[INFRA] CI and Compilers #3270

eseiler · 2024-07-17T10:32:10Z

Uses docker images from seqan/actions
Removes CMake 3.5 test (prone to errors with newer operating systems, and hard to get running)
gcc-14 support
clang-18 support
Change default of SEQAN3_VERBOSE_TESTS to OFF. ON is a bad default because we basically always set it to OFF explicitly
Simplify coverage generation. I don't think we ever use it to build local HTML reports. And not doing everything in CMake makes it much easier to maintain (and read).

I did not yet update the cron jobs (follow-up).

vercel · 2024-07-17T10:32:21Z

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Updated (UTC)
seqan3	✅ Ready (Inspect)	Visit Preview	Jul 18, 2024 3:52pm

codecov · 2024-07-17T12:38:58Z

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.12%. Comparing base (69284e8) to head (a151471).

❗ Current head a151471 differs from pull request most recent head 82e9727

Please upload reports for the commit 82e9727 to get more accurate results.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3270      +/-   ##
==========================================
- Coverage   98.18%   98.12%   -0.07%     
==========================================
  Files         270      270              
  Lines       11885    11921      +36     
  Branches        0      105     +105     
==========================================
+ Hits        11669    11697      +28     
- Misses        216      224       +8

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

rrahn

Looks all good to me.
I have only some open questions. Otherwise this can be merged after squahsing the debug commit.

rrahn · 2024-07-18T13:44:22Z

.github/workflows/ci_documentation.yml

-      - name: Install CMake
-        uses: seqan/actions/setup-cmake@main
-        with:
-          cmake: 3.16.9



Not running inside a docker container?

Not yet, because I didnt do doxygen-containers, and we don't need compilers for the documentation build.

I just use the default cmake version.

rrahn · 2024-07-18T13:45:35Z

.github/workflows/ci_linux.yml

    if: github.repository_owner == 'seqan' || github.event_name == 'workflow_dispatch' || github.event.label.name == 'lint'
    strategy:
-      fail-fast: true
+      fail-fast: false


Why is this change necessary?

It's not necessary, but I come to find it rather annoying that all jobs are cancelled when a single job fails :)

rrahn · 2024-07-18T14:01:06Z

test/external_project/install-sharg.cmake

@@ -4,6 +4,8 @@

 cmake_minimum_required (VERSION 3.14)

+cmake_policy (SET CMP0135 NEW)


Can you please add a comment what this policy is for.
Maybe that explains why it is set already. Otherwise, please enlighten me :)

I'll add a link to https://cmake.org/cmake/help/latest/policy/CMP0135.html

There was a segfault in `enforce_random_iterator_test::adaptor`. `auto v3 = test_range | seqan3::views::enforce_random_access | std::views::drop(1);` llvm-18 caches some iterator. It will then call `begin() + pos`, where position is the cached distance, in this case 1 from the drop view. This then calls `operator+` from, at some point, the inherited_iterator_base. The actual iterator is `sentinel_pseudo_random_access_range::test_iterator`. With `derived_t{as_base() - skip}`, the constructor with only the base iterator is called `inherited_iterator_base(base_t it)`. However, the `test_iterator` also has a member `last`, which in this case would be defaulted to a `nullptr` (the std::vector<int> iterator decays to a int*). This means that for `std::ranges::copy` `first != last` will always be false. The fix is to properly implement `inherited_iterator_base`'s operators that return a copy to first copy the derived iterator instead of constructing a new one.