I'll let Adrian weigh in - but I suspect, if possible, it'd be better to just finalize them all once IRGen for the function finishes. On Tue, May 30, 2017 at 6:02 PM Keno Fischer <k...@juliacomputing.com> wrote:
> For some reason your comments aren't showing up on Phabricator, so > replying via email - hope everybody sees it. > > If SPs are now all finalized early - I'm assuming there's some other code >> that finalizes SPs that's no longer needed? >> > > Not all of them, just the one before cloning. I wanted to make the minimal > change possible here, since this already seems like a bit of a hack to work > around the fact that we can't properly forward variadic function calls at > the IR level. > > >> Also - this seems like a layering/separation issue - does other code call >> into the DI as casually/explicitly as this code being added to CGVTables? >> (Or should the callback go through some more generic entry point in >> CGDebugInfo to delegate/handle that?) & is there nothing in CGDebugInfo >> that is already called back in a timely fashion to handle this? I'm >> guessing what's needed is a "IR has finished being generated for this >> function, opt/codegen is about to start"? That seems like a good generic >> callback & probably shouldn't be coming from something called CGVTables? >> (presuming there's a more general place to put that) >> > > I suppose there's CGDebugInfo::EmitFunctionEnd, so if we did want to > finalize every SP as soon as we're done with it, we could put an explicit > call to that extent there. Frankly though I don't really know the clang > code too well, I just want to unbreak LLVM ;). >
_______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits