rustc: Add the ability to not run dsymutil

[alexcrichton]

This commit adds the ability for rustc to not run dsymutil by default
on OSX. A new codegen option, -Z run-dsymutil=no, was added to specify
that dsymutil should not run and instead the compiler should
unconditionally keep the object files around in a compilation if
necessary for debug information.

cc #47240

[rust-highfive]

r? @estebank

(rust_highfive has picked a reviewer for you, use r? to override)

[alexcrichton]

I'm sort of hoping that we can use a function like this to enable split dwarf on Linux over time as well!

[alexcrichton]

r? @michaelwoerister

[kennytm]

Would -C run-dsymutil=no cause RUST_BACKTRACE=1 to not have any line numbers?

[alexcrichton]

A good question and one I forgot to verify! Looks like the answer is "no" though which is a bummer!

@JohnColanduoni, out of curiosity, do you know if there's a way to get libbacktrace to relatively easily do the same logic that lldb is doing? (looking for object files in addition to the dSYM)

[JohnColanduoni]

@alexchrichton Not easily, it would probably make more sense to add this feature as part of a pure-Rust rewrite of libbacktrace. I didn’t know lldb had this functionality on Mac, I’ve had issues where lldb refuses to give me line numbers if Spotlight hasn’t yet indexed tbe dSYM.

[alexcrichton]

Ah ok, thanks for the info @JohnColanduoni! In light of that I've now switched this to a -Z flag instead of a -C flag so we can continue to tweak it. I think that we'll want this option though to play around with over time still with incremental compilation.

[kennytm]

I think you've changed the wrong variable...

[00:18:57] error[E0609]: no field `linker` on type `rustc::session::config::DebuggingOptions`
[00:18:57]   --> librustc_trans/back/link.rs:80:56
[00:18:57]    |
[00:18:57] 80 |     if let Some(ref linker) = sess.opts.debugging_opts.linker {
[00:18:57]    |                                                        ^^^^^^ unknown field
[00:18:57]    |
[00:18:57]    = note: available fields are: `codegen_backend`, `verbose`, `span_free_formats`, `identify_regions`, `emit_end_regions` ... and 89 others

[alexcrichton]

er oops!

[kennytm]

... the cg ↦ debugging_opts should be applied here 😄

[00:18:57] error[E0609]: no field `run_dsymutil` on type `rustc::session::config::CodegenOptions`
[00:18:57]    --> librustc_trans/back/link.rs:224:28
[00:18:57]     |
[00:18:57] 224 |         match sess.opts.cg.run_dsymutil {
[00:18:57]     |                            ^^^^^^^^^^^^ unknown field
[00:18:57]     |
[00:18:57]     = note: available fields are: `ar`, `linker`, `link_arg`, `link_args`, `link_dead_code` ... and 28 others

[BatmanAoD]

@kennytm @alexcrichton Something seems odd here; GitHub isn't displaying the commits in which kennytm's review comments were resolved, if indeed they were resolved. What is the status on this? (Also, does @michaelwoerister still need to review, since kennytm already did?)

[alexcrichton]

Ah I think we're just waiting on review! @michaelwoerister I believe is on leave right now, but this isn't a particularly pressing PR so I'd be fine waiting.

[bors]

☔ The latest upstream changes (presumably #48053) made this pull request unmergeable. Please resolve the merge conflicts.

[michaelwoerister]

Looks good to me! Thanks, @alexcrichton!

[michaelwoerister]

Not that it makes much of a difference but this if could be pulled out of the loop, I think.

[alexcrichton]

@bors: r=michaelwoerister

[bors]

📌 Commit 0f16eee has been approved by michaelwoerister

[alexcrichton]

@bors: rollup