Offer different allocator options for building Cargo itself

[weihanglo]

Problem

In #14691, some benchmarks show that using jemalloc or mimalloc boosts Cargo's performance. There is a discussion on Zulip about whether we should switch to a different allocator. However, there are concerns raised, such as the risk of changing the allocator on some platforms. Besides, allocations are not the largest contributor to Cargo's invocation time—algorithms in dependency resolution play a more significant role. It might not be worth the risk.

This issue is open for collecting feedback and use cases.

Proposed Solution

Do something like #14691 adding different Cargo feature to enable different allocators for downstream packagers.

Notes

No response

[daniel-pfeiffer]

Finding the optimal allocator for a given program is useful!

At the same time the bigger question would seem: how can a builder change this easily? Rather than downloading and first patching 2 files, I’d like to

cargo install --allocator=mimalloc …

Especially given glibc version-incompatibilities between different distros, if you want to prebuild binaries, you either use a very old distro or libc-musl. That however comes with an even slower allocator.