Loïc Lecrenier

haha! @irevoire it's your turn to try afterwards!

I was starting to wonder as well. Since it is compiled with `opt-level = 0`, the binary will be bigger indeed :/ I'll check on my computer if it seems...

On my computer (macOS, so not necessarily the same), there doesn't seem to be any significant difference between the size of the `target/` folder before and after the PR. So...

nooooo 😭 I guess there really is something about that PR that makes rustc on Windows sad

It looks like mimalloc is indeed faster than jemalloc, so I have enabled it on all platforms now. ``` group indexing_main_950d8e4c indexing_mimalloc-benchmarks_fb2b6c0c ----- ---------------------- ------------------------------------- indexing/-geo-delete-facetedNumber-facetedGeo-searchable- 1.00 35.6±2.96ms ? ?/sec...

mimalloc is used for everything now. Well, almost. It could also be used as a dev-dependency for milli's tests I guess. But I don't think it is very important, and...

Yes, done :-)

Does anybody know why the tests are failing on Windows? 🤔

This didn't turn out to be good in terms of performance, so I'm dropping this PR for now.

Sorry about this :( I will try to have a look at it this weekend (along with your PR!)