Should lto = false be treated as LTO "off"?

[ehuss]

Currently, a Cargo profile of lto = false is actually "thin local" LTO where LTO is performed across the local crate's codegen units. I'm wondering if that should actually be treated as -C lto=off where LTO is completely disabled?

This would change the behavior for manifests that set lto = false, which shows up occasionally since people copy/paste all the defaults.

Unfortunately there would be no way to explicitly express the default if this was changed, which I think would be a problem (or annoyance). We could add an explicit string ('thin-local'), and translate that to no -C lto flag, or add that to rustc itself.

Another option is to keep the current behavior, and just document that the default of false actually still performs some form of LTO, and that the string "off" should be used to completely turn it off. cc rust-lang/rust#65136 where I am proposing to document these flags on the rustc side.

I kinda lean towards the documentation side, since otherwise it would change behavior for some projects, and probably isn't too important to be absolutely correct here.

[alexcrichton]

I would personally be fine either way here, but I agree that if we change how we interpret lto=false that we whould add lto = 'thin-local' (or a better name).

[Rua]

I got pointed here via the bug I submitted above.

I think things would be the clearest for the user if this option didn't accept boolean values at all, as the setting isn't boolean anyway and making it seem like it is, is just confusing. Only the strings full/fat, thin, thin-local and off would be accepted. This ensures that people's existing settings don't get reinterpreted, existing boolean values would give an error instead which is more explicit. full is my own proposal as alternative (or replacement?) for fat.