NoQ updated this revision to Diff 148684.
NoQ added a comment.
Add an explicit brute-force protection against re-entering `simplifySVal()`.
Remove the threshold completely.
https://reviews.llvm.org/D47155
Files:
include/clang/StaticAnalyzer/Core/PathSensitive/SValBuilder.h
NoQ added a comment.
I only essentially did one optimization - introduce a short path that returns
the original value if visiting its sub-values changed nothing, which is a
relatively common case. The reason it works though is that `evalBinOp()` will
be called later to combine the sub-values,
NoQ updated this revision to Diff 148681.
NoQ added a comment.
Optimize `simplifySVal()` instead of reducing the threshold.
I'll address the memoization separately.
https://reviews.llvm.org/D47155
Files:
include/clang/StaticAnalyzer/Core/PathSensitive/SValBuilder.h
baloghadamsoftware added a comment.
Hello! Thank you for addressing this problem. Are these kinds of symbols common
in real code? For me it seems very artificial. However, I agree with George, it
would be better to have this value as an analyzer option with a default value
(of 20).
george.karpenkov added a comment.
Also we should make sure that all recursive transformations on expressions
represented as DAGs should be memoized.
Repository:
rC Clang
https://reviews.llvm.org/D47155
___
cfe-commits mailing list
george.karpenkov added a comment.
@NoQ I'm really wary of magic numbers.
- We should expose it through an analyzer-config flag. We already do so for the
budget.
- We should probably have both positive and negative tests. What scenarios
_stop_ working after the threshold is decreased? [point 1
NoQ added a reviewer: mikhail.ramalho.
NoQ added a subscriber: mikhail.ramalho.
NoQ added a comment.
@mikhail.ramalho Does it solve your problems with ffmpeg as well? :)
Repository:
rC Clang
https://reviews.llvm.org/D47155
___
cfe-commits
NoQ created this revision.
NoQ added reviewers: dcoughlin, xazax.hun, a.sidorin, george.karpenkov, szepet,
rnkovacs.
Herald added subscribers: cfe-commits, baloghadamsoftware.
Reported by Mikael Holmén on
http://lists.llvm.org/pipermail/cfe-commits/Week-of-Mon-20180416/225349.html -
a
