[MLIR][doc] Clarify `allowUnregisteredDialects()` as "testing option" (NFC)

[jpienaar]

mlir/include/mlir/IR/MLIRContext.h

139

convenient during testing ? (testing way sounds awkard).

140

I don't know if I follow the "and ..." part. I'd just stop before then. E.g., MLIR here is opaque and I think rather refers to passes/lowerings (where these should be handled conservatively but be supported really) but you are more likely to run into unmet expectations/surprises wrt conservative behavior. But it does't seem to be specific to MLIR for me there.

[mehdi_amini]

mlir/include/mlir/IR/MLIRContext.h

140

I'm trying to address "crashes" and other stuff around our "system" invariants, like https://github.com/llvm/llvm-project/issues/60569

[mehdi_amini]

Address comments

[mehdi_amini]

mlir/include/mlir/IR/MLIRContext.h

140

I made it a bit more explicit and pushed, but I'll adjust if you have other suggestions!

mlir/include/mlir/IR/MLIRContext.h

public:
   /// If no dialect is loaded for this name and none is available in the
   /// registry, returns nullptr.
   Dialect *getOrLoadDialect(StringRef name);
 
   /// Return true if we allow to create operation for unregistered dialects.
   bool allowsUnregisteredDialects();
 
   /// Enables creating operations in unregistered dialects.
+  /// This option is **heavily discouraged**: it is convenient during testing
+  /// but it is not a good practice to use it in production code. Some system
+  /// invariants can be broken (like loading a dialect after creating
+  ///  operations) without being caught by assertions or other means.
   void allowUnregisteredDialects(bool allow = true);
 
   /// Return true if multi-threading is enabled by the context.
   bool isMultithreadingEnabled();
 
   /// Set the flag specifying if multi-threading is disabled by the context.
   /// The command line debugging flag `--mlir-disable-threading` is overriding
   /// this call and making it a no-op!