[PATCH] D149119: [CMake] Use LLVM own tools in extract_symbols.py

Wed May 10 00:00:39 PDT 2023

jhenderson added a comment.

In D149119#4331207 <https://reviews.llvm.org/D149119#4331207>, @ikudrin wrote:

> In D149119#4329274 <https://reviews.llvm.org/D149119#4329274>, @tmatheson wrote:
>
>> LGTM, thank you for doing this. Please give it a couple of days in case others have comments.
>
> Thanks!
>
> In D149119#4329285 <https://reviews.llvm.org/D149119#4329285>, @jhenderson wrote:
>
>> I've not really looked into this patch significantly, so this may well be addressed in the patch, given I see you have modified stuff to do with the NATIVE build, but in the past I have seen LLVM using its own tools to build other parts of its system. I believe it was using llvm-nm to extract the list of symbols needed for export, possibly to do with part of the clang build, possibly even using this script, I don't remember. The problem was that it was using the just-built version of llvm-nm, rather than specifically one from a release build. On a debug build this caused particularly slow builds for me, so much so that I stopped building the relevant parts of LLVM. Please don't introduce a similar situation/make the situation worse (it's quite possible this was fixed some time ago, but I haven't tried recently, nor do I remember the exact thing causing the issue): much like tablegen, any parts of the LLVM build that use just-built tools should make use of release builds, even in debug configuration, at least if an appropriate cmake option is specified.
>
> Your concerns are legit, but the tools in this patch follow the same principle as `TableGen`, i.e. if `LLVM_OPTIMIZED_TABLEGEN` is `ON` then the tools are forced to be built with optimization.

Thanks - that's fine with me (though raises the question as to whether we should be renaming that variable at some point...).

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D149119/new/

https://reviews.llvm.org/D149119