stage0 libstd fails to build when jemalloc is disabled

[cuviper]

Since #43320 bumped the bootstrap compiler, stage0 libstd fails to build when jemalloc is disabled:

error[E0463]: can't find crate for `alloc_jemalloc`

error: aborting due to previous error

error: Could not compile `std`.

I believe it needs to restore the logic removed here:
https://github.com/rust-lang/rust/pull/42727/files#diff-242481015141f373dcb178e93cffa850L236

I tried to add this to libstd/lib.rs:

// Always use alloc_system during stage0 since we don't know if the alloc_*
// crate the stage0 compiler will pick by default is available (most
// obviously, if the user has disabled jemalloc in `./configure`).
#![cfg_attr(any(stage0, feature = "force_alloc_system"), feature(global_allocator))]
#![cfg(any(stage0, feature = "force_alloc_system"))]
#[global_allocator]
static ALLOC: alloc_system::System = alloc_system::System;

Then stage0 passed, but stage1 std failed like:

error[E0463]: can't find crate for `std`

error: aborting due to previous error

error: Could not compile `std`.

😕

[Mark-Simulacrum]

cc @alexcrichton

[alexcrichton]

We should probably add a builder for this...

[cuviper]

or just get on with evicting jemalloc... 😉

[aidanhs]

force_alloc_system is not the right way to fix this - I added that so I could build rust with a local rust that does not have jemalloc i.e. unless the following two conditions are true:

• you're using a local rust (as opposed to the automatic stage0 download) and
• your local rust does not have jemalloc (you compiled the whole thing yourself with --disable-jemalloc)

it's irrelevant.

Edit: to be clear, by "not the right way to fix this" I mean "it worked fine before I added that flag, so its removal is probably not the root cause". That's doesn't exclude the possibility that we could perform some incantations with this flag and fix the issue.

Edit2: sorry, I thought you were exclusively referring to force_alloc_system rather than all the stage0 logic. I read #43524 and thought this issue was about the same thing.

[cuviper]

Generally, the two conditions you list are actually true for me. I build Fedora's rustc without jemalloc, and I use that as the local stage0 to build the next release. And i think the force flag does come into play too, because if we need to rebuild the current release then the local rust gets treated as a stage1.

But right now it's more direct. I am using the downloaded stage0, which wants to default to jemalloc. Since I'm disabling that for my build, stage0 can't find it.

[aidanhs]

Interesting, I kinda figured I was the only one! Rereading #39086, I see I'm wrong about my two conditions above - it actually applies when trying to build something without jemalloc, from something that defaults to it.

Anyway, it should all be fixed soonish...