fix: invalidate cache when install directory is missing

[nikhilramakrishna3005]

Summary

Fixes a stale package-cache state where PackageCache.is_built() could return True even when the cached package's install/ directory no longer existed.

A valid .built marker alone does not guarantee that the cached package is usable, because callers rely on the install/ directory for the package's installed headers and libraries.

Type of Change

• feat — New feature
• fix — Bug fix
• docs — Documentation only
• style — Formatting, no code change
• refactor — Code restructuring without behavior change
• test — Add or fix tests
• build — Build system or dependency changes
• ci — CI/CD pipeline changes
• perf — Performance improvement

Changes

• Updated PackageCache.is_built() to require the package install/ path to exist as a directory before treating the cache as valid.
• Added a regression test that verifies a valid .built marker is not sufficient when the corresponding install/ directory has been removed.

Testing

• Unit tests pass (ctest --test-dir build --output-on-failure)
• Integration tests pass
• Manual testing performed
• New tests added for new functionality

Focused regression test:

python3 -m pytest tests/ebuild/test_package_cache.py -v --tb=short

Result: 1 passed

Lint:

python3 -m flake8 ebuild/packages/cache.py tests/ebuild/test_package_cache.py

Result: passed

The regression test was also verified against the original implementation, where it failed because is_built() incorrectly returned True after the install/ directory was removed.

The broader test suite currently has unrelated failures on upstream master, including existing dispatcher and Ninja backend issues. The new package-cache regression test passes independently.

Pre-Submission Checklist

• Code compiles without warnings (-Wall -Wextra -Werror for C)
• All existing tests pass
• New tests added for new functionality
• Documentation updated if API changed
• Commit messages follow <type>(<scope>): <description> convention
• Branch is rebased on latest master

Related Issues

None.

Screenshots / Logs

Not applicable.

Additional Notes

This change uses is_dir() rather than only exists() because a non-directory path named install is not a usable package installation prefix.

The existing .built metadata validation for package name, version, and checksum is unchanged. No public APIs or cache metadata formats are modified.

[srpatcha]

Approving. Small, correct, and the test genuinely pins the behaviour.

is_built() is read by callers as "the cached install prefix is usable", but it
was only checking the .built marker. A marker outliving its install directory —
a partially cleaned cache, an interrupted rm, a pruned build tree, anything
that removes install/ without removing the marker beside it — made
is_built() return true for a prefix that is not there. The build then skips
the rebuild and fails later, somewhere that gives no hint the cache was the
cause.

Checking both is the right shape: the marker says "we finished", the directory
says "and the result still exists", and the caller needs both to be true.

Verified

Applied onto #66 (necessary — master itself does not parse, #87), clean apply:

with fix       5 passed, 202 deselected
without fix    FAILED test_is_built_is_false_when_install_dir_is_missing
               1 failed, 4 passed

Removing only your two-line guard turns the test red, so it fails on the unfixed
code and passes on the fixed one. That is what separates a regression test from a
demonstration, and a surprising number of "regression tests" do not manage it —
one earlier in this repo asserted on values that were correct both before and
after the bug it claimed to cover.

Whole suite on the same base: 206 passed, up from 205, no failures.

The docstring naming the invariant — "a valid .built marker is not enough if
install/ is gone" — is the part that will stop someone deleting this check in
two years as a redundant stat call.

Needs #66 merged first for CI here to run at all.