[PATCH] D62780: msabi: Fix exponential mangling time for even more contrived inputs

[Nico Weber via Phabricator via cfe-commits]

thakis marked an inline comment as done.
thakis added inline comments.


================
Comment at: clang/lib/AST/MicrosoftMangle.cpp:819-823
     ArgBackRefMap::iterator Found = TemplateArgBackReferences.find(ND);
     if (Found == TemplateArgBackReferences.end()) {
-      // Mangle full template name into temporary buffer.
-      llvm::SmallString<64> TemplateMangling;
-      llvm::raw_svector_ostream Stream(TemplateMangling);
-      MicrosoftCXXNameMangler Extra(Context, Stream);
-      Extra.mangleTemplateInstantiationName(TD, *TemplateArgs);
-
-      // Use the string backref vector to possibly get a back reference.
-      mangleSourceName(TemplateMangling);
-
-      // Memoize back reference for this type.
-      BackRefVec::iterator StringFound =
-          llvm::find(NameBackReferences, TemplateMangling);
-      if (StringFound != NameBackReferences.end()) {
-        TemplateArgBackReferences[ND] =
-            StringFound - NameBackReferences.begin();
+
+      TemplateArgStringMap::iterator Found = TemplateArgStrings.find(ND);
+      if (Found == TemplateArgStrings.end()) {
----------------
rnk wrote:
> Here's a thought: could we get by with one map? I think we can be certain that nothing mangles to integer literals between 0 and 9, since otherwise the demangler couldn't distinguish those from back references. So, can we keep the string map, and insert the string "0", "1", etc for back referenced things?
I had wondered too. In the end, I figured the int map needs less memory and is almost always enough, so I went with this approach.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D62780/new/

https://reviews.llvm.org/D62780