[PATCH][gdb] Don't set initial language using previous language

[PATCH][gdb] Don't set initial language using previous language

[Tom de Vries]

Hi,

When language is set to auto, part of loading an executable is to update the
language accordingly.  This is implemented by set_initial_language.

In case of a c++ executable without DW_AT_main_subprogram,
set_initial_language finds "main" in the minimal symbols, and does a lookup of
"main" in the symbol tables to determine the language of the symbol, and uses
that as initial language.

The symbol lookup is done using lookup_symbol which is a wrapper around
lookup_symbol_in_language, using the current language.

So, consider two c++ executables a.out and b.out, which we'll load one after
another.  If we track the resulting lookup_symbol_in_language calls:
...
$ gdb -batch \
    -ex "b lookup_symbol_in_language" \
    -ex r -ex c -ex c \
    --args gdb
...
we find that indeed lookup_symbol_in_language is called once using language_c, and
once using language_c_plus:
...
(gdb) file a.out
Reading symbols from a.out...

Breakpoint 1, lookup_symbol_in_language (name=0x5555568c2050 "main", \
  block=0x0, domain=VAR_DOMAIN, lang=language_c, is_a_field_of_this=0x0) \
  at ../../gdb/symtab.c:1905
1905    {
(gdb) file b.out
Load new symbol table from "b.out"? (y or n) y
Reading symbols from b.out...

Breakpoint 1, lookup_symbol_in_language (name=0x5555568c2030 "main", \
  block=0x0, domain=VAR_DOMAIN, lang=language_cplus, is_a_field_of_this=0x0) \
  at ../../gdb/symtab.c:1905
1905    {
(gdb)
...

It seems like a bad idea to have the previous language play a role
in determining the executable language.

Fix this by using lookup_symbol_in_language in set_initial_language with the
default language c as argument.

Tested on x86_64-linux.

OK for trunk?

Thanks,
- Tom

[gdb] Don't set initial language using previous language

gdb/ChangeLog:

2020-02-28  Tom de Vries  <tdevries@suse.de>

	* symfile.c (set_initial_language): Use default language for lookup.

---
 gdb/symfile.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/gdb/symfile.c b/gdb/symfile.c
index f1edf2dca5..01c3f5af12 100644
--- a/gdb/symfile.c
+++ b/gdb/symfile.c
@@ -1684,11 +1684,15 @@ set_initial_language (void)
   if (language_mode == language_mode_manual)
     return;
   enum language lang = main_language ();
+  /* Make C the default language.  */
+  enum language default_lang = language_c;
 
   if (lang == language_unknown)
     {
       const char *name = main_name ();
-      struct symbol *sym = lookup_symbol (name, NULL, VAR_DOMAIN, NULL).symbol;
+      struct symbol *sym
+	= lookup_symbol_in_language (name, NULL, VAR_DOMAIN, default_lang,
+				     NULL).symbol;
 
       if (sym != NULL)
 	lang = sym->language ();
@@ -1696,8 +1700,7 @@ set_initial_language (void)
 
   if (lang == language_unknown)
     {
-      /* Make C the default language */
-      lang = language_c;
+      lang = default_lang;
     }
 
   set_language (lang);

Re: [PATCH][gdb] Don't set initial language using previous language

[Tom Tromey]

>>>>> "Tom" == Tom de Vries <tdevries@suse.de> writes:

Tom> It seems like a bad idea to have the previous language play a role
Tom> in determining the executable language.

Does it make a difference in practice?

Tom> Fix this by using lookup_symbol_in_language in set_initial_language with the
Tom> default language c as argument.

Tom> Tested on x86_64-linux.

Tom> OK for trunk?

Yes, it seems fine to me.  Thank you.

Tom

Re: [PATCH][gdb] Don't set initial language using previous language

[Tom de Vries]

On 28-02-2020 16:04, Tom Tromey wrote:
>>>>>> "Tom" == Tom de Vries <tdevries@suse.de> writes:
> 
> Tom> It seems like a bad idea to have the previous language play a role
> Tom> in determining the executable language.
> 
> Does it make a difference in practice?
> 

I don't know of a concrete case where it does.

> Tom> Fix this by using lookup_symbol_in_language in set_initial_language with the
> Tom> default language c as argument.
> 
> Tom> Tested on x86_64-linux.
> 
> Tom> OK for trunk?
> 
> Yes, it seems fine to me.  Thank you.

Ack, committed.

Thanks,
- Tom