Fix crash for diagnostics around source with embedded nulls (microsoft#8164)

Previously if the source contained embedded nulls and a diagnostic range
included that this would result in an assert firing (and then a crash in
builds without asserts).

This change makes it quietly return in this case, and adds a regression
test for this situation.

---------

Co-authored-by: Copilot <[email protected]>

Author: damyanp

tools/clang/lib/Frontend/TextDiagnostic.cpp

@@ -965,9 +965,10 @@ static void highlightRange(const CharSourceRange &R,
       EndColNo = map.startOfPreviousColumn(EndColNo);
 
     // If the start/end passed each other, then we are trying to highlight a
-    // range that just exists in whitespace, which must be some sort of other
-    // bug.
-    assert(StartColNo <= EndColNo && "Trying to highlight whitespace??");
+    // range that just exists in whitespace. This can happen with null bytes or
+    // other unusual characters in the source.
+    if (StartColNo > EndColNo)
+      return;
   }
 
   assert(StartColNo <= map.getSourceLine().size() && "Invalid range!");

tools/clang/test/DXC/highlight-range-null-byte.test

@@ -0,0 +1,7 @@
+// Regression test: dxc crashes on HLSL with embedded null bytes due to
+// out-of-bounds access in highlightRange() (TextDiagnostic.cpp).
+
+// RUN: python -c "import sys; sys.stdout.buffer.write(b'f(){a b int4\x00(1}')" > %t.hlsl
+// RUN: not %dxc /T ps_6_0 %t.hlsl 2>&1 | FileCheck %s
+
+// CHECK: error: HLSL requires a type specifier for all declarations