Implement variable lifetimes

[sapir]

I mean the BuilderMethods::lifetime_start and lifetime_end methods which would be implemented here:

rustc_codegen_gcc/src/builder.rs

Lines 1661 to 1667 in 764770d

	fn lifetime_start(&mut self, _ptr: RValue<'gcc>, _size: Size) {
	// TODO(antoyo)
	}
	fn lifetime_end(&mut self, _ptr: RValue<'gcc>, _size: Size) {
	// TODO(antoyo)
	}

If I understand correctly, this would allow reusing variable stack space.

How could this be done? Could the variable lifetimes be mapped to gcc blocks?

[antoyo]

I am not sure. I asked on IRC and they mentionned CLOBBER_OBJECT_BEGIN, but that it is not equivalent.

[sapir]

I didn't know that existed. How is it different? I found this mail with an explanation that does sound similar

https://gcc.gnu.org/pipermail/gcc/2024-February/243353.html

[antoyo]

I do not know. Here's what people said about it on IRC the other day:

to get exactly the same .. I suppose we’d have to introduce a new internal fn or so .. and wire it up in some way to indicate that object initialization was complete .. seems like moderately heavy lifting.
(or maybe we could have some flag on a VAR_DECL .. that the init code sets .. dunno)

the tree clobbers are used for optimisation (especially that the lifetime has ended - and therefore the object is not longer usable / memory can be reused) .. but not I can’t really answer whether the start-of-lifetime would be more useful than the start of the CTOR .. (and I’ve not been amongst the LLVM middle end for like 8 years .. so it’s likely all changed :) )
you’d need ME folks to answer whether the lifetime start is poentially useful (although maybe the idea has been discussed)

You might need to ask on IRC or the mailing list to get a proper answer.