[PATCH] D63533: [analyzer] Fix clang-tidy crash on GCCAsmStmt

[Nick Desaulniers via Phabricator via cfe-commits]

nickdesaulniers added inline comments.


================
Comment at: clang/test/Analysis/egraph-asm-goto-no-crash.cpp:1
+// RUN: %clang_analyze_cc1 -analyzer-checker=core -analyzer-dump-egraph=%t.dot %s
+// RUN: cat %t.dot | FileCheck %s
----------------
NoQ wrote:
> NoQ wrote:
> > NoQ wrote:
> > > Ugh, you picked an exotic test as an example.
> > > 
> > > Let's try the following:
> > > ```lang=c++
> > > // RUN: %clang_analyze_cc1 -analyzer-checker=core,debug.ExprInspection -verify %s
> > > 
> > > // expected-no-diagnostics
> > > 
> > > void clang_analyzer_warnIfReached();
> > > 
> > > void testAsmGoto() {
> > >   asm goto("xor %0, %0\n je %l[label1]\n jl %l[label2]"
> > >            : /* no outputs */
> > >            : /* inputs */
> > >            : /* clobbers */
> > >            : label1, label2 /* any labels used */);
> > > 
> > >   label1:
> > >   // FIXME: Should be reachable.
> > >   clang_analyzer_warnIfReached();
> > >   return;
> > > 
> > >   label2:
> > >   // FIXME: Should be reachable.
> > >   clang_analyzer_warnIfReached();
> > >   return;
> > > }
> > > ```
> > > 
> > >  (and the egraph part in the main file is also out of place)
> > (wait, one of these shouldn't be reachable, right?)
> (i mean, let's do something similar, just with the correct amount of FIXMEs)
You'd have to "peak" into the assembly to tell.  Essentially `asm goto` is treated as a "black box" throughout Clang and LLVM, similar to vanilla inline assembly.  Basically, the explicit list of labels are valid branch targets from the inline assembly, as is fallthrough.  It's undefined behavior if the assembly jumps to a label not explicitly listed in the asm statement (but would likely fail to link, in the best case).


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D63533/new/

https://reviews.llvm.org/D63533