test cases/linuxlike/14 static dynamic linkage: Add matrix by static #13842

dememax · 2024-10-30T13:44:52Z

Not only Solaris doesn't ship static libraries, it's common practice for Linux distributions in general (security updates, size/space, etc.).
See, for example 'test cases/frameworks/1 boost'.

Enhancements:

Added flexibility: Introduced the SKIP_STATIC_ZLIB environment variable to optionally skip static linkage (similar to SKIP_STATIC_BOOST).
Expanded testing: Included testing for dynamic linkage, which was previously limited to static in verify_static.py (renamed to verify_zlib_linkage.py).
Improved test coverage: Ensured that each case is tested for its specific linkage type and the opposite case fails.
Addressed Cygwin issue: Ensured that the corresponding symbol is present on Cygwin.

Renames, comments, etc.

dememax · 2024-10-30T14:05:57Z

https://github.com/mesonbuild/meson/actions/runs/11594506125/job/32280901101#step:5:587

Total failed tests:  1

All failures:
  -> common: 273 both libraries    (default_library=both default_both_libraries=static use_dep=True)

Is it my fault?

dememax · 2024-10-30T19:20:39Z

During "cygwin / gccx64ninja (pull_request) Failing", with the help of debug print, I've found:

For dynamic linkage (static=false), nm output in "linuxlike / static dynamic linkage" filtered by "zlibVersion" is:

0000000100408150 I __imp_zlibVersion
00000001004010d0 T zlibVersion

For static, it doesn't contain the symbol with "_imp".

eli-schwartz · 2024-10-31T14:43:35Z

Small nit: commit messages should use imperative mood: https://cbea.ms/git-commit/#imperative

dememax · 2024-11-02T11:49:46Z

@eli-schwartz Done! Thanks!

eli-schwartz · 2024-11-05T04:56:10Z

test cases/linuxlike/14 static dynamic linkage/meson.build

-else
-  has_static = true
-endif
+project('zlib linkage', 'c')


The project description doesn't really need to change here (it is still testing static dynamic linkage, not zlib specifically, anyway :))

Shall be fixed! Thanks.

eli-schwartz · 2024-11-05T04:57:53Z

test cases/linuxlike/14 static dynamic linkage/meson.build

-# Solaris does not ship static libraries
-if host_machine.system() == 'sunos'


We should probably still explicitly skip on Solaris. Your new approach would cause test failure if SKIP_STATIC_ZLIB is not exported, even though we know that if s and host_machine.system() == 'sunos' there will not be a static library.

You are perfectly right that this changes the behavior, and it's better to keep the previous logic if we don't want to break something. Nice catch!

I will do what you are asking, but let's go a little farther...

When my new approach cause test failure

As far as I know, there are no SunOS tests in the current CI. And, we both know, the green status is showing this.

So there should be some test cases outside the project, the dark matter, where people will have this issue.

When my new approach fixes problem where it could possibly cause test failure

AFAIK, SunOS could be configured to have the static zlib version in addition to the default dynamic one. In this case, keeping the old condition is a bug.

Let's ask the author of those lines @alanc

With my approach the user of this test case will be able to change his environment to adopt the usage on SunOS for his/her setup.

Symmetry / Special case

What is so special about SunOS to include it into this build script?

What about the default configuration on MS Windows (no Cygwin) or MacOS?

What about IBM i (AS/400) and BSD families?

Gentoo setup is configured for Boost with the help of SKIP_STATIC_BOOST in the GHA yaml file. Why could not SunOS be configured in the same way?

Bottom line

As I said, I will fix this to keep this special case for SunOS, but IMHO it must go.

While there is no SunOS (Solaris or illumos) instance in the CI, we run the test suite on Solaris when building the meson packages for Solaris. It is true that it is not impossible for a user to build and install a static version of libz on Solaris - but since we provide a dynamic library for libz as part of the OS packages, I don't know of anyone who bothers to build their own static version. I can't speak to other OS'es or why they didn't hit the problem - perhaps their default configuration doesn't have dynamic libz either instead of just dynamic and not static.

I'm not opposed to moving this from a line of code to a configuration file - what's the "GHA yaml file"?

$ git --no-pager grep -C3 STATIC_BOOST .github/workflows/cygwin.yml- env: .github/workflows/cygwin.yml- # Cygwin's static boost installation is broken (some static library .github/workflows/cygwin.yml- # variants such as boost_thread are not present) .github/workflows/cygwin.yml: SKIP_STATIC_BOOST: 1 .github/workflows/cygwin.yml- shell: C:\cygwin\bin\bash.exe --noprofile --norc -o igncr -eo pipefail '{0}' .github/workflows/cygwin.yml- .github/workflows/cygwin.yml- - uses: actions/upload-artifact@v4 -- ci/ciimage/fedora/image.json- "base_image": "fedora:latest", ci/ciimage/fedora/image.json- "env": { ci/ciimage/fedora/image.json- "CI": "1", ci/ciimage/fedora/image.json: "SKIP_STATIC_BOOST": "1", ci/ciimage/fedora/image.json- "MESON_CI_JOBNAME": "linux-fedora-gcc" ci/ciimage/fedora/image.json- } ci/ciimage/fedora/image.json-} -- ci/ciimage/gentoo/image.json- "env": { ci/ciimage/gentoo/image.json- "CI": "1", ci/ciimage/gentoo/image.json- "MESON_CI_JOBNAME": "linux-gentoo-gcc", ci/ciimage/gentoo/image.json: "SKIP_STATIC_BOOST": "1" ci/ciimage/gentoo/image.json- } ci/ciimage/gentoo/image.json-} -- ci/ciimage/opensuse/image.json- "base_image": "opensuse/tumbleweed:latest", ci/ciimage/opensuse/image.json- "env": { ci/ciimage/opensuse/image.json- "CI": "1", ci/ciimage/opensuse/image.json: "SKIP_STATIC_BOOST": "1", ci/ciimage/opensuse/image.json- "SINGLE_DUB_COMPILER": "1", ci/ciimage/opensuse/image.json- "MESON_CI_JOBNAME": "linux-opensuse-gcc" ci/ciimage/opensuse/image.json- } -- test cases/frameworks/1 boost/test.json- "matrix": { test cases/frameworks/1 boost/test.json- "options": { test cases/frameworks/1 boost/test.json- "static": [ test cases/frameworks/1 boost/test.json: { "val": "true", "skip_on_env": [ "SKIP_STATIC_BOOST" ] }, test cases/frameworks/1 boost/test.json- { "val": "false" } test cases/frameworks/1 boost/test.json- ], test cases/frameworks/1 boost/test.json- "b_vscrt": [

For all our docker-based CI testing we generate the docker images themselves with /ci/env_vars.sh as a kind of "/etc/profile for CI".

For Cygwin, which isn't docker-based, we use the GitHub Actions (GHA) yaml file .github/workflows/cygwin.yml. We don't maintain any such thing for Solaris' packaging.

Thank you, @alanc !
@eli-schwartz, should we throw away the SunOS special condition, or merge the PR in the current state?

Excuse me, @eli-schwartz, what do you think?

Not only Solaris doesn't ship static libraries, it's common practice for Linux distributions in general (security updates, size/space, etc.). See, for example 'test cases/frameworks/1 boost'. Enhancements: * Added flexibility: Introduced the SKIP_STATIC_ZLIB environment variable to optionally skip static linkage (similar to SKIP_STATIC_BOOST). * Expanded testing: Included testing for dynamic linkage, which was previously limited to static in verify_static.py (renamed to verify_zlib_linkage.py). * Improved test coverage: Ensured that each case is tested for its specific linkage type and the opposite case fails. * Addressed Cygwin issue: Ensured that the corresponding symbol is present on Cygwin. Renames, comments, etc.

dememax · 2024-12-03T09:39:26Z

Only one fatal error not caused by my changes:

Ubuntu Bionic
failed Dec 3, 2024 in 46s

Run actions/checkout@v3
/usr/bin/docker exec  7f49c64d95a49a1cc1327eb78b5f56bc3e01d8e7f8e38f1f2dcc91e71e0b1614 sh -c "cat /etc/*release | grep ^ID"
/__e/node20/bin/node: /lib/x86_64-linux-gnu/libc.so.6: version `GLIBC_2.28' not found (required by /__e/node20/bin/node)

https://github.com/mesonbuild/meson/actions/runs/12135619615/job/33835043547?pr=13842

dememax requested a review from jpakkane as a code owner October 30, 2024 13:44

dememax force-pushed the optional-zlib-linkage-test branch from 78a74fd to b7f1d3c Compare October 31, 2024 11:31

dememax force-pushed the optional-zlib-linkage-test branch from b7f1d3c to 629e79f Compare November 2, 2024 11:47

dememax changed the title ~~test cases/linuxlike/14 static dynamic linkage: Adding matrix by static~~ test cases/linuxlike/14 static dynamic linkage: Add matrix by static Nov 2, 2024

dememax force-pushed the optional-zlib-linkage-test branch from 629e79f to a509016 Compare November 4, 2024 17:12

eli-schwartz reviewed Nov 5, 2024

View reviewed changes

dememax force-pushed the optional-zlib-linkage-test branch from a509016 to edabf58 Compare November 5, 2024 09:58

dememax requested a review from eli-schwartz November 5, 2024 10:20

dememax force-pushed the optional-zlib-linkage-test branch from edabf58 to 9cec3d6 Compare November 24, 2024 19:25

dememax force-pushed the optional-zlib-linkage-test branch from 9cec3d6 to c9157c9 Compare December 3, 2024 08:22

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

test cases/linuxlike/14 static dynamic linkage: Add matrix by static #13842

test cases/linuxlike/14 static dynamic linkage: Add matrix by static #13842

dememax commented Oct 30, 2024 •

edited

Loading

dememax commented Oct 30, 2024 •

edited

Loading

dememax commented Oct 30, 2024

eli-schwartz commented Oct 31, 2024

dememax commented Nov 2, 2024

eli-schwartz Nov 5, 2024

dememax Nov 5, 2024

eli-schwartz Nov 5, 2024

dememax Nov 5, 2024

alanc Nov 5, 2024

eli-schwartz Nov 5, 2024

dememax Nov 6, 2024 •

edited

Loading

dememax Nov 24, 2024

dememax commented Dec 3, 2024

		# Solaris does not ship static libraries
		if host_machine.system() == 'sunos'

test cases/linuxlike/14 static dynamic linkage: Add matrix by static #13842

Are you sure you want to change the base?

test cases/linuxlike/14 static dynamic linkage: Add matrix by static #13842

Conversation

dememax commented Oct 30, 2024 • edited Loading

dememax commented Oct 30, 2024 • edited Loading

dememax commented Oct 30, 2024

eli-schwartz commented Oct 31, 2024

dememax commented Nov 2, 2024

eli-schwartz Nov 5, 2024

Choose a reason for hiding this comment

dememax Nov 5, 2024

Choose a reason for hiding this comment

eli-schwartz Nov 5, 2024

Choose a reason for hiding this comment

dememax Nov 5, 2024

Choose a reason for hiding this comment

When my new approach cause test failure

When my new approach fixes problem where it could possibly cause test failure

Symmetry / Special case

Bottom line

alanc Nov 5, 2024

Choose a reason for hiding this comment

eli-schwartz Nov 5, 2024

Choose a reason for hiding this comment

dememax Nov 6, 2024 • edited Loading

Choose a reason for hiding this comment

dememax Nov 24, 2024

Choose a reason for hiding this comment

dememax commented Dec 3, 2024

dememax commented Oct 30, 2024 •

edited

Loading

dememax commented Oct 30, 2024 •

edited

Loading

dememax Nov 6, 2024 •

edited

Loading