I had a look and i don't observe any regressions here. Both this test
alone and the whole test suite in general take as much time as on
r369520 for me. Additionally, -analyzer-stats doesn't indicate that any
significant amount of time was spent in bug report post-processing.
On 8/23/19 11:41 A
Totally possible, thanks for letting me know! There should be plenty of
room for caching, because I do calculate control dependencies in an excess
for the same CFG, and the retrieval of a basic block from an ExplodedNode
is anything but efficient, though I honestly didnt expect a performance hit
th
I suspect that this patch makes analysis much slower in certain cases. For
example, the clang/test/Analysis/pr37802.cpp test has become ~5 times
slower in some configurations in our environment. This happened somewhere
between r369520 and r369679, and your series of patches seems most
suspicious :)
