Valentin Churavy
Valentin Churavy
Copying what I just posted in Slack here: > One thought with regards to [sources] I just came across a package that has dependencies that exist in a different registry....
I am bad at `sed`.
Perhaps, but @lcw GridArray is a wrapper type... So it felt a bit silly to add a second wrapper type to present a vector to Krylov.jl
When it doesn't crash I get: ``` Expression: ((autodiff(Reverse, f9, Active, Active(2.0)))[1])[1] == 8 Evaluated: 0.0 == 8 ```
``` Thread 1 hit Breakpoint 2.1, gc_mark_outrefs (ptls=0x55f8f928d3c0, mq=0x55f8f928da00, _new_obj=0x7fac6a6b7e70 ) at /home/vchuravy/src/julia-1.12/src/gc-stock.c:2239 2239 uintptr_t vtag = o->header & ~(uintptr_t)0xf; (rr) p new_obj $2 = (jl_value_t *) 0x7fac6a6b7e70 (rr) p...
Interestingly running under ASAN ``` vchuravy@loki ~/s/Enzyme (main)> ~/src/julia-1.12-asan/asan/julia -g2 --project=. 2825.jl Error During Test at /home/vchuravy/src/Enzyme/2825.jl:12 Test threw exception Expression: ((autodiff(Reverse, f9, Active, Active(2.0)))[1])[1] == 8 EnzymeInternalError: Enzyme compilation...
ASAN changed the way object are laid out in memory and so the corruption memory write goes somewhere else and doesn't cause the GC crash
Would it make sense to open a PR upstream and ask for feedback?
Sure, but before we commit to the API. Maybe we get lucky with a quick review ;)
Bump since this is now on LLVM main?