[TableGen] Emit more helpful error messages on empty type set

[arichardson]

Summary

I have seen this error quite frequently in our out-of-tree CHERI backends
and the lack of location information sometimes makes it quite difficult
to track down the actual source of the error.
This patch changes the llvm_unreachable() to a PrintFatalError() so that
tablegen prints a stack of source locations.

[jrtc27]

The fact that this is under NDEBUG at all seems rather dubious

[craig.topper]

Seems reasonable to me. Whether we want to move that out of NDEBUG is a different issue. We get a lot of compile time complaints about tablegen in release builds.

[foad]

In D99468#2654612, @jrtc27 wrote:

The fact that this is under NDEBUG at all seems rather dubious

See D147127.

llvm/utils/TableGen/CodeGenDAGPatterns.cpp

 #ifndef NDEBUG
 TypeInfer::ValidateOnExit::~ValidateOnExit() {
   if (Infer.Validate && !VTS.validate()) {
     dbgs() << "Type set is empty for each HW mode:\n"
               "possible type contradiction in the pattern below "
               "(use -print-records with llvm-tblgen to see all "
               "expanded records).\n";
     Infer.TP.dump();
-    llvm_unreachable(nullptr);
+    dbgs() << "Generated from record:\n";
+    Infer.TP.getRecord()->dump();
+    PrintFatalError(Infer.TP.getRecord()->getLoc(),
+                    "Type set is empty for each HW mode in '" +
+                        Infer.TP.getRecord()->getName() + "'");
   }
 }
 #endif
 
 
 //===----------------------------------------------------------------------===//
 // ScopedName Implementation
 //===----------------------------------------------------------------------===//