Re: docs: confusing gimple "Statement and operand traversals"

On Mon, Aug 24, 2026 at 4:46 AM Georg-Johann Lay via Gcc
<[email protected]> wrote:
>
> https://gcc.gnu.org/onlinedocs/gccint/Statement-and-operand-traversals.html
>
> has:
>
> tree walk_gimple_op (gimple stmt, walk_tree_fn callback_op, struct
> walk_stmt_info *wi)
>
> but according to gimple-walk.h, the 1st arg is of type gimple*.

Looks like the documentation was not updated after
r6-3402-g355fe0884b8097 which changed the name of the struct from
gimple_statement_base to gimple and then removed the typedef and added
the `*` everywhere.

>
> Moreover:
>
> tree walk_gimple_seq (gimple_seq seq, walk_stmt_fn callback_stmt,
> walk_tree_fn callback_op, struct walk_stmt_info *wi)
>
> This function walks all the statements in the sequence SEQ calling
> walk_gimple_stmt on each one. [...] If walk_gimple_stmt returns non-
> NULL, the walk is stopped and the value returned. [...]
>
> - According to gimple-walk.h, the return value of walk_gimple_seq
>    is gimple* and not tree.
>
> - The return value of walk_gimple_stmt is of type tree, and it is
>    unclear how that is mapped to the return value of walk_gimple_seq
>    which is gimple*.

And it looks like the documentation there was not updated way back in
2008 by r0-88576-g726a989a8b74bf for the change to tuples.

So in this case walk_stmt_info's callback_result will contain the
return value of walk_gimple_stmt.
But the return value is either the gimple stmt where the non-null
walk_gimple_stmt returned a non-null result or NULL if
walk_stmt_info's removed_stmt was set (removed_stmt will be unset
still).
callback_result will be non null in the case of the removed_stmt case.

>
> Johann
>