The arena_inlines_a.h / arena_inlines_b.h split, and the parallel
jemalloc_internal_inlines_b.h file that defined arena_choose() between
them, all existed to manage one ordering constraint: arena_choose()
had to be defined before arena_choose_maybe_huge() (which calls it),
but had to be defined after the tsd/tcache inlines it depends on.
Three files, one staged include order, no real semantic boundary.
After the malloc_dispatch refactor moved the heaviest tcache-pulling
inlines (the malloc/dalloc routing) out of arena_inlines_b.h, the
arena-side inlines that remain all belong together. Merge them into
a single arena_inlines.h that:
- explicitly includes jemalloc_internal_inlines_a.h (for tsd
accessors) and tcache.h (for tcache_arena_associate /
reassociate externs) -- both were previously pulled
transitively;
- orders functions so each caller appears after its callee
(cheap accessors -> arena_choose family -> the rest), so no
forward references are needed;
- drops the load-bearing-split comment, which is no longer true.
All consumers that included any of the three old headers now include
arena_inlines.h. background_thread_inlines.h now pulls a heavier set
of transitive includes (prof.h, large.h, mutex.h, ...) than when it
only needed cheap accessors; this is acceptable because every TU that
includes background_thread_inlines.h already pulls those headers via
other paths.
Converted: hpa_utils, ecache, extent_mmap, util, safety_check,
prof_stats, peak_event, inspect, log.
(src/div.c was already minimal; skipped.)
One latent hermeticity bug surfaced: peak_event.h declared
`extern te_base_cb_t peak_te_handler;` but didn't include
thread_event_registry.h where te_base_cb_t is typedef'd. Added the
include to peak_event.h. peak_event.c also needs thread_event.h
directly for TE_MIN_START_WAIT.
Step 6 (Option B) of the cyclical-dep cleanup, batch 2 of N.
- `-Wmissing-prototypes` and `-Wmissing-variable-declarations` are
helpful for finding dead code and/or things that should be `static`
but aren't marked as such.
- `-Wunused-macros` is of similar utility, but for identifying dead macros.
- `-Wunreachable-code` and `-Wunreachable-code-aggressive` do exactly
what they say: flag unreachable code.
This saves us a cache miss when lookup up the arena bin offset in a remote
arena during tcache flush. All arenas share the base offset, and so we don't
need to look it up repeatedly for each arena. Secondarily, it shaves 288 bytes
off the arena on, e.g., x86-64.
