[Lldb-commits] [lldb] [lldb-dap] Fix global-scope evaluate always routing to command interpreter (PR #225243)

[Nathan Fusselman via lldb-commits]

================
@@ -225,14 +230,22 @@ def assert_eval_fails(expression: str):
         session.verify_evaluate(eval_body, matches="20")
 
         if context_parses_expressions:
-            # Access global variable without a frame
-            # Run in variable mode to avoid interpreting it as a command.
-            session.evaluate("`lldb-dap repl-mode variable", context="repl")
-
+            # Access a global variable with no frame at all (frameId omitted).
             eval_body = session.evaluate("static_int", context=context)
             session.verify_evaluate(eval_body, matches="42", type="int")
 
-            session.evaluate("`lldb-dap repl-mode auto", context="repl")
+            assert_eval_fails_frameless("var1")  # local, not global.
+            assert_eval_fails_frameless("totally_bogus_expr_xyz")
+
+            if context in ("repl", None):
+                # `help` shares its name with a real lldb command (see main.cpp);
+                # it should resolve to the global, with an ambiguity warning.
+                eval_body = session.evaluate("help", context=context)
+                session.verify_evaluate(eval_body, matches="99", type="int")
+
+                # `version` has no matching global; still runs as a command frameless.
+                eval_body = session.evaluate("version", context=context)
+                session.verify_evaluate(eval_body, matches=r"^lldb[ -]\d", has_mem_ref=False)
----------------
nathanfusselman wrote:

Even better, I will just remove the usage of `version` here, it was a convenient one-word command but I can create a deterministic one that is not platform-dependent.

https://github.com/llvm/llvm-project/pull/225243