[Mlir-commits] [mlir] [MLIR][LLVM] Support Recursive DITypes (PR #80251)

[Billy Zhu]

================
@@ -289,7 +316,19 @@ DINodeAttr DebugImporter::translate(llvm::DINode *node) {
     return nullptr;
   };
   if (DINodeAttr attr = translateNode(node)) {
-    nodeToAttr.insert({node, attr});
+    // If this node was marked as recursive, set its recId.
+    if (auto recType = dyn_cast<DIRecursiveTypeAttrInterface>(attr)) {
+      if (DistinctAttr recId = translationStack.lookup(node)) {
+        attr = cast<DINodeAttr>(recType.withRecId(recId));
+        // Remove the unbound recursive ID from the set of unbound self
+        // references in the translation stack.
+        unboundRecursiveSelfRefs.back().erase(recId);
+      }
+    }
+
+    // Only cache fully self-contained nodes.
+    if (unboundRecursiveSelfRefs.back().empty())
----------------
zyx-billy wrote:

Oh thank you so much for the example. I see now that caching on the set of unbound self-refs doesn't solve the compile time problem... It probably does make sense to instead just create a new cache for each level of the translation stack like you said. Every time we put something that is potentially recursive onto the stack, we start a new cache. And one trick we can play is every time we finish a recursive type, we go into the current layer cache, replace all instances of its rec-self with the just-finished recursive type, and insert into the outer layer cache. This way `translate` never walks into a recursive type twice. Let me try to make something like this that's efficient 🤔 

https://github.com/llvm/llvm-project/pull/80251