[ci] Cover all supported Pythons and both LLVM majors per PR

[aaronj0]

No description provided.

[github-actions]

Test Results

Configuration	Result
macos-26-intel-llvm21-py3.14-cxx20	= 478 passed, 44 skipped, 83 xfailed, 1 xpassed, 100 warnings in 61.43s (0:01:01) =
macos-26-intel-llvm22-py3.12-cxx20	= 493 passed, 29 skipped, 83 xfailed, 1 xpassed, 100 warnings in 81.53s (0:01:21) =
macos-26-llvm21-py3.12-cxx20	========== 483 passed, 30 skipped, 93 xfailed, 100 warnings in 35.84s ==========
macos-26-llvm21-py3.14-cxx20	========== 483 passed, 30 skipped, 93 xfailed, 100 warnings in 53.66s ==========
macos-26-llvm22-py3.13-cxx20	========== 483 passed, 30 skipped, 93 xfailed, 100 warnings in 45.86s ==========
ubuntu-24.04-llvm21-py3.12-cxx20	============ 564 passed, 25 skipped, 17 xfailed in 66.91s (0:01:06) ============
ubuntu-24.04-llvm21-py3.13-cxx20	============ 564 passed, 25 skipped, 17 xfailed in 64.06s (0:01:04) ============
ubuntu-24.04-llvm21-py3.14-cxx20-vg	============ 564 passed, 25 skipped, 17 xfailed in 70.16s (0:01:10) ============
ubuntu-24.04-llvm22-py3.12-cxx20	============ 563 passed, 25 skipped, 18 xfailed in 72.66s (0:01:12) ============
ubuntu-24.04-llvm22-py3.13-cxx20	================= 563 passed, 25 skipped, 18 xfailed in 49.13s =================
ubuntu-24.04-llvm22-py3.14-cxx20	================= 563 passed, 25 skipped, 18 xfailed in 48.52s =================
ubuntu-24.04-llvm22-py3.14-cxx20-cling	============ 555 passed, 25 skipped, 26 xfailed in 72.13s (0:01:12) ============
ubuntu-24.04-llvm22-py3.14-cxx23	============ 584 passed, 3 skipped, 19 xfailed in 73.52s (0:01:13) =============

[mcbarton]

I see xpassed in the test results comment above. I thought this

cppjit/pyproject.toml

Line 64 in 8be966d

xfail_strict = true

was supposed to make the ci fail if that happens? I don't know if it points towards what the issue is, but the xpasses are not tripping the ci are the macos x86 jobs.

[aaronj0]

Behaves as expected. As documented on that test: flaky on intel macs as it checks for a delta RSS threshold which is not deterministic and can fail intermittently. The xfail_strict = true you've pointed to in pyproject.toml works correctly, test08_list_creation is the only test that overrides it conditionally on intel mac. (see the marker). So if any test on any other platform xpasses, it would in fact, trip the CI