don't let rustc-dep-of-std enable backtrace-sys

[RalfJung]

Fixes #253

[alexcrichton]

I sort of forget the context for this at this point again, but is this somethign that we can condition the target string on? For example could a new target be added to print out the empty cfg? Or is this required as part of a build elsewhere or something like that?

[RalfJung]

The point of this is to be able to do a cross check build of libstd without the backtrace feature for foreign architectures with only a Rust toolchain. This is needed for rust-lang/miri#1198: since Miri is just an interpreter, in principle we should be able to emulate any target platform on any host platform. Since recently Miri just does check builds for everything, so rust-lang/rust#56443 is not a problem any more.

The only thing that still fails when I run xargo-check check --target x86_64-apple-darwin on my Linux box is that, even though the backtrace feature in libstd is disabled, libstd still depends on backtrace-sys (transitively) and its build script fails (as I have no apple toolchain and headers installed).

libstd without the backtrace feature depends on an "empty" backtrace with no features (you requested that in rust-lang/rust#64444 to avoid lots of cfg in libstd). Unfortunately, due to cargo limitations, the fact that backtrace forwards the rustc-dep-of-std feature to backtrace-sys means that the rustc-dep-of-std backtrace feature implicitly enables backtrace-sys. So, that "empty" backtrace crate is actually not quite so empty when rustc-dep-of-std is enabled.

is this somethign that we can condition the target string on

I don't follow... how would we condition the target string on anything? We don't control the target string?

[alexcrichton]

Ah ok that makes sense, thanks for the explanation!

The motivating case looks to be "I disabled backtraces in libstd but backtrace-sys is still getting built when checking for standard platforms", specifically in your cross build case. That's good enough for me at least :)

Ah and for conditioning the target string, I thought this was originally motivated by "whenever I build for platform X I want to disable backtraces" which I think is best handled in the code itself, not with more features. That's not the motivation for this, though, so it's all good!

[RalfJung]

The motivating case looks to be "I disabled backtraces in libstd but backtrace-sys is still getting built when checking for standard platforms", specifically in your cross build case.

Yes that's a good summary. :)