Rémy Rakic
Rémy Rakic
@rust-timer build 7c1e66778214e445126667614fe1d6a72eb7d2ea
BTW do we know if the [+18% cycles](https://perf.rust-lang.org/compare.html?start=0fa7feaf3f287b900176061053619c06306c67b0&end=380a85bad098d06e75548877d9061302a9dd5312&stat=cycles%3Au&showRawData=true&name=externs) on `externs` are noise or something?
Absolutely, @bors try @rust-timer queue
I believe this should be due to the soundness fix in #121679.
(random passerby note about these last results, they show the worst case: for stability reasons the perf runs are done with incremental hash verification turned on all the time, instead...
> @lqd, I think I'm a bit confused by the [self_contained_components](https://github.com/rust-lang/rust/blob/20d8e3c2b94fe1f2823d2e0afd20377adcec8a6e/compiler/rustc_codegen_ssa/src/back/link.rs#L1746) function. It seems to take the CLI arguments into account, but only if it is `yes` or `no`. Otherwise...
> Maybe, but it's a big churn, probably not worth doing until everything stabilizes for sure. Looking at the diff, I agree. Let's keep the 3-variant enum for now and...
> So, I thought maybe we should do a 180 degree turn and shift _all_ the logic except principal flavors to `-Clinker-features`? > Then, for example, instead of the clang...
Sure, like I described in my earlier comment. Overall, I personally find this direction promising. A bunch of the support code could also be based on renaming the self-contained linking...
I meant you could imagine the possible linker features as a generalization of self-contained linking components. And get an approximation of how that'd look like by renaming `LinkSelfContainedComponents` to `LinkerFeatures`,...