[4/9] gdb: remove one user of the executable changed observer

Message ID 5e9c1c90c7ba3adf0a7062285ec40617cd0ef14a.1694858967.git.aburgess@redhat.com
State New
Headers
Series Add executable_changed event to Python API |

Checks

Context Check Description
linaro-tcwg-bot/tcwg_gdb_check--master-arm success Testing passed
linaro-tcwg-bot/tcwg_gdb_build--master-aarch64 success Testing passed
linaro-tcwg-bot/tcwg_gdb_build--master-arm success Testing passed
linaro-tcwg-bot/tcwg_gdb_check--master-aarch64 success Testing passed

Commit Message

Andrew Burgess Sept. 16, 2023, 10:18 a.m. UTC
  My goal for the next few commits is to expose the executable_changed
observable from the Python API.

However, there is call to the executable_changed observable in the
reread_symbols function (in symfile.c), and this doesn't actually
correspond to the executable changing.  My idea then, is to remove
this use of the executable_changed observable, but, before I can do
that, I need to check that nothing is going to break, and that
requires my to think about the current users of this observable.

One current user of executable_changed is in symtab.c.  We add an
executable_changed observer that calls:

  set_main_name (nullptr, language_unknown);

to discard all information about the main function when the executable
changes.

However, changing the executable doesn't actually change the debug
information.  The debug information changes when the symbol-file
changes, so I think this observer is in slightly the wrong place.

The new_objfile observable is (unfortunately) overloaded, it is called
when a new objfile is loaded, and also (when its argument is nullptr),
when all debug information should be discarded.

It turns out that there is already a new_objfile observer in
symtab.c.  I propose that, when the argument is nullptr (indicating
all debug info should be discarded), that we should call set_main_name
to discard the information about the main function.  We can then
remove the executable_changed observer from symtab.c.

All tests still pass, and, in my local world, I added some debug
printf calls, and I think we are still discarded the main information
everywhere we need to.
---
 gdb/symtab.c | 18 +++++++-----------
 1 file changed, 7 insertions(+), 11 deletions(-)
  

Patch

diff --git a/gdb/symtab.c b/gdb/symtab.c
index c0c2454d967..2d446331ed1 100644
--- a/gdb/symtab.c
+++ b/gdb/symtab.c
@@ -100,6 +100,8 @@  static struct block_symbol
 			    enum block_enum block_index,
 			    const char *name, const domain_enum domain);
 
+static void set_main_name (const char *name, enum language lang);
+
 /* Type of the data stored on the program space.  */
 
 struct main_info
@@ -1694,6 +1696,11 @@  symtab_new_objfile_observer (struct objfile *objfile)
 {
   /* Ideally we'd use OBJFILE->pspace, but OBJFILE may be NULL.  */
   symbol_cache_flush (current_program_space);
+
+  /* When all objfiles have been removed (OBJFILE is nullptr), then forget
+     everything we know about the main function.  */
+  if (objfile == nullptr)
+    set_main_name (nullptr, language_unknown);
 }
 
 /* This module's 'free_objfile' observer.  */
@@ -6329,15 +6336,6 @@  main_language (void)
   return info->language_of_main;
 }
 
-/* Handle ``executable_changed'' events for the symtab module.  */
-
-static void
-symtab_observer_executable_changed (void)
-{
-  /* NAME_OF_MAIN may no longer be the same, so reset it for now.  */
-  set_main_name (NULL, language_unknown);
-}
-
 /* Return 1 if the supplied producer string matches the ARM RealView
    compiler (armcc).  */
 
@@ -7020,8 +7018,6 @@  the use of prologue scanners."),
 		     class_maintenance, 0, &maintenancelist);
   deprecate_cmd (c, "maintenancelist flush symbol-cache");
 
-  gdb::observers::executable_changed.attach (symtab_observer_executable_changed,
-					     "symtab");
   gdb::observers::new_objfile.attach (symtab_new_objfile_observer, "symtab");
   gdb::observers::free_objfile.attach (symtab_free_objfile_observer, "symtab");
 }