Pass --strip-debug to GccLinker when building without debuginfo

[kyrias]

C.f. #46034

---

This brings a hello-world built by passing rustc no command line options from 2.9M to 592K on Linux.

(This might need to special case MacOS or Windows, not sure if the linkers there support --strip-debug, and there is an annoying lack of dependable docs for the linkers there.)

[rust-highfive]

r? @petrochenkov

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

[retep998]

This might need to special case MacOS or Windows

For pc-windows-gnu it depends on whether the MinGW linker needs to shell out to another binary to handle stripping, or whether it does it itself, and also whether it is even capable of doing so.

pc-windows-msvc already uses split debuginfo so this is a non-issue for that target.

[kyrias]

It's a linker argument, so it's passed to the linker. Specifically I am not sure whether ld.exe and the MacOS ld supports it, and there's no good docs for either of them that I can find at all. So the only way to find out that I know of would be a compile test.

[petrochenkov]

r? @michaelwoerister probably

[michaelwoerister]

Looks good, @kyrias! On OSX the flag is called -S. Can't hurt to add it there too.

[kyrias]

Ah, you sure that they don't have --strip-debug as well? I prefer the long and readable names, but the Linux ld has -S as well so we don't need to match on the different platforms either way.

[alexcrichton]

Are we sure we want to do this? Just because the final compilation doesn't have -g doesn't mean that the intermediate debuginfo is not desired, right?

This indeed does fix a "first impression" of Rust on linux sort of but I think it could hurt legitimate use cases where debuginfo is only applied to intermediate crates, right?

[michaelwoerister]

Good point, @alexcrichton!

I see two options:

• We leave things as they are and tell people to run strip if binary size is a concern for them; and point them to @alexcrichton's reasoning.
• We strip by default (since this is probably the more common case) and add a flag that allows to control this behavior.

[michaelwoerister]

@kyrias The man page at least does not list the long version.

[kyrias]

I would prefer the latter option, especially since anyone who would want to do it already needs to do some funky custom build system calling rustc directly.

[kyrias]

I added the boolean -Znever-strip-debug-info now. So now we by default tell the linker to strip the debug info whenever we're building with debuginfo=0, unless -Znever-strip-debug-info is passed to rustc.

[alexcrichton]

Can the defaults remain as they were? Until we stabilize the option I don't think we should change the default behavior

[kyrias]

Done, now you need to pass -Zstrip-debuginfo-if-disabled=yes.

[michaelwoerister]

Thanks, @kyrias!

@bors r+

[bors]

📌 Commit 2f0dd4a has been approved by michaelwoerister

[bors]

⌛ Testing commit 2f0dd4a with merge d351805...

[bors]

☀️ Test successful - status-appveyor, status-travis
Approved by: michaelwoerister
Pushing d351805 to master...

[ghost]

shouldnt this be --strip-all (-s) instead of --strip-debug (-S)?

rust-lang/cargo#4122 (comment)

[mati865]

@cup for most users -strip-debug is a safer bet, --strip-all isn't good choice for everything as removes things like relocation table.