Szelethus added inline comments.
================
Comment at: clang/test/Analysis/stream.c:274-284
// Check that "location uniqueing" works.
// This results in reporting only one occurence of resource leak for a stream.
void check_leak_noreturn_2() {
FILE *F1 = tmpfile();
if (!F1)
return;
if (Test == 1) {
----------------
balazske wrote:
> NoQ wrote:
> > balazske wrote:
> > > Szelethus wrote:
> > > > Why did this change? Is there a sink in the return branch?
> > > The change is probably because D83115. Because the "uniqueing" one
> > > resource leak is reported from the two possible, and the order changes
> > > somehow (probably not the shortest is found first).
> > The shortest should still be found first. I strongly suggest debugging
> > this. Looks like a bug in suppress-on-sink.
> There is no code that ensures that the shortest path is reported. In this
> case one equivalence class is created with both bug reports. If
> `SuppressOnSink` is false the last one is returned from the list, otherwise
> the first one (`PathSensitiveBugReporter::findReportInEquivalenceClass`),
> this causes the difference (seems to be unrelated to D83115).
> There is no code that ensures that the shortest path is reported.
There absolutely should be -- See the summary of D65379 for more info, CTRL+F
"shortest" helps quite a bit as well. For each bug report, we create a bug path
(a path in the exploded graph from the root to the sepcific bug reports error
node), and sort them by length.
It all feels super awkward --
`PathSensitiveBugReporter::findReportInEquivalenceClass` picks out a bug report
from an equivalence class as you described, but that will only be reported if
it is a `BasicBugReport` (as implemented by
`PathSensitiveBugReporter::generateDiagnosticForConsumerMap`), otherwise it
should go through the graph cutting process etc.
So at the end of the day, the shortest path should appear still?
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D83120/new/
https://reviews.llvm.org/D83120
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
