Conversation
Diff: - Don't look for old RUSTC_DEBUGINFO vars (rust-lang/backtrace-rs#313)
rust-highfive
added
the
S-waiting-on-review
|
Just to be sure, we didn't accidentally regress something here? i.e. we didn't stop putting debuginfo in unintentionally? Regardless these are unused so @bors r+ rollup (can't make the situation worse) |
|
📌 Commit 2bafb1b has been approved by |
bors
added
S-waiting-on-bors
It seems like we probably did regress at some point. I'm not sure of the history myself, but from the code it looks like backtrace-sys's build was coupled with some details of how bootstrap works (which then got removed). This removes the coupling and makes it behave like any other crate, which bootstrap now supports. |
|
Hm, well, we did switch away from custom handling via the rustc shim to proper setting of Cargo's environment variables I believe for the debuginfo, so we may not have regressed? I don't personally care too much, I suspect it doesn't matter in practice. |
Rollup of 8 pull requests Successful merges: - rust-lang#71148 (Vec drop and truncate: drop using raw slice *mut [T]) - rust-lang#71465 (Add a convenience method on `TyCtxt` for checking for thread locals) - rust-lang#71567 (Handle build completion message from Cargo) - rust-lang#71590 (MIR dump: print pointers consistently with Miri output) - rust-lang#71682 (Bump pulldown-cmark) - rust-lang#71688 (Allow `Downcast` projections unconditionally in const-checking) - rust-lang#71691 (Allow `Unreachable` terminators unconditionally in const-checking) - rust-lang#71719 (Update backtrace-sys) Failed merges: r? @ghost
|
I made the fix in rust-lang/backtrace-rs#313 because I noticed missing debuginfo for libbacktrace, and this fixes that. Whether it was always missing, I don't know. |
Could #71060 be related to that? |
|
I believe this is fixing a regression from #64316 that wasn't noticed until recently. Before that PR we used custom env vars to specify debuginfo to our wrapper script, and the build script in backtrace-sys, which was extracted long ago, tried to mirror this and respect the debuginfo choices of libstd. In #64316 we switched to using recently-added Cargo features to specify things like debuginfo, which causes Cargo to pass vars to build scripts conventionally rather than in a custom manner. Basically 1.39.0 (first release with 7fdea7a) I think libbacktrace hasn't had debuginfo when it should (since libstd has line-tables debuginfo). This I believe fixes that regression where libbacktrace should have debuginfo again. FWIW #71060 looks like debuginfo is just missing completely and/or something about symbolication is totally busted. That's probably a MinGW-specific (or Windows-specific?) issue about either finding debuginfo or dealing with maybe something like ASLR incorrectly. |
6 participants
Diff:
This fixes an issue of libbacktrace never being built with debuginfo.
r? @Mark-Simulacrum
cc @alexcrichton