Update mimalloc to microsoft/mimalloc@9b75377 #23027
Conversation
```
../../../system/lib/mimalloc/src/os.c:36:3: error: implicit conversion from 'unsigned long long' to 'size_t' (aka 'unsigned long') changes value from 4294967296 to 0 [-Werror,-Wconstant-conversion]
32 | static mi_os_mem_config_t mi_os_mem_config = {
| ~
33 | 4096, // page size
34 | 0, // large page size (usually 2MiB)
35 | 4096, // allocation granularity
36 | MI_DEFAULT_PHYSICAL_MEMORY,
| ^~~~~~~~~~~~~~~~~~~~~~~~~~
../../../system/lib/mimalloc/src/os.c:26:40: note: expanded from macro 'MI_DEFAULT_PHYSICAL_MEMORY'
26 | #define MI_DEFAULT_PHYSICAL_MEMORY 4*MI_GiB
| ~^~~~~~~
1 error generated.
```
|
Context: this PR seems to address a suspicious(?) OOM error on wasm-vips' test suite. See for example the CI results for commit kleisauke/wasm-vips@605b53f and kleisauke/wasm-vips@5cb048b. It's generally recommended to use tagged upstream versions, but I prefer to avoid those that are known to have issues. Note that the previous version of mimalloc (used prior to PR #21548) was commit microsoft/mimalloc@4e50d67, which is also an untagged version. |
| #define MIMALLOC_H | ||
|
|
||
| #define MI_MALLOC_VERSION 217 // major + 2 digits minor | ||
| #define MI_MALLOC_VERSION 188 // major + 2 digits minor |
There was a problem hiding this comment.
Ah, I just noticed that I synced with the dev branch. Perhaps it should be synced with the dev-slice branch instead.
emscripten/system/lib/mimalloc/readme.md
Lines 71 to 76 in 8d59f19
|
Updating to the latest commit on the Upon further investigation, it appears to be the same issue as outlined in #20645 (comment), i.e. mimalloc requests an alignment of 4 MiB, causing emmalloc to allocate 8 MiB (or perhaps was that fixed via 3cd6861?), which somehow exceeds the Closing in favor of #23037. |
No description provided.