[Lldb-commits] [lldb] [lldb] Fix dwim-print to not delete non-result persistent variables (PR #85152)

[Dave Lee via lldb-commits]

================
@@ -170,6 +170,14 @@ void CommandObjectDWIMPrint::DoExecute(StringRef command,
     ExpressionResults expr_result = target.EvaluateExpression(
         expr, exe_scope, valobj_sp, eval_options, &fixed_expression);
 
+    auto persistent_name = valobj_sp->GetName();
+    // EvaluateExpression doesn't generate a new persistent result (`$0`) when
+    // the expression is already just a persistent variable (`$var`). Instead,
+    // the same persistent variable is reused. Take note of when a persistent
+    // result is created, to prevent unintentional deletion of a user's
+    // persistent variable.
+    bool did_persist_result = persistent_name != expr;
----------------
kastiglione wrote:

I agree that it's not ideal, but I felt it was sufficient.

> what you want to test, which is whether the result of the expression was a new expression result variable

exactly. However I think peeking at the result variables is also not ideal. Both my initial stab at this, and your suggestion are deducing whether a new expression result variable was created. If I'm to change the API, I think it'd be ideal to make the API communicate have means to communicate this explicitly.

There's another approach I considered taking. Before calling `EvaluateExpression`, try the expression as a persistent variable and check in the target to see if it exists, and if it does exist, use it (and avoid `EvaluateExpression` altogether). This is similar to how expressions are first treated as frame variables (and as @felipepiovezan has requested, we should try `target variable` too).

https://github.com/llvm/llvm-project/pull/85152