Printing to stdout/stderr no longer works during rustc bootstrap

[petrochenkov]

Any attempt to do "printf debugging" with println, or bootstrapping with -Z time-passes, or outputting any other useful information to stdout from rustc results in "failed to parse process output" reported from Cargo (https://github.com/rust-lang/cargo/blob/668d55cc6d54c7e9635c924c252621f5e6664e00/src/cargo/util/process_builder.rs#L205).

This is a regressions from the recent few months, I believe.
Printf debugging can be emulated with hacks like sess.span_warn() or similar, or by recompiling everything and using debug!(...), but I haven't found a workaround for time-passes yet.

Writing to stderr (e.g. eprintln) doesn't work as well, the output is silently ignored for some reason, but I haven't investigated why this happens. (UPDATE: The stderr issue was fixed in #47634)

Any workarounds are appreciated.

cc someone familiar with Cargo and infrastructure @Mark-Simulacrum @alexcrichton

[alexcrichton]

I believe this is because we switched to calling Cargo with --message-format json but we force the compiler through "hacks" to not actually emit json but continue to emit data on stderr. In this mode Cargo asserts stdout is empty and forwards stderr but tries to parse JSON.

That seems definitely fixable in Cargo (just forwarding stdout from the compiler) but I'm not entirely sure why printing to stderr wouldn't work, in theory it should get forwarded all the way out. In rustbuild we don't even try to capture stderr...

[devurandom]

This might be somewhat fixed by #47634 and is probably duplicated by #44244

[wesleywiser]

Just ran into this myself. eprintln!() instead of println!() does work though.

[petrochenkov]

Not sure about println, but eprintln works now and that's enough.