Use raw_c_allocator when not in Debug

[steeve]

In our case (lots of C headers), codegen went from ~90s to less than 1s.

[Vexu]

Is SmpAllocator suitable for a single threaded application? In Aro we link libc and use raw_c_allocator in release mode.

[steeve]

My understanding was that it seeks to replace std.heap.GeneralPurposeAllocator but also happens to be somewhat faster than std.heap.c_allocator, without the need for libc, too.

According to https://ziglang.org/documentation/0.15.2/#toc-Choosing-an-Allocator

If you are compiling in ReleaseFast mode, std.heap.smp_allocator is a solid choice for a general purpose allocator.

See also: ziglang/zig#22808

[Vexu]

Yes but it also explicitly asserts that it is not used in single threaded mode. I think it would be stick to the C allocator for now.

[steeve]

@Vexu so force linking with -lc for the translate-c binary ? i'm good with that if you are

also, does translate-c have to be built in single threaded mode ?

[Vexu]

so force linking with -lc for the translate-c binary ? i'm good with that if you are

No need to force it, just link libc in release mode and use raw_c_allocator if libc is linked.

also, does translate-c have to be built in single threaded mode ?

It doesn't have to but it should since it is single threaded.