[PATCH] D87272: [lld] Buffer writes when composing a single diagnostic

Sun Oct 4 21:49:03 PDT 2020

dblaikie added a comment.

In D87272#2310448 <https://reviews.llvm.org/D87272#2310448>, @MaskRay wrote:

> In D87272#2295704 <https://reviews.llvm.org/D87272#2295704>, @dblaikie wrote:
>
>> Not sure I follow - the patch description seems to be quite focussed on the performance implications of this change - it sounds like, reading the patch description/commit message that that is the motivation.
>>
>> Oh, perhaps I see another reason that might be less nuanced: This will probably break color printing on Windows when using the normal Windows color API - because it's a stateful API, not something that can be encoded in the stream: https://github.com/llvm/llvm-project/blob/d3f6972abb9c0ac06ddabf61697754c3c6f5d11b/llvm/lib/Support/raw_ostream.cpp#L513
>
> I find another reason, probably still weak: I want to check a large binary with many out of range relocations. lld will emit a large number of warnings (if --noinhibit-exec). This patch can make it faster.

Could the uninteresting warnings be disabled? if all the warnings are interesting, I'd guess a bunch of processing would need to be done to make it actionable - or a lot of human time to read through them all, either way that time probably dwarfs the extra time lld takes to print those errors.

In any case - I think colour can't be portably buffered (well, in addition to the existing buffering in the ostream itself) so far as I can see in the implementation. Does that understanding seem right? Is there a way that could be addressed?

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D87272/new/

https://reviews.llvm.org/D87272