[PATCH] D109632: [clang] de-duplicate methods from AST files

Thu Oct 21 18:53:08 PDT 2021

dexonsmith added a comment.

In D109632#3079745 <https://reviews.llvm.org/D109632#3079745>, @vsapsai wrote:

> In D109632#3079455 <https://reviews.llvm.org/D109632#3079455>, @rmaz wrote:
>
>> So given these numbers are we good to go ahead with set dedupe approach?
>
> I'd rather get an opinion on this from other reviewers. It's not purely a numbers game, there can be other reasons to prefer one solution over another. I am biased, so I don't think I can make this call. If reviewers want a short summary instead of checking the entire discussion, we can prepare that.

Can you summarize how each of the two proposed new architectures differ from the baseline?

Note also that clang/lib/Serialization/MultiOnDiskHashTable.h is an option if there's a tradeoff between redundantly repeating information in module files (optimize for storage) and the number of module files that need to be visited (optimize for having information available). I think I mentioned it way up thread (indirectly, when talking about identifier tables), but I'm not sure if it was considered.

> If anybody is interested in the raw measurement data, I can provide it. The data is suitable for calculating the statistical significance between different methods using t-test.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D109632/new/

https://reviews.llvm.org/D109632