[PATCH] D60974: Clang IFSO driver action.

Mon Jul 1 16:05:37 PDT 2019

plotfi added a comment.

In D60974#1565632 <https://reviews.llvm.org/D60974#1565632>, @jfb wrote:

> In D60974#1565621 <https://reviews.llvm.org/D60974#1565621>, @plotfi wrote:
>
> > In D60974#1565577 <https://reviews.llvm.org/D60974#1565577>, @jfb wrote:
> >
> > > Looking at the code quickly, I'm not sure that this should be in clang itself. It sounds like a better fit for a clang-based tool, and not clang. Why does it need to be part of clang?
> >
> >
> > @jfb I think this could actually be refactored into a clang based tool come to think of it, but I am also looking at how to add some features to the driver so that it can support job pipelines that are different from the standard PP -> CC1 -> BE -> ASM -> LINK pipeline. I can work on a change once I sort out this pipeline work to move the interface generation into a clang tool so that instead of invoking clang -cc1 to generate the interfaces the clang driver invokes the new clang interface generation tool.
>
>
> That sounds pretty good, thanks! Maybe someone like @ributzka / @arphaman / @Bigcheese have opinions on how to best do this.
>  Once you do this, can you also revert rL364855 <https://reviews.llvm.org/rL364855>?

That is definitely possible, because then only the clang-tool that emits the interfaces needs to verify against llvm-readelf. The the driver invocation path can just test against the llvm-nm output of the final .so file. More collaboration and feedback on this project is very welcome.

Currently @compnerd and I have been discussing allowing for a more flexible compilation pipeline generation, where possibly instead of the standard PP -> CC -> BE -> ASM -> LNK we might have something like PP -> clang-ifsgen -> llvm-ifsmerge -> llvm-elfabi.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D60974/new/

https://reviews.llvm.org/D60974