[Mlir-commits] [mlir] [mlir] Expand error message to include unregistered dialects. (PR #158028)

[llvmlistbot at llvm.org]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-mlir-core

Author: Jacques Pienaar (jpienaar)

<details>
<summary>Changes</summary>

It is possible to load unregistered dialects, this can result in a confusing error message. Mark unregistered but loaded dialects.

This is currently done as a merged list as that is most concise but requires some additional preprocessing (did merge sort given the other two lists are, could do it shorter and probably at not too much extra cost if I just used SetVectors - so alternative which uses less code and may be preferred as performance not critical here).

---
Full diff: https://github.com/llvm/llvm-project/pull/158028.diff


1 Files Affected:

- (modified) mlir/lib/AsmParser/Parser.cpp (+39-3) 


``````````diff
diff --git a/mlir/lib/AsmParser/Parser.cpp b/mlir/lib/AsmParser/Parser.cpp
index 435ff713a1b29..86ffe41092547 100644
--- a/mlir/lib/AsmParser/Parser.cpp
+++ b/mlir/lib/AsmParser/Parser.cpp
@@ -2077,9 +2077,45 @@ OperationParser::parseCustomOperation(ArrayRef<ResultRecord> resultIDs) {
         diag << " (tried '" << opName << "' as well)";
       auto &note = diag.attachNote();
       note << "Registered dialects: ";
-      llvm::interleaveComma(getContext()->getAvailableDialects(), note,
-                            [&](StringRef dialect) { note << dialect; });
-      note << " ; for more info on dialect registration see "
+      std::vector<StringRef> registered = getContext()->getAvailableDialects();
+      auto loaded = getContext()->getLoadedDialects();
+
+      // Merge the sorted lists of registered and loaded dialects.
+      SmallVector<std::pair<StringRef, bool>> mergedDialects;
+      auto regIt = registered.begin(), regEnd = registered.end();
+      auto loadIt = loaded.rbegin(), loadEnd = loaded.rend();
+      bool isRegistered = false;
+      bool isOnlyLoaded = true;
+      while (regIt != regEnd && loadIt != loadEnd) {
+        StringRef reg = *regIt;
+        StringRef load = (*loadIt)->getNamespace();
+        if (reg < load) {
+          mergedDialects.emplace_back(*regIt++, isRegistered);
+        } else if (load < reg) {
+          mergedDialects.emplace_back(load, isOnlyLoaded);
+          loadIt++;
+        } else {
+          mergedDialects.emplace_back(*regIt++, isRegistered);
+          loadIt++;
+        }
+      }
+      for (; regIt != regEnd; ++regIt)
+        mergedDialects.emplace_back(*regIt, isRegistered);
+      for (; loadIt != loadEnd; ++loadIt)
+        mergedDialects.emplace_back((*loadIt)->getNamespace(), isOnlyLoaded);
+
+      bool loadedUnregistered = false;
+      llvm::interleaveComma(mergedDialects, note, [&](auto &pair) {
+        note << pair.first;
+        if (pair.second) {
+          loadedUnregistered = true;
+          note << " (*)";
+        }
+      });
+      note << " ";
+      if (loadedUnregistered)
+        note << "(* corresponding to loaded but unregistered dialects)";
+      note << "; for more info on dialect registration see "
               "https://mlir.llvm.org/getting_started/Faq/"
               "#registered-loaded-dependent-whats-up-with-dialects-management";
       return nullptr;

``````````

</details>


https://github.com/llvm/llvm-project/pull/158028