alexcrichton commented on issue #7576:
My first reaction to this is that this isn't where I'd like to see an env var being tested. I think currently for example
WASMTIME_TEST_NO_HOG_MEMORY
is only checked in test binaries, never in Wasmtime runtime code. Ideally we'd do the same here and only change test-related code to check this env var. That being said I can see how that's not easy to do since there isn't one location where all theEngine
s are created.Perhaps just the
wast
tests and maybe a few others could get mpk force-enabled with this env var though? Get some broad coverage but not necessarily 100%. Another option would be RUSTFLAGS with--cfg force_mpk
or something like that
abrown commented on issue #7576:
My first reaction to this is that this isn't where I'd like to see an env var being tested.
Makes sense. The commit I just pushed, 3d2102a, limits the use of this variable to the
wast
tests and the 30 uses ofsmall_pool_config
. Forcing MPK in tests that usesmall_pool_config
is not _too_ useful: they only really have one memory slot so we don't check the multiple slices. But it may exercise a few more of the MPK code paths. In any case, thewast
tests _are_ useful: any time in the past I've tested an incorrect implementation of MPK, thewast
tests have had some failures.
Last updated: Nov 22 2024 at 16:03 UTC