Re: [PATCH] gdb: search local symbols before global symbols in completion

[Oleg Tolmatcev <oleg.tolmatcev@gmail.com>]

пн, 3 авг. 2026 г. в 13:03, Andrew Burgess <aburgess@redhat.com>:
>
> Oleg Tolmatcev <oleg.tolmatcev@gmail.com> writes:
>
> > When symbol completion is performed from a selected frame, search the
> > local blocks before scanning minimal symbols and global/static symbols.
> > This makes completion prefer names that are visible in the current
> > context, including local variables.
> > ---
> >  gdb/symtab.c | 72 ++++++++++++++++++++++++++--------------------------
>
> It feels like this really needs a test in the gdb.dap/ directory.
>
> Thanks,
> Andrew
>
>
> >  1 file changed, 36 insertions(+), 36 deletions(-)
> >
> > diff --git a/gdb/symtab.c b/gdb/symtab.c
> > index 5d5076f2e77..85665bebdf4 100644
> > --- a/gdb/symtab.c
> > +++ b/gdb/symtab.c
> > @@ -5931,6 +5931,42 @@ default_collect_symbol_completion_matches_break_on
> >
> >    lookup_name_info lookup_name (sym_text, name_match_type, true);
> >
> > +  /* Search upwards from currently selected frame (so that we can
> > +     complete on local vars).  Also catch fields of types defined in
> > +     this places which match our text string.  Only complete on types
> > +     visible from current context.  */
> > +
> > +  b = get_selected_block ();
> > +  surrounding_static_block = b == nullptr ? nullptr : b->static_block ();
> > +  surrounding_global_block = b == nullptr ? nullptr : b->global_block ();
> > +  if (surrounding_static_block != NULL)
> > +    while (b != surrounding_static_block)
> > +      {
> > +     QUIT;
> > +
> > +     for (struct symbol *sym : block_iterator_range (b))
> > +       {
> > +         if (code == TYPE_CODE_UNDEF)
> > +           {
> > +             completion_list_add_symbol (tracker, sym, lookup_name,
> > +                                         sym_text, word);
> > +             completion_list_add_fields (tracker, sym, lookup_name,
> > +                                         sym_text, word);
> > +           }
> > +         else if (sym->domain () == STRUCT_DOMAIN
> > +                  && sym->type ()->code () == code)
> > +           completion_list_add_symbol (tracker, sym, lookup_name,
> > +                                       sym_text, word);
> > +       }
> > +
> > +     /* Stop when we encounter an enclosing function.  Do not stop for
> > +        non-inlined functions - the locals of the enclosing function
> > +        are in scope for a nested function.  */
> > +     if (b->function () != NULL && b->inlined_p ())
> > +       break;
> > +     b = b->superblock ();
> > +      }
> > +
> >    /* At this point scan through the misc symbol vectors and add each
> >       symbol you find to the list.  Eventually we want to ignore
> >       anything that isn't a text symbol (everything else will be
> > @@ -5974,42 +6010,6 @@ default_collect_symbol_completion_matches_break_on
> >        SEARCH_ALL_DOMAINS);
> >      }
> >
> > -  /* Search upwards from currently selected frame (so that we can
> > -     complete on local vars).  Also catch fields of types defined in
> > -     this places which match our text string.  Only complete on types
> > -     visible from current context.  */
> > -
> > -  b = get_selected_block ();
> > -  surrounding_static_block = b == nullptr ? nullptr : b->static_block ();
> > -  surrounding_global_block = b == nullptr ? nullptr : b->global_block ();
> > -  if (surrounding_static_block != NULL)
> > -    while (b != surrounding_static_block)
> > -      {
> > -     QUIT;
> > -
> > -     for (struct symbol *sym : block_iterator_range (b))
> > -       {
> > -         if (code == TYPE_CODE_UNDEF)
> > -           {
> > -             completion_list_add_symbol (tracker, sym, lookup_name,
> > -                                         sym_text, word);
> > -             completion_list_add_fields (tracker, sym, lookup_name,
> > -                                         sym_text, word);
> > -           }
> > -         else if (sym->domain () == STRUCT_DOMAIN
> > -                  && sym->type ()->code () == code)
> > -           completion_list_add_symbol (tracker, sym, lookup_name,
> > -                                       sym_text, word);
> > -       }
> > -
> > -     /* Stop when we encounter an enclosing function.  Do not stop for
> > -        non-inlined functions - the locals of the enclosing function
> > -        are in scope for a nested function.  */
> > -     if (b->function () != NULL && b->inlined_p ())
> > -       break;
> > -     b = b->superblock ();
> > -      }
> > -
> >    /* Add fields from the file's types; symbols will be added below.  */
> >
> >    if (code == TYPE_CODE_UNDEF)
> > --
> > 2.55.0.windows.1
>

I tried 3 times, but every time my test causes regressions on ARM. I
don't know how to write a passing test.

Is a test really necessary for this trivial patch that only changes the
order of completion results?

Oleg