Drop the relocatable build flag; test sysdir resolution

[charlesstaats]

resolveSysdir(), the per-platform executablePath(), and the preference for a base/ beside the executable are all already on master. This branch finishes that work: it deletes the build-time opt-in and the two extra locations that opt-in gated, adds a FreeBSD branch to executablePath(), and adds the tests.

Remove --enable-relocatable and IS_RELOCATABLE, and with them two of the three candidate locations (<exedir>/../share/asymptote, and base files flat beside the binary). One candidate remains: <exedir>/base, accepted only if it contains plain.asy — true only in a build tree or a distribution that deliberately ships base/ beside the binary, which is why it needs no opt-in. Otherwise the compiled-in ASYMPTOTE_SYSDIR is returned unchanged. An empty compiled-in sysdir no longer short-circuits resolution; it is just the fallback that initDir() hands to kpsewhich, so a TeXLive-configured binary run from its build tree finds its own base/. The macOS bundle now relocates by layout (make install-asy bindir=$STAGE/Asymptote asydir=$STAGE/Asymptote/base), and the devcontainer preset template drops its ENABLE_RELOCATABLE cache entries, which matched no CMake option and did nothing.

Fix ASYMPTOTE_SYSDIR on the CTAN binary. Definitions inherited from asycore are emitted after a target's own, so the global -DASYMPTOTE_SYSDIR beat the per-target override and asy-ctan was silently built with the normal sysdir. It is no longer set globally; add_settings_obj takes it as a required parameter.

Tests. tests/test_executable_path.py is a straight-line smoke test of executablePath() on the host it runs on — the one part written once per OS, so the part likeliest to be wrong where nobody has tried it (the new, untested FreeBSD kern.proc.pathname branch exists because /proc is not mounted there). tests/test_relocatable.py is the matrix: the 18 layout states of the candidate plus the two former candidates as negative controls, exhaustively, with axes for launch route, -sysdir/-dir/ASYMPTOTE_SYSDIR, the compiled-in value including the empty CTAN one, and the Windows registry. Both run under ctest and make check. New asy-check-test-deps target, since ctest builds nothing and asy-with-basefiles does not pull in asy-ctan; tests/Makefile also stops trying to run __pycache__/*.asy as a suite.

Python checking. mypy.ini (allowlist-driven, floor 3.7), run once per platform in CI since mypy prunes sys.platform branches; pyrightconfig.json for editors; mypy~=1.8.0, the last release able to target 3.7; PY3_MINIMUM_VERSION enforced by the CMake interpreter search. A pylint plugin replaces too-many-lines with a statement-based limit, so documentation does not count against a module, and print_non_gui_py_files_for_linting.py asks git for candidates instead of walking the tree. Annotating generate_enums.py turned up a real bug: -xopt spaces=N arrives as a string and was used as a repeat count.

Makefile.in cleanups salvaged from the abandoned libatomic_ops work. License installation factored into one define shared by copy-licenses and install-licenses (which no longer installs by way of doc/licenses, and creates $(docdir) itself, since --with-licensedir can point elsewhere); GL shaders removed by uninstall-asy.

Build and packaging fixes. A stamp rebuilds gc when the macOS architecture flags or gc's configure options change; a touch after macOS Vulkan bundling stops asy being relinked on every later make. The bundler strips the absolute LC_RPATHs the linker recorded — they point into the builder's home and ship in the .dmg — refusing to if any @rpath reference survived the rewrite. sync-vcpkg-baseline.sh records the baseline it bootstrapped for, so an interrupted bootstrap is retried.

Plus doc corrections in INSTALL-VCPKG.md and .devcontainer/README.md: the real ctest test names, --tests-list for running a subset, and why AC_PREREQ stays at 2.71.

[Copilot]

🟡 Changes recommended

The GC stamp reruns its recipe without invalidating existing objects, so changed architecture flags may not rebuild GC correctly.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Review details

Suppressed comments (1)

Makefile.in:345

• Changing the stamp does rerun this outer recipe, but the nested make can still consider the existing gc object files and library up to date: make does not track changes to compiler flags, and rerunning configure does not itself invalidate those objects. A switch such as single-arch to universal can therefore keep the old objects, defeating the stamp's stated purpose. Clean gc after configuring (or make its objects depend on the stamp) before rebuilding.

$(GCLIB): $(GCSTAMP)
	-cd $(GC) && ln -sf ../libatomic_ops libatomic_ops
	cd $(GC) && \
		./configure CC="$(CC)" CXX="$(CXX)" CFLAGS="$(MACOS_CFLAGS) $(CFLAGS)" CXXFLAGS="$(MACOS_CFLAGS) $(CXXFLAGS)" $(GCOPTIONS)
	$(MAKE) -C $(GC) all CFLAGS_EXTRA="$(MACOS_CFLAGS)"

• Files reviewed: 31/32 changed files
• Comments generated: 1
• Review effort level: Balanced