[lldb/docs] Fix/improve the gdb command map for dynamic types

[labath]

The setting and option value names were wrong. I'm assuming this changed over time, but I haven't tried to figure out when.

[llvmbot]

@llvm/pr-subscribers-lldb

Author: Pavel Labath (labath)

Changes

The setting and option value names were wrong. I'm assuming this changed over time, but I haven't tried to figure out when.

---

Full diff: https://github.com/llvm/llvm-project/pull/138538.diff

1 Files Affected:

• (modified) lldb/docs/use/map.rst (+12-4)

diff --git a/lldb/docs/use/map.rst b/lldb/docs/use/map.rst
index ed285b2d1f6e9..c648b212006e0 100644
--- a/lldb/docs/use/map.rst
+++ b/lldb/docs/use/map.rst
@@ -800,16 +800,24 @@ Print the dynamic type of the result of an expression
   (gdb) p someCPPObjectPtrOrReference
   (Only works for C++ objects)
 
+LLDB does this automatically if determining the dynamic type does not require
+running the target (in C++, running the target is never needed). This default is
+controlled by the `target.prefer-dynamic-value` setting. If that is disabled, it
+can be re-enabled on a per-command basis:
+
 .. code-block:: shell
 
-  (lldb) expr -d 1 -- [SomeClass returnAnObject]
-  (lldb) expr -d 1 -- someCPPObjectPtrOrReference
+  (lldb) settings set target.prefer-dynamic-value no-dynamic-values
+  (lldb) frame variable -d no-run-target someCPPObjectPtrOrReference
+  (lldb) expr -d no-run-target -- someCPPObjectPtr
 
-or set dynamic type printing to be the default:
+Note that printing of the dynamic type of references is not possible with the
+`expr` command. The workaround is to take the address of the reference and
+instruct lldb to print the children of the resulting pointer.
 
 .. code-block:: shell
 
-  (lldb) settings set target.prefer-dynamic run-target
+  (lldb) expr -P1 -d no-run-target -- &someCPPObjectReference
 
 Call a function so you can stop at a breakpoint in it
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

[labath]

I spent some time trying to make this work, but I came to the conclusion that this would require a relatively big change in how the expression evaluator works. Because we're creating persistent copies of the expression result, it's not just a matter of getting something to recognise the dynamic type of the result -- we would actually need to determine the dynamic type of the object before we create the copy (so that we copy the entire object) -- and then I guess somehow reset the original (non-dynamic) ValueObject to point to the subobject of the result. And that might be too much work to put on the shoulders of the "use-dynamic" setting?

[labath]

(If you're wondering why we're making a copy of a reference -- but not of a pointer -- I suspect that's because in the expression evaluator we do not differentiate between an object and a reference to it. I suppose we could do it, and then somehow treat references as pointers, but I don't exactly know what would that entail.)

[Michael137]

Makes sense to me