chore(release): 2.7.6 -- separate debug and release build configurations

[LESdylan]

Cuts v2.7.6 and separates the build you develop against from the build you ship.

Build configurations

make now takes MODE=debug|release|relwithdebinfo:

MODE	flags	for
debug (default)	-O0 -g3 -ggdb, ASan+LSan, libc malloc	developing
release	-O3 -DNDEBUG, LTO, --gc-sections, no -g, no sanitizer	shipping
relwithdebinfo	-O2 -g -DNDEBUG, no sanitizer, no LTO	bugs that only appear optimized

Measured on this machine:

mode	size	debug data	ASan syms
release	462 KB	807 B	0
relwithdebinfo	3.0 MB	2.6 MB	0
debug	5.6 MB	2.2 MB	154

Nothing is stripped after the fact — release carries no debug information because release never compiles -g in. The 807 bytes that remain are a single compilation unit, libgcc's crtfastmath.c, pulled in by the pre-existing -ffast-math and shipped by the distro already built with -g. None of our 493 translation units contribute any debug info to release. No sanitizer, assertion, symbol or dependency was removed from the debug build to get there.

OPT=1 — which the release workflow, the platform matrix, both install targets, the Docker build and make bench all pass — still means exactly MODE=release, byte for byte, and a test pins that.

Two real defects found while writing the test

OPT=1 disagreed with itself. SAFE defaulted off ifdef OPT, so the two spellings of one configuration picked different allocators and produced different binaries. SAFE now keys off MODE.

Build modes shared an object tree. OBJ_DIR also keyed on ifdef OPT, covering the benchmark build alone and leaving MODE=release and MODE=relwithdebinfo parked in the debug tree. make rebuilds on a changed prerequisite, never on a changed flag, so a plain make MODE=release after a debug build fed the linker ASan-instrumented objects under a link line with no -fsanitize:

func_retire.o: undefined reference to `__asan_report_load4'

make re hid it. Objects now live in build/obj-<mode>-<allocator>.

Tests

tests/build_modes_test.py — 20 checks reading make flags, which resolves a configuration without compiling, so it costs nothing and cannot be fooled by a stale object tree. 3 of its checks fail against the pre-fix OBJ_DIR, verified red-then-green.

tests/link_closure_test.py now globs build/obj* and picks the newest tree rather than the first name it recognised — which is what its docstring always claimed, and what a fixed list could not keep doing once modes were added.

Both are picked up automatically by tests/pty_suite.sh (it globs tests/*.py), so no CI wiring was needed.

Also in this release

67508ad (already on develop) — make user-install now leaves hellish on PATH rather than only on disk. Release notes updated to cover it, since it ships under this tag.

Verified locally

• make norm — clean
• golden suite — 3790/3790
• pty suite — 30 ok / 0 failed / 4 skipped
• build_modes_test.py 20/20, user_install_path_test.py 22/22