fix: only executables should choose a global allocator

[0xaatif]

This is general rust wisdom, and it prevents me linking two different versions of evm_arithmetization in #275

[0xaatif]

FWIW I would normally want to see profiling before opting-in to any allocator change, but I don't want to get bogged down in that discussion for now

[Nashtare]

Sinilarly to your comment about opting-in a global allocator, I would expect profiling results before justifying removal. I believe Jacqui had good reasons bringing in the past and if there is a performance gain, and it is non-negligible, it may be worth considering alternative options?

Especially as there is little point in having both type1 / type2 live concurrently, hence the issues faced in the past by Brendan and possibly you now as well shouldn't be actually happening once we get to a more evolved state.

If we make it the default on zero-bin side then it should be fine though, but full removal should wait IMHO.

[0xaatif]

Thanks for the review,

Sinilarly to your comment about opting-in a global allocator, I would expect profiling results before justifying removal. I believe Jacqui had good reasons bringing in the past and if there is a performance gain, and it is non-negligible, it may be worth considering alternative options?

I don't think PR constitutes removal - it's changing who is specifying the global allocator (which is not changing) - perhaps your comment belongs in #302?

There's a bit of nuance in this PR: each executable has a global allocator, and I count 5 executables in the repo:

• evm_arithmetization/src/bin/assemble.rs
• zero_bin/leader/src/main.rs
• zero_bin/rpc/src/main.rs
• zero_bin/verifier/src/main.rs
• zero_bin/worker/src/main.rs

I expect only worker to be on the allocation fast path, so that's the only one I've annotated - happy to be corrected on this though - I'll add #[global_allocator]s to the execs as needed.

To summarize:

• we agree that we should do profiling before committing to/from an allocator.
• we agree that executables should decide the allocator

  If we make it the default on zero-bin side then it should be fine though

• we (may) disagree on which binaries opt-in to an allocator

What are the next steps to approve this PR?

[Nashtare]

I don't think PR constitutes removal - it's changing who is specifying the global allocator (which is not changing) - perhaps your comment belongs in #302?

Right it probably have been better suited within #302, I was just sharing concerns about removing jemalloc without further evidences it didn't bring significant savings, just for the benefits of "convenience".