Lukasz Dorau
Lukasz Dorau
> [@ldorau](https://github.com/ldorau) thanks for following up. The error path is triggered by the extent split failure, which I suspect is not the expected behavior by your custom extent hooks --...
Running with `MALLOC_CONF="retain:false"` nor adding `const char *malloc_conf = "retain:false";` to the code does not help, the output is the same like it had been before: ``` $ MALLOC_CONF="retain:false" ./test/test_jemalloc_pool...
> Did you build with a custom prefix? If so you need to prefix the `malloc_conf` as well. > Yes, we build jemalloc with a custom prefix, but we cannot...
@AlexeySachkov `UMF_DISABLE_HWLOC` is going to be removed soon.
@vinser52
Maybe this ASAN error has something to do with it? https://github.com/ldorau/unified-memory-framework/actions/runs/14083824988/job/39443119450 ``` ==12506==ERROR: AddressSanitizer: use-after-poison on address 0x7fa3f2a69188 at pc 0x55fdfc47fde2 bp 0x7fa3e93feb30 sp 0x7fa3e93feb20 READ of size 8 at...
See: https://github.com/oneapi-src/unified-memory-framework/pull/1224
> I am not sure because the issue above is related to the allocation flow, right? > Most probably, or the free() path.
The latest occurrence: https://github.com/oneapi-src/unified-memory-framework/actions/runs/16470636284/job/46558694579 and https://github.com/oneapi-src/unified-memory-framework/actions/runs/16470636284/job/46558694507
The latest occurrence: https://github.com/oneapi-src/unified-memory-framework/actions/runs/16617013654/job/47013068561