Skip to content

Fix make dist, distcheck and gettext infra - #5201

Open
zyv wants to merge 18 commits into
masterfrom
5193_fix_distcheck
Open

zyv wants to merge 18 commits into
masterfrom
5193_fix_distcheck

Conversation

@zyv

@zyv zyv commented Oct 22, 2026 •

Copy link
Copy Markdown
Member

Proposed changes

This is a batch of release blocker fixes. Unfortunately, this doesn't contain anything pertaining to #3866, but at least it makes mc theoretically releasable. I've updated the PO stuff after all the fixes, so ideally this should be merged soon/next.

Checklist

  • I have referenced the issue(s) resolved by this PR (if any)
  • I have signed-off my contribution with git commit --amend -s
  • Lint and unit tests pass locally with my changes (make indent && make check)
  • I have added tests that prove my fix is effective or that my feature works
  • I have added the necessary documentation (if appropriate)

zyv added 16 commits October 22, 2026 11:01
…wired files

Assisted-By: Claude Opus 5.5
Signed-off-by: Yury V. Zaytsev <yury@shurup.com>
… Fedora

Signed-off-by: Yury V. Zaytsev <yury@shurup.com>
Builds run inside a VM, and workspace files are not available to the
artifact upload step upon failure under the default sync strategy (rsync
on success only). Switch to the nfs strategy establishing a live
bidirectional workspace bridge between the host and the VM.

Made possible by vmactions/solaris-vm#70.

Signed-off-by: Yury V. Zaytsev <yury@shurup.com>
* usqfs.args
* uwim.mock.sh
* ualz.*
* uarj.*

Assisted-By: Claude Opus 5.5
Signed-off-by: Yury V. Zaytsev <yury@shurup.com>
It has been a regular tracked file since a07797f, when it stopped
being generated from `doc.sh.in`.

Assisted-By: Claude Opus 5.5
Signed-off-by: Yury V. Zaytsev <yury@shurup.com>
The unanchored `mc.charsets` pattern was added in 48d3470 for the
file generated in `tests/lib/vfs`, but it also matches the fixture that
is tracked since 1336422. `misc/mc.charsets` is already ignored in
`misc/.gitignore`, so only list the generated test file.

Assisted-By: Claude Opus 5.5
Signed-off-by: Yury V. Zaytsev <yury@shurup.com>
`po/mc.pot` is tracked, because it is pushed to Transifex from the
repository, so listing it in `po/.gitignore` is misleading.

Assisted-By: Claude Opus 5.5
Signed-off-by: Yury V. Zaytsev <yury@shurup.com>
The `config_string` test creates this file in the build directory and
never removes it, so `make distcheck` fails in `distcleancheck`.

Assisted-By: Claude Opus 5.5
Signed-off-by: Yury V. Zaytsev <yury@shurup.com>
`po/POTFILES.in` was generated by running `xgettext` over all sources in
the tree, which has the following problems:

* Strings from the tests (and any other sources lying around in the
  tree) ended up in `mc.pot` and were sent out for translation

* Bootstrapping required GNU `xgettext`, otherwise an empty
  `po/POTFILES.in` was silently created, e.g., on Solaris. The check
  also ignored `XGETTEXT`, so setting it as suggested didn't help.

* Splitting the `#:` comments into file names with `sed` was fragile
  and reportedly glued file names together on some systems (#4619).

Instead, list the sources in `lib` and `src` that look like they call a
gettext keyword using a basic `grep` that works everywhere:

* `_ *(` matches calls of any identifier ending with an underscore,
  i.e., `_ ()`, `N_ ()` and `Q_ ()`, but also `C_ ()` or `NC_ ()`
  should we ever start using them, so the keywords don't have to be
  listed here.

* `gettext *(` matches `ngettext ()` and other `*gettext ()` functions.

False positives such as `__attribute__ (` are harmless: `configure`
only accepts GNU `xgettext` for updating `mc.pot`, and it simply finds
nothing to extract in such files.

Assisted-By: Claude Opus 5.5
Signed-off-by: Yury V. Zaytsev <yury@shurup.com>
The gettext infrastructure files from 0.18.2 are from 2012 and lack the
`PO_DEPENDS_ON_POT` and `DIST_DEPENDS_ON_UPDATE_PO` settings. 0.19.8 is
the oldest version still shipped by the supported systems, e.g., RHEL 8,
Ubuntu 20.04 and Solaris 11.4.

Keep pinning the version instead of switching to
`AM_GNU_GETTEXT_REQUIRE_VERSION`, because the files from gettext >= 0.24
require GNU make >= 4.0 or BSD make. Releases are made on recent Fedora,
so the tarballs would no longer build with the native make on macOS,
Solaris or AIX.

Assisted-By: Claude Opus 5.5
Signed-off-by: Yury V. Zaytsev <yury@shurup.com>
Since `mc.pot` is tracked in git, which doesn't preserve timestamps, PO
files could be rewritten by a regular `make` whenever `mc.pot` happened
to be newer, which also fails if the source tree is read-only.

Moreover, `make dist` (and hence `make distcheck`) always regenerated
`mc.pot` and merged all PO files, rewriting more than 60 tracked files
in the source tree with output depending on the installed gettext.

Disable both, as recommended by gettext when the POT file is tracked
and the PO files are maintained externally, on Transifex in our case.

This is safe even if `make -C po update-po` is forgotten before a
release: `msgmerge` only adds untranslated or fuzzy entries, which
`msgfmt` doesn't compile, and translations are looked up by `msgid`,
so the installed catalogs are exactly the same either way. Only the
`mc.pot` in the tarball can get stale, but translators work on
Transifex, which is fed from the `mc.pot` in git, and `make dist`
never helped with that anyway.

Without `update-po`, the `.gmo` files are no longer built before
`dist2`, which exposes a race in the stock rules: `make -j dist` builds
the same `.gmo` files both as prerequisites of `dist2` and from
`stamp-po`. Hence, disable parallel builds in `po`, and run `distcheck`
in parallel on CI to make sure that it keeps working.

Assisted-By: Claude Opus 5.5
Signed-off-by: Yury V. Zaytsev <yury@shurup.com>
The `.gmo` files are only rebuilt via `stamp-po`, which depends on
`mc.pot` alone, so after the PO files were updated, e.g., by `git pull`,
`make` didn't rebuild the catalogs, and the only way to get them was to
remove `po/stamp-po` manually.

Make `stamp-po` depend on the PO files, so that `make` rebuilds exactly
the catalogs whose PO files have changed.

Assisted-By: Claude Opus 5.5
Signed-off-by: Yury V. Zaytsev <yury@shurup.com>
Overriding `MSGMERGE_UPDATE` in `po/Makevars` replaced the GNU
`msgmerge` found by `configure` with whatever `msgmerge` happens to be
first in `PATH`, for example, when GNU gettext is installed in a
non-standard location.

`MSGMERGE_OPTIONS` is the supported way to pass extra options, and the
stock rules already add it to every `msgmerge` invocation.

Assisted-By: Claude Opus 5.5
Signed-off-by: Yury V. Zaytsev <yury@shurup.com>
The `update-po` rule from `po/Rules-pot-defaults` rewrote the header of
`mc.pot` after it was generated by `xgettext` to work around limitations
of the old Transifex client (325c266).

As a result, the stock `mc.pot-update` rule always considered `mc.pot` to
be outdated and tried to replace it, which fails if the source tree is
read-only, i.e. during `make distcheck`. The same rule also
unconditionally rewrote all PO files to strip location comments, which
are already omitted by `msgmerge --no-location`.

The modern Transifex client accepts the stock header just fine, so use
the stock `update-po` rule instead.

Assisted-By: Claude Opus 5.5
Signed-off-by: Yury V. Zaytsev <yury@shurup.com>
Parse `po/LINGUAS` the same way as gettext does, which also allows
comments at the end of a line.

Assisted-By: Claude Opus 5.5
Signed-off-by: Yury V. Zaytsev <yury@shurup.com>
Signed-off-by: Yury V. Zaytsev <yury@shurup.com>
@zyv
zyv requested a review from mc-worker October 22, 2026 09:38
@github-actions github-actions Bot added needs triage Needs triage by maintainers prio: medium Has the potential to affect progress labels Oct 22, 2026
@github-actions github-actions Bot added this to the Future Releases milestone Oct 22, 2026
@zyv zyv changed the title 5193 fix distcheck Fix make dist, distcheck and gettext infra Oct 22, 2026
@zyv zyv added area: build Build system and (cross-)compilation and removed needs triage Needs triage by maintainers labels Oct 22, 2026
@zyv zyv modified the milestones: Future Releases, 4.9.0 Oct 22, 2026
buildsys: pin gettext infrastructure to 0.19.8

The gettext infrastructure files from 0.18.2 are from 2012 and lack the
`PO_DEPENDS_ON_POT` and `DIST_DEPENDS_ON_UPDATE_PO` settings. 0.19.8 is
the newest version still shipped by the supported systems, e.g., RHEL 8,
Ubuntu 20.04 and Solaris 11.4.

Keep pinning the version instead of switching to
`AM_GNU_GETTEXT_REQUIRE_VERSION`, because the files from gettext >= 0.24
require GNU make >= 4.0 or BSD make. Releases are made on recent Fedora,
so the tarballs would no longer build with the native make on macOS,
Solaris or AIX.

Assisted-By: Claude Opus 5.5
Signed-off-by: Yury V. Zaytsev <yury@shurup.com>

@ossilator ossilator left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

419e195: "since it now runs on Fedora" - one could read that as the bad emulation now running on fedora ... 😂
anyway, i'd move that side sentence to the description. "we now have a proper implementation running on fedora."
alternatively, one could consider squashing this one up, making the replacement atomic. debatable.

67ede21: it's not really necessary to name the files, as git log --stat provides the info on demand. debatable.

1f8884b: kind of hints at an additional fix: actually anchor all paths (incl. files in the root dir).

f4c93b1: i'd put that last, so the actual build instructions aren't torn apart.

ce02722: autogen.sh is expected to be run only from git checkouts, no? then the more elegant solution would be using git ls-tree -r --name-only HEAD -- src lib for the raw file list. that can be filtered down with grep '\.c$' (are there strings in .h files?). content filtering seem unnecessary.

bd31de4: you could try injecting the missing deps from the outside. it's perfectly legit to have multiple rules for the same file, as long as only one has commands. ... and the next commit actually depends on that ...

make install \
make distcheck && \
\
make -C doc/man \

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

you could just use the artifacts from distcheck's build. it doesn't auto-clean after itself.

make indent && \
git diff --exit-code && \
\
make -j$(nproc) && \

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

(apparently) losing the parallel build seems unfortunate.
in principle, every target should work with -j. if it doesn't, some dependencies are missing.

git diff --exit-code && \
\
make distcheck && \
make -j$(nproc) distcheck && \

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ah, here we go (cf. first commit). this is asking for a re-cut, but i'm not sure how yet.

Comment thread doc/INSTALL
- https://www.jedsoft.org/slang/
- https://invisible-island.net/ncurses/ncurses.html
- gettext >= 0.18.2
- gettext >= 0.19.8

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

i'd add "(upstream commits require exactly this version)" or something like that.

@ossilator

Copy link
Copy Markdown
Contributor

so, this series shows quite well why it's actually mostly pointless to clean up before migrating to a new build tool, if that was to happen soon. meson's implementation is most probably different enough to make most of this just unnecessary.

on a totally unrelated note, where do you keep getting claude credits?

The warnings were (re-)enabled for main `./configure` script in #3029,
but we don't ship Samba anymore, and with a large number of historical
options removed, it's safer to alert distro maintainers by failing upon
(now) unknown options by default.

Signed-off-by: Yury V. Zaytsev <yury@shurup.com>
@zyv zyv mentioned this pull request Nov 8, 2026
2 tasks done
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

area: build Build system and (cross-)compilation prio: medium Has the potential to affect progress

3 participants