[musl is] slower than glibc, and in occasionally difficult to predict spots
Note that sometimes (/often) this purely due to the allocator, and as far as I remember Chimera uses a different allocator than the stock musl one.
Note that sometimes (/often) this purely due to the allocator, and as far as I remember Chimera uses a different allocator than the stock musl one.
What’s actually amazing about chimera is the musl + clang/llvm/compiler-rt combination. It allows you to do all kinds of amazing things easily from a programming perspective.
The apk 3 package manager, build system (cbuild) is also excellent. Everything feels modern and clean. Package availability is still limited and you would probably need to compile many of your favourite packages from source but I would still highly recommend this distribution.
Any examples?
As such, I have stuck with GCC which requires much less space.
Is this any different on Chimera.
I have 2TB of internal SSD storage on my laptop, and I can buy extra TB's for $50 each.
Why on earth would I balk at 100MB?
Some webpages serve half of that in optimized images for crying out loud, lol.
Yet I can't find many benchmarks.
Frustratingly, it's listed here, but not actually included in the results: https://github.com/daanx/mimalloc-bench Might be fun trying to run the benchmarks myself.
that said, scudo is highly configurable, and the performance reflects the configuration (you can even swap out the primary allocator, secondary cache, you can tune all the parameters, the TSD registry implementation has a big impact on multithreaded code, etc.), so it's not 100% representative - chimera's current configuration is further tweaked for both better performance and lower memory usage
scudo being this configurable is excellent because it is what enables such integration in the first place (other allocators e.g. frequently rely on ELF TLS, i.e. __thread and the likes, which would make libc integration very difficult and would require major changes to the dynamic linker, with scudo we instead implement a custom TSD registry and simply shove an extra pointer in the pthread structure)