[PATCH] D16569: Emit line 0 line information for interesting 'orphan' instructions

[Paul Robinson via llvm-commits]

probinson added a comment.

Our FastISel patch dealt with line info for post-call local-value instructions mainly as an artifact of the implementation, not because we had complaints or thought it was an important case.  I think omitting that case is fine.  The line info for the call instruction implicitly covers the following local-value instructions, which may not be pedantically "correct" but does not have the funny stepping behavior problems that the top-of-block case has.
Your description of the inlining case sounds like a reasonable justification for the PrevLabel case, and yes it would be interesting to see the example as an experiment for our debugger.

Did you happen to try an always_inline+nodebug case?  Lots of the x86 intrinsics do this, for example.


http://reviews.llvm.org/D16569