ikudrin abandoned this revision.
ikudrin added a comment.
This change is not needed anymore because the whole issue was fixed in
http://reviews.llvm.org/D20286.
http://reviews.llvm.org/D20287
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
ikudrin added a comment.
In http://reviews.llvm.org/D20287#431055, @davidxl wrote:
> On the other hand, I wonder what is the real root cause of the problem. The
> dummy function record does not have its 'own' profile counts, so
>
> if (std::error_code EC = ProfileReader.getFunctionCounts(
>
davidxl added a comment.
Strictly speaking, this patch requires a version bump of the indexed format.
The profile reader also needs to adjust the FunctionHash computation (either
using 0 or simple function hash) based on the version of the profile data.
Check with Justin/vsk to see if it is imp
ikudrin added a comment.
Does anyone known, why we need dummy coverage mapping records for unused
functions? How are they used? Isn't it better to remove these dummy records to
prevent confusion with the real ones?
http://reviews.llvm.org/D20287
_
ikudrin added a comment.
The motivation sample (using llvm-cov with http://reviews.llvm.org/D20286
applied):
$ cat > sample.h << EOF
inline int sample_func(int A) {
return A;
}
EOF
$ cat > dummy.cpp << EOF
#include "sample.h"
EOF
$ cat > sample.cpp << EOF
#include "sample.h
ikudrin created this revision.
ikudrin added reviewers: bogner, davidxl, vsk.
ikudrin added a subscriber: cfe-commits.
A valid function might not have any statement which affects the hash value, so
the hash for that function was zero. The hash value for an unused function is
also zero, so the lo
