vitalybuka wrote:
> > @MaskRay can you talk more about your concerns? This won't affect compile
> > time, and instrumenting a shadow load will cause an immediate crash of the
> > application at runtime. This patch seems much more robust than counting on
> > toolchains to always arrange to avoid double instrumentation.
>
> From Eli: I'm a little skeptical that you pass pipeline is really doing what
> you want if you end up running sanitizer passes twice; that indicates you're
> running a bunch of other optimization passes twice without a good reason.
>
> The other passes will affect compile time. There are many more
> instrumentation passes than sanitizers. I am concerned this leads us to the
> rabbit hole to add others (or others would blindly copy the pattern here,
> when the use cases aren't recommended/endorsed for other platforms).
If we wanted to prevent for real we'd implement a diagnostics. The current
state looks worth the the state with the patch.
Maybe this way?
```
copt<bool> IgnoreRerundantIntrumentation;
bool checkIfAlreadyIntrumenter(string_view flag) {
if (M.getModuleFlag(flag)) {
if (IgnoreRerundantIntrumentation)
return PreservedAnalyses::all();
diag();
}
M.addModuleFlag(Module::ModFlagBehavior::Override, flag, 1);
}
::run() {
if (checkIfAlreadyIntrumenter("my_sanitizer"))
return;
}
```
https://github.com/llvm/llvm-project/pull/99439
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
