Re: [PATCH 02/10] Return gdbpy_ref<> from symbol_to_symbol_object

[Simon Marchi <simark@simark.ca>]

On 2026-02-20 16:03, Tom Tromey wrote:
> @@ -622,9 +614,10 @@ gdbpy_lookup_static_symbols (PyObject *self, PyObject *args, PyObject *kw)
>  
>  		  if (symbol != nullptr)
>  		    {
> -		      PyObject *sym_obj = symbol_to_symbol_object (symbol);
> +		      gdbpy_ref<> sym_obj = symbol_to_symbol_object (symbol);
>  		      if (sym_obj == nullptr
> -			  || PyList_Append (return_list.get (), sym_obj) == -1)
> +			  || PyList_Append (return_list.get (),
> +					    sym_obj.get ()) == -1)

I asked Claude to review your comments, and it said:

    Bonus finding: The gdbpy_lookup_static_symbols function in
    py-symbol.c had a pre-existing reference leak — the old code passed
    the raw PyObject * from symbol_to_symbol_object directly to
    PyList_Append (which does its own incref) without decrementing.  The
    new code stores it in a gdbpy_ref<> and passes .get(), fixing the
    leak.

It seems true.  It's really hard to remember which functions steal
references and which functions get their own new reference.  Good job on
your part to call .get() here and not .release().  I wish we had a C++
layer around the Python API to make that clear through the type
system...

Simon