Add support for -Cforce-frame-poiners=non-leaf #125127
Conversation
rustbot
added
S-waiting-on-review
|
Some changes occurred in compiler/rustc_codegen_cranelift cc @bjorn3 |
|
Doesn't #124733 try to do this already? |
This brings the command-line option up to parity with the frame-pointers field in the json target specs.
jsgf
force-pushed
the
frame-pointers-non-leaf
branch
2 times, most recently
from
3be218c
to
3112454
Compare
|
Yes, I'm sorry @jsgf but I believe my PR already completely supersedes yours, which would require all the additional bits that mine handles (namely, landing behind |
|
I suppose I didn't add the "always" option, I would be happy to include that. |
jsgf
force-pushed
the
frame-pointers-non-leaf
branch
from
3112454
to
49ea4e2
Compare
|
I believe the "never" option is confusing as it suggests we will not force frame pointers or that we will force-them-to-not-occur, but we do demand frame pointers from the backend when it is required for ABI compliance, and not allowing people to turn this off is quite intentional as I understand it, as then it breaks important elements of runtime support. |
|
@workingjubilee @jieyouxu Oh yeah, looks like it. I guess I should have done a search. Like you I was surprised this wasn't already an option. |
Yeah I was in two minds about this - the alternative is to just reflect the "MayOmit" terminology from the enum |
| fp = FramePointer::Always; | ||
| match (opts.unstable_opts.instrument_mcount, opts.cg.force_frame_pointers) { | ||
| (true, _) => fp = FramePointer::Always, | ||
| (_, Some(fp_opt)) => fp = fp_opt, |
There was a problem hiding this comment.
This changes the behavior. The code did not previously override the value of fp if force_frame_pointers was Some(false).
This brings the command-line option up to parity with the frame-pointers field in the json target specs. It's still backwards compatible with before but now also accepts
always,neverandnon-leaf.