[PATCH] D109400: [LLD] Add archive Name to relocaiton overflow printout

Thu Nov 4 21:40:41 PDT 2021

wenlei added a comment.

> The description is not what this patch does. Do you mean D94141 <https://reviews.llvm.org/D94141>?

Yes, I was referring to that patch. It is all related to improving diagnostics on error.

> It is also not much to ask the user to re-link with an option.

I don't agree. 1) When builds are done on build agents, it's often not easy to just do "re-link". Then for people to repro this, a full rebuild may be needed which could take hours; 2) as I said above, it's too much to ask for an average non-compiler developer to deal with all of this, to know the `-M` + `sed` trick; 3) toolchain maintainers often don't have the setup to redo those builds locally.

Let's use an analogy - you're a user of a product, say a phone app, and when the app crashed, wouldn't it be nice if there's an diagnostic report you can send to app developers right way? Isn't it a bad user experience if after the crash, what users have to do is to go to phone settings, do some trick, and repeat all you've done again to repro the crash? Ofc, it's doable. But isn't it better to make things easier for everyone? Toolchain is also a product that serves its users. I'm yet to be convinced why such a harmless diagnostic improvements is not acceptable, given it certainly helps some users.

> As I said, I do not think auto -M or section output behavior is useful.

Well, for the reasons I mentioned above, evidently it's proven to be quite useful for us. It helped us to triage some known issues quickly without needing to rebuild and point people to proper solutions very efficiently. It routinely saves many people hours of extra work from extra build for repro. Also in case there's confusion, this is not unconditional auto `-M`, this is just better diagnostics when error actually happens.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D109400/new/

https://reviews.llvm.org/D109400