================
@@ -1054,10 +1054,24 @@ const VarRegion *MemRegionManager::getVarRegion(const 
VarDecl *D,
     assert(!Ty.isNull());
     if (Ty.isConstQualified()) {
       sReg = getGlobalsRegion(MemRegion::GlobalImmutableSpaceRegionKind);
-    } else if (Ctx.getSourceManager().isInSystemHeader(D->getLocation())) {
-      sReg = getGlobalsRegion(MemRegion::GlobalSystemSpaceRegionKind);
     } else {
-      sReg = getGlobalsRegion(MemRegion::GlobalInternalSpaceRegionKind);
+      StringRef N = D->getNameAsString();
+      QualType FILETy = D->getASTContext().getFILEType();
+      if (!FILETy.isNull())
+        FILETy = FILETy.getCanonicalType();
+      Ty = Ty.getCanonicalType();
+      bool IsStdStreamVar = Ty->isPointerType() &&
+                            Ty->getPointeeType() == FILETy &&
+                            (N == "stdin" || N == "stdout" || N == "stderr");
+      // Pointer value of C standard streams is usually not modified by system
+      // calls. This means they should not get invalidated at system calls and
+      // can not belong to the system memory space.
----------------
NagyDonat wrote:

```suggestion
      // Pointer value of C standard streams is usually not modified by calls
      // to functions declared in system headers. This means that they should
      // not get invalidated by calls to functions declared in system headers,
      // so they are placed in the global internal space, which is not
      // invalidated by calls to functions declared in system headers.
```
I would prefer avoiding the phrase "system call" because it has a [very 
specific meaning](https://en.wikipedia.org/wiki/System_call) which is not what 
you speak about here.

Also, I extended the comment to describe the reason why is the global internal 
space a "better" place for these streams, because I felt that this makes it 
easier to quickly understand the goals of this logic.

https://github.com/llvm/llvm-project/pull/147766
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to