Re: [PATCH 3/3] Cache the result of find_file_and_directory

[Simon Marchi via Gdb-patches <gdb-patches@sourceware.org>]

On 2021-12-05 15:17, Tom Tromey wrote:
>>> Sorry about that.  I'll fix it soon.  Or you can back it out if you
>>> prefer.
> 
> Simon> I'm not personally affected so I do not mind, I can wait for the fix.
> 
> Here is what I am checking in.
> I was able to reproduce the failure with valgrind, and confirmed that
> this patch fixes it.
> 
> Tom

Thanks.  And while looking at that fix, I had this this question:
looking at find_file_and_directory:

    static file_and_directory &
    find_file_and_directory (struct die_info *die, struct dwarf2_cu *cu)
    {
      if (cu->per_cu->fnd != nullptr)
        return *cu->per_cu->fnd;

      /* Find the filename.  Do not use dwarf2_name here, since the filename
         is not a source language identifier.  */
      file_and_directory res (dwarf2_string_attr (die, DW_AT_name, cu),
    			  dwarf2_string_attr (die, DW_AT_comp_dir, cu));

      if (res.get_comp_dir () == nullptr
          && producer_is_gcc_lt_4_3 (cu)
          && res.get_name () != nullptr
          && IS_ABSOLUTE_PATH (res.get_name ()))
        res.set_comp_dir (ldirname (res.get_name ()));

      cu->per_cu->fnd.reset (new file_and_directory (std::move (res)));
      return *cu->per_cu->fnd;
    }

`dwarf2_string_attr (die, DW_AT_name, cu)` is used as the name in
`fnd`, and `fnd` is saved in the dwarf2_per_cu_data structure.  Is
`dwarf2_string_attr (die, DW_AT_name, cu)` value guaranteed to live at
least as long as the dwarf2_per_cu_data structure?  IIUC, the die_info
objects are tied to a dwarf2_cu.  And dwarf2_cus can be deleted
(by free_cached_comp_units).  I might be wrong, but I thought I'd
raise the point just in case.

Simon