Add mimalloc on feature flag

[passcod]

No description provided.

[passcod]

I think the system linux allocator is pretty good, but perhaps we can use it for musl builds. No strong opinions, though.

[NobodyXu]

I think the system linux allocator is pretty good, but perhaps we can use it for musl builds. No strong opinions, though.

Musl's malloc is indeed very slow compared to glibc, but honestly, I don't think that is a bottleneck, at least I didn't observe that now.

Maybe when #175 is implemented it might make a difference.

[NobodyXu]

Looking at the CI build, looks like it doesn't add much to the binary size (0.1M).
So perhaps we can also do that for musl.

BTW, I think we have a couple of C libraries (zstd, libz-ng, xz2, bzip2), not sure whether they use malloc.

[NobodyXu]

BTW, I think we have a couple of C libraries (zstd, libz-ng, xz2, bzip2), not sure whether they use malloc.

Oh wait, that shouldn't be a problem since mimalloc defines malloc and free, which overrides the libc implementation.

[NobodyXu]

LGTM

[passcod]

Just for fun, ran some basic hyperfine benchmarks on linux (pulling the same archive every run, from a localhost server, not from the internet):

Benchmark 1: binches/x64-gnu-mimalloc --no-confirm --manifest-path . --install-path /tmp cargo-binstall
  Time (mean ± σ):     139.7 ms ±   7.1 ms    [User: 114.3 ms, System: 32.5 ms]
  Range (min … max):   128.5 ms … 161.8 ms    100 runs

Benchmark 2: binches/x64-gnu-sysalloc --no-confirm --manifest-path . --install-path /tmp cargo-binstall
  Time (mean ± σ):     139.1 ms ±   8.8 ms    [User: 114.7 ms, System: 30.5 ms]
  Range (min … max):   127.5 ms … 176.6 ms    100 runs

Benchmark 3: binches/x64-musl-mimalloc --no-confirm --manifest-path . --install-path /tmp cargo-binstall
  Time (mean ± σ):     137.7 ms ±   7.4 ms    [User: 113.2 ms, System: 30.3 ms]
  Range (min … max):   125.6 ms … 164.0 ms    100 runs

Benchmark 4: binches/x64-musl-muslalloc --no-confirm --manifest-path . --install-path /tmp cargo-binstall
  Time (mean ± σ):     134.9 ms ±   7.5 ms    [User: 110.8 ms, System: 27.5 ms]
  Range (min … max):   124.2 ms … 167.1 ms    100 runs

Summary
  'binches/x64-musl-muslalloc --no-confirm --manifest-path . --install-path /tmp cargo-binstall' ran
    1.02 ± 0.08 times faster than 'binches/x64-musl-mimalloc --no-confirm --manifest-path . --install-path /tmp cargo-binstall'
    1.03 ± 0.09 times faster than 'binches/x64-gnu-sysalloc --no-confirm --manifest-path . --install-path /tmp cargo-binstall'
    1.04 ± 0.08 times faster than 'binches/x64-gnu-mimalloc --no-confirm --manifest-path . --install-path /tmp cargo-binstall'

[NobodyXu]

Looks like cargo-binstall does not allocate/deallocate enough to have heap as a bottleneck.

It will also be interesting to run this again once we got #175