Linker --sort-section name is broken

[Sonal Santan]

Hello H.J.

Looks like ld lacks the kind of examples you tried in its testsuite. I 
can think of the following two ways to fix this:

[1] A quick fix would be to change the logic which triggers BST sorting 
in wild() to the following--
if (s->handler_data[0]
    && s->handler_data[0]->spec.sorted == by_name
    && !s->filenames_sorted
    && !handler_data[1])
    {...
[2] A better fix would be grow the handler_data[] array in struct 
lang_wild_statement_struct to size 5. Make the BST logic exclusively use 
handler_data[4] for all its needs.

What do you think?

Sonal

H. J. Lu wrote:
> Hi Sonal,
>
> Your linker patch breaks --sort-section name. In wild (), you hae
>
>   if (s->handler_data[0]
>       && s->handler_data[0]->spec.sorted == by_name
>       && !s->filenames_sorted)
>     {
>       lang_section_bst_type *tree;
>
>       walk_wild (s, output_section_callback_fast, output);
>
>       tree = (lang_section_bst_type *) s->handler_data[1];
>       if (tree)
>         output_section_callback_tree_to_list (s, tree, output);
>       s->handler_data[1] = NULL;
>     }
>
> With "--sort-section name" on
>
> *(.text .text.*)
>
> handler_data[1] is a wildcard spec. But you use it for BST in
> wild_sort_fast. I don't see how it can work. I really don't like
>
> tree = (lang_section_bst_type **) &wild->handler_data[1];
>
> It is hard to check if it is really valid.
>
>
> H.J.
>
>