e6c2fd44fc
The pthread slot approach is somewhat buggy (pretty visible with the stress unit test which segfault more or less randomly, but the stats never show up). Using the default approach instead, the test passes eventough it s relatively slow (e.g 1.5 sec on FreeBSD vs 4.5 on DragonFly with same machine).
Testing allocators is difficult as bugs may only surface after particular
allocation patterns. The main approach to testing mimalloc is therefore
to have extensive internal invariant checking (see page_is_valid in page.c
for example), which is enabled in debug mode with -DMI_DEBUG_FULL=ON.
The main testing strategy is then to run mimalloc-bench using full
invariant checking to catch any potential problems over a wide range of intensive
allocation benchmarks and programs.
However, this does not test well for the entire API surface and this is tested
with test-api.c when using make test (from out/debug etc). (This is
not complete yet, please add to it.)
The main.c and main-override.c are there to test if building and overriding
from a local install works and therefore these build a separate test/CMakeLists.txt.