Re: [PATCH 2/4] gdb: remove use of iterate_over_inferiors in mi/mi-interp.c

[Simon Marchi <simark@simark.ca>]

On 2020-01-16 4:01 a.m., Aktemur, Tankut Baris wrote:
>> @@ -140,8 +138,22 @@ mi_interp::init (bool top_level)
>>        /* The initial inferior is created before this function is
>>  	 called, so we need to report it explicitly.  Use iteration in
>>  	 case future version of GDB creates more than one inferior
>> -	 up-front.  */
>> -      iterate_over_inferiors (report_initial_inferior, mi);
>> +	 up-front.
>> +
>> +         This function is called from mi_interpreter_init, and since
> 
> There is an 8-spaces/tab issue above.  And also, the comment piece "This
> function is called from mi_interpreter_init, and" seems obsolete now.

Thanks for spotting these.

That made me realize there are many issues with that comment.

- The part about not being able to use mi_inferior_added because top_level_interpreter_data
  is not set is stale.  That was true before the interps were made into C++ classes.  Back
  then, the mi_interpreter_init function would return the pointer to be used as interpreter
  data, which the caller (the interp_set function) would assign to interp->data.  So it was
  indeed not possible to call mi_inferior_added, because it required interp->data to be set.

- It is still not possible today to use mi_inferior_added, because it iterates over all MI UIs
  and to print the event on all of them.  So if you do:

  1. Start GDB with `-i mi`
  2. Add some inferiors
  3. Create a new MI UI, with `new-ui mi /dev/pts/X`

  When initializing the new MI UI, it will print the inferior-added events on both MI UIs.

- It is necessary to use iteration, not only in case GDB creates more than one inferior
  up front, but also because there might be multiple inferiors when initializing a secondary
  MI UI, as shown above.

I don't know if it is by design, that the =inferior-added events are sent to a new MI UI added
with new-ui, but in any case I don't want to change the behavior with this patch.

Also, when testing this, I realized it was wrong to put the gdb_flush out of the loop.  It's not
a simple flush as in "make sure the data is all sent to the underlying device", it's also what
adds the final \n!  So with the gdb_flush outside the loop, all the events were printed on the same
line.  I've put the terminal setting calls and the gdb_flush back into the loop, to minimize the
changes.

The patch would now look like this:

From cb845a4f28d1761feb463914c13b0a45c97d9974 Mon Sep 17 00:00:00 2001
From: Simon Marchi <simon.marchi@efficios.com>
Date: Wed, 15 Jan 2020 14:12:20 -0500
Subject: [PATCH] gdb: remove use of iterate_over_inferiors in mi/mi-interp.c

Replace it with a range-based for.  I've updated the comment in
mi_interp::init, which was a bit stale.

gdb/ChangeLog:

	* mi/mi-interp.c (report_initial_inferior): Remove.
	(mi_interp::init): Use range-based for to iterate over inferiors.
---
 gdb/mi/mi-interp.c | 49 ++++++++++++++++++++--------------------------
 1 file changed, 21 insertions(+), 28 deletions(-)

diff --git a/gdb/mi/mi-interp.c b/gdb/mi/mi-interp.c
index 2ac4c119961c..e77093cfa282 100644
--- a/gdb/mi/mi-interp.c
+++ b/gdb/mi/mi-interp.c
@@ -91,8 +91,6 @@ static void mi_memory_changed (struct inferior *inf, CORE_ADDR memaddr,
 			       ssize_t len, const bfd_byte *myaddr);
 static void mi_on_sync_execution_done (void);

-static int report_initial_inferior (struct inferior *inf, void *closure);
-
 /* Display the MI prompt.  */

 static void
@@ -137,12 +135,27 @@ mi_interp::init (bool top_level)

   if (top_level)
     {
-      /* The initial inferior is created before this function is
-	 called, so we need to report it explicitly.  Use iteration in
-	 case future version of GDB creates more than one inferior
-	 up-front.  */
-      iterate_over_inferiors (report_initial_inferior, mi);
-    }
+      /* The initial inferior is created before this function is called, so we
+	 need to report it explicitly when initializing the top-level MI
+	 interpreter.
+
+	 This is also called when additional MI interpreters are added (using
+	 the new-ui command), when multiple inferiors possibly exist, so we need
+	 to use iteration to report all the inferiors.  mi_inferior_added can't
+	 be used, because it would print the event on all the other MI UIs.  */
+
+      for (inferior *inf : all_inferiors ())
+	{
+	  target_terminal::scoped_restore_terminal_state term_state;
+	  target_terminal::ours_for_output ();
+
+	  fprintf_unfiltered (mi->event_channel,
+			      "thread-group-added,id=\"i%d\"",
+			      inf->num);
+
+	  gdb_flush (mi->event_channel);
+	}
+  }
 }

 void
@@ -1253,26 +1266,6 @@ mi_user_selected_context_changed (user_selected_what selection)
     }
 }

-static int
-report_initial_inferior (struct inferior *inf, void *closure)
-{
-  /* This function is called from mi_interpreter_init, and since
-     mi_inferior_added assumes that inferior is fully initialized
-     and top_level_interpreter_data is set, we cannot call
-     it here.  */
-  struct mi_interp *mi = (struct mi_interp *) closure;
-
-  target_terminal::scoped_restore_terminal_state term_state;
-  target_terminal::ours_for_output ();
-
-  fprintf_unfiltered (mi->event_channel,
-		      "thread-group-added,id=\"i%d\"",
-		      inf->num);
-  gdb_flush (mi->event_channel);
-
-  return 0;
-}
-
 ui_out *
 mi_interp::interp_ui_out ()
 {
-- 
2.24.1