + scripts + recipes) labels Oct 15, 2026

jimklimov added this to the 2.8.6 milestone Oct 15, 2026

jimklimov added a commit to nmbro/nut that referenced this pull request Oct 15, 2026

tools/gitlog2version.sh, tests/nut-driver-enumerator-test.sh: handle …

0e7554e

…SEMVER_ACINIT corner case of `X.Y.Z.0(.0...)` [networkupstools#3759] Signed-off-by: Jim Klimov <jimklimov+nut@gmail.com>

jimklimov added a commit to nmbro/nut that referenced this pull request Oct 15, 2026

NEWS.adoc: CI_AUTOGEN_STABLE_VERSION never appeared in a release so…

72140e7

… not worth mentioning in these notes [networkupstools#3759] Signed-off-by: Jim Klimov <jimklimov+nut@gmail.com>

jimklimov approved these changes Oct 15, 2026

View reviewed changes

nmbro mentioned this pull request Oct 15, 2026

CI: report script and doc check findings on the changed lines #3773

Merged

7 tasks done

jimklimov added 2 commits October 15, 2026 22:33

tools/gitlog2version.sh, tests/nut-driver-enumerator-test.sh: handle …

b056457

…SEMVER_ACINIT corner case of `X.Y.Z.0(.0...)` [networkupstools#3759] Signed-off-by: Jim Klimov <jimklimov+nut@gmail.com>

NEWS.adoc: CI_AUTOGEN_STABLE_VERSION never appeared in a release so…

6056077

… not worth mentioning in these notes [networkupstools#3759] Signed-off-by: Jim Klimov <jimklimov+nut@gmail.com>

jimklimov force-pushed the build-stable-config-h branch from 1759fa5 to 6056077 Compare October 15, 2026 20:38

Merge branch 'master' into build-stable-config-h

1590292

Merge branch 'master' into build-stable-config-h

6e32b14

github-actions Bot deleted a comment from AppVeyorBot Oct 18, 2026

Merge branch 'master' into build-stable-config-h

b6ff843

nmbro mentioned this pull request Oct 20, 2026

AppVeyor: fix cmd.exe-eaten "||", keep the ccache cache from getting lost, no cache saves from PRs #3790

Merged

jimklimov merged commit 3a703f8 into networkupstools:master Oct 21, 2026
127 of 134 checks passed

nmbro deleted the build-stable-config-h branch October 21, 2026 15:52

Conversation

nmbro commented Oct 14, 2026

Problem

Change

Testing

Open question

Checklist

Uh oh!

github-actions Bot commented Oct 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jimklimov left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

user01010111 commented Oct 15, 2026

Uh oh!

nmbro commented Oct 15, 2026

Uh oh!

jimklimov left a comment

Choose a reason for hiding this comment

Uh oh!

AppVeyorBot commented Oct 19, 2026

Uh oh!

AppVeyorBot commented Oct 19, 2026

Uh oh!

AppVeyorBot commented Oct 19, 2026

Uh oh!

AppVeyorBot commented Oct 19, 2026

Uh oh!

nmbro commented Oct 20, 2026

Uh oh!

jimklimov commented Oct 20, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

github-actions Bot commented Oct 14, 2026 •

edited

Loading