Use thread_info and inferior pointers more throughout

From: Pedro Alves <palves@redhat.com>

  On 06/07/2018 07:28 PM, Tom Tromey wrote:
>>>>>> "Pedro" == Pedro Alves <palves@redhat.com> writes:
> 
> Pedro> In many cases though, we already have the thread_info or inferior
> Pedro> pointer handy, but we "lose" it somewhere along the call stack, only
> Pedro> to look it up again by ptid_t/pid.  Since thread_info or inferior
> Pedro> objects know their parent target, if we pass around thread_info or
> Pedro> inferior pointers when possible, we avoid having to add extra
> Pedro> target_ops parameters to many functions, and also, we eliminate a
> Pedro> number of by ptid_t/int lookups.
> 
> This seems like a good idea.  I've sometimes wondered why ptid was used
> rather than object pointers; but now it seems that there was no deep
> reason :)

I think it's just a consequence of how thread support grew in gdb.
In the really old days, gdb did not support debugging multiple threads at all,
because threads weren't even a thing, and the inferior process was just identified
by its integer PID.  Later, when support for threads was added, threads were bolted
on as kind of a second-class citizen, with gdb context-switching
a bunch of globals, filling them with the context of the now-selected thread.
See e.g.:

  https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;a=blob;f=gdb/thread.c;hb=ca67fcb81a300f8e2e58246cc64d971514e95d5a#l295

ptid_t was introduced to replace a single integer PIDs, and I think before
that for a period the LWP/TID info was crammed into that single
integer PID with a shifting and masking, even.

When non-stop support was introduced, gdb's internal threading
model was changed to "always a thread", see e.g.:

  https://sourceware.org/ml/gdb-patches/2008-08/msg00169.html

So before _that_, we really had to use ptid instead of object pointers,
because when debugging a single threaded process, there was really
no thread object to point to.

After that, the many run-control-related globals were eliminated, in favor
of thread_info fields, and that funny context-switching was
mostly eliminated, see e.g.,:

 https://sourceware.org/ml/gdb-patches/2008-08/msg00441.html

Other time more globals have been eliminated since that, but we're
still left with a few.  In fact, I'm about to send a patch that
eliminates another one (stop_pc).  :-)

That was a fun trip down the memory lane.  :-)

> 
> Pedro> - Related, there's a spot where using a RAII object to handle the
> Pedro>   refcount would be handy, so a new scoped_inc_dec_ref class is added
> Pedro>   to common/refcounted-object.h.
> 
> It seems to me that this could be done via gdb::ref_ptr plus a simple
> policy class that interfaces it to refcounted_object.

Yeah, good idea -- I think that is likely to come in handy in other
spots too, so I've done it.

> 
> Pedro> +/* A RAII type that increments/decrements the refcount of an object on
> Pedro> +   enter/exit of a scope.  */
> Pedro> +
> Pedro> +class scoped_inc_dec_ref
> Pedro> +{
> Pedro> +public:
> Pedro> +  explicit scoped_inc_dec_ref (refcounted_object *obj)
> Pedro> +    : m_obj (obj)
> Pedro> +  {
> Pedro> +    m_obj->incref ();
> Pedro> +  }
> Pedro> +
> Pedro> +  ~scoped_inc_dec_ref ()
> Pedro> +  {
> Pedro> +    m_obj->decref ();
> Pedro> +  }
> 
> ... if you do keep this it probably needs DISABLE_COPY_AND_ASSIGN.
> 

You're right.  I removed it instead though.

I'm squashing this into the patch that I'm merging.

(Also noticed that there was a pair of incref/decref in
run_inferior_call that is not necessary.)

From ccf4c1b904f0c7b8ee339c7fb14af5f857c305bb Mon Sep 17 00:00:00 2001
From: Pedro Alves <palves@redhat.com>
Date: Fri, 15 Jun 2018 22:51:49 +0100
Subject: [PATCH] thread_info_ref

---
 gdb/common/refcounted-object.h | 19 +++++++------------
 gdb/gdbthread.h                |  5 +++++
 gdb/infcall.c                  | 28 ++++++++++++----------------
 gdb/thread.c                   |  8 ++++----
 4 files changed, 28 insertions(+), 32 deletions(-)

Use thread_info and inferior pointers more throughout

Commit Message

Patch