[zfergus]

Description

Fixes the CI caching and Debug test setup across the Build, Python, Coverage, CUDA, and Docs workflows, plus one pathological unit test. On #300, CI took ~40 min wall-clock, and Linux Debug failed on a timeout.

• ccache keys — every workflow except the ODR check used a fixed key (Linux-Debug-cache, coverage-cache, …). Cache entries are immutable, so each was saved once on 2026-10-12 and never updated. Every build job on Fix Eigen ABI pin when a parent project shadows the Eigen recipe #300 got 0/465 hits, while the ODR job, which keys by run_id, got 459/465 on the same commit.
• Windows — Windows Debug's /Zi writes to a shared PDB that ccache can't cache (465/465 uncacheable). Windows Python used the Visual Studio generator, which ignores CMAKE_CXX_COMPILER_LAUNCHER and compiles each target's files serially (2,340 s build vs 616 s on Linux).
• Debug tests — at -O0, the Debug test step took 1,079–1,713 s vs 131–166 s in Release. CCD stops at the thickness alone took 1,319 s on macOS and timed out at 1,500 s on Linux, also failing main in run 42642942563.

Impact:

• PR builds restore the latest main ccache instead of compiling from scratch.
• The Debug test step is no longer bound by one 20+ minute test, and the flaky timeout is gone.
• No library or API changes.

CI changes

• Restore with actions/cache/restore using key <prefix>-ccache-${{ github.run_id }} and restore-keys: <prefix>-ccache-, the same pattern as odr-check.yml.
• Save with actions/cache/save only on pushes to main (always for docs.yml, which only runs on push), so PR caches don't evict main's under the 10 GB limit.
• Windows Debug configures with /Od /Ob0 /RTC1, matching what cuda.yml already does.
• Windows Python sets up MSVC and builds with CMAKE_GENERATOR=Ninja.
• Linux/macOS Debug builds at -O1 instead of -O0; NDEBUG stays undefined, so asserts are still checked.

Test change

• CCD stops at the thickness uses a TightInclusionCCD with a 1e-3 tolerance instead of the default 1e-6.
• Tight Inclusion slows down sharply as the minimum separation grows, and this test uses 0.2. Its assertions only check the step to within 0.1.
• Removed the Windows-Debug SKIP, since the test is now cheap everywhere.

	before	after
CCD stops at the thickness, Release (5 broad phases)	15.6 s	0.83 s
Full suite, Debug -O1, 456 tests	—	12.8 s

• The old *-cache entries no longer match any key and will expire after 7 days unused.
• This PR's own builds will still be cold: the new keys only get populated by the first push to main after merge.

Type of change

• Bug fix (non-breaking change which fixes an issue)
• Enhancement (non-breaking change which improves existing functionality)

How Has This Been Tested?

• ipc_toolkit_tests "[thickness]" in Release: all 9 test cases pass (6,347 assertions).
• Full ctest -j 8 on a Debug build configured with -DCMAKE_CXX_FLAGS_DEBUG=-O1: 456/456 pass, 12.8 s.
• Full ctest -j 8 in Release: 457/457 pass, 43.4 s.
• All five edited workflows parse as YAML.
• Workflow behavior (cache restore/save, Windows Ninja Python build) is verified only by this PR's CI and the first post-merge run on main.
• Before numbers come from the job logs of Fix Eigen ABI pin when a parent project shadows the Eigen recipe #300.

Test Configuration:

• OS and Version: macOS 26 (Apple Silicon)
• Compiler and Version: Apple Clang 21

Checklist

• I have followed the project style guide
• My code follows the clang-format style guidelines of this project
• I have performed a self-review of my code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• Any dependent changes have been merged and published in downstream modules

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 96.86%. Comparing base (f388443) to head (2827265).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #301   +/-   ##
=======================================
  Coverage   96.86%   96.86%           
=======================================
  Files         229      229           
  Lines       23047    23047           
  Branches     1619     1618    -1     
=======================================
  Hits        22324    22324           
  Misses        723      723           

Flag	Coverage Δ
unittests	96.86% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.