Re: [PATCH v2 3/3] Replace home-grown linked-lists in FreeBSD's native target with STL lists.

[Simon Marchi <simon.marchi@polymtl.ca>]

On 2017-08-09 21:15, John Baldwin wrote:
> On Wednesday, August 09, 2017 08:16:29 PM Simon Marchi wrote:
>> On 2017-08-09 19:47, John Baldwin wrote:
>> > diff --git a/gdb/fbsd-nat.c b/gdb/fbsd-nat.c
>> > index 721e72b85c..c89343a24f 100644
>> > --- a/gdb/fbsd-nat.c
>> > +++ b/gdb/fbsd-nat.c
>> > @@ -41,6 +41,8 @@
>> >  #include "elf-bfd.h"
>> >  #include "fbsd-nat.h"
>> >
>> > +#include <list>
>> > +
>> >  /* Return the name of a file that can be opened to get the symbols for
>> >     the child process identified by PID.  */
>> >
>> > @@ -712,13 +714,7 @@ fbsd_update_thread_list (struct target_ops *ops)
>> >    sake.  FreeBSD versions newer than 9.1 contain both fixes.
>> >  */
>> >
>> > -struct fbsd_fork_info
>> > -{
>> > -  struct fbsd_fork_info *next;
>> > -  ptid_t ptid;
>> > -};
>> > -
>> > -static struct fbsd_fork_info *fbsd_pending_children;
>> > +static std::list<ptid_t> fbsd_pending_children;
>> 
>> Can't this be a forward_list as well?
> 
> Not trivially.  fbsd_is_child_pending has to walk the list looking for 
> a
> specific PID and remove that specific list entry (not always the first
> entry).  Right now this is using the following loop:
> 
>    for (auto it = fbsd_pending_children.begin ();
>        it != fbsd_pending_children.end (); it++)
>      if (it->pid () == pid)
>        {
>          ptid_t ptid = *it;
>          fbsd_pending_children.erase (it);
>          return ptid;
>        }
>    return null_ptid;
> 
> I'm not sure if it's legal to do something like this for a 
> forward_list:
> 
>    for (auto it = fbsd_pending_children.before_begin ();
>         it + 1 != fbsd_pending_children.end (); it++)
>     if ((it + 1)->pid () == pid)
>        {
>          ptid_t ptid = *(it + 1);
>          fbsd_pending_childern.erase_after (it);
>          return ptid;
>        }
>    return null_ptid;
> 
> Even if it is legal, I'm not sure it is more readable.  These lists 
> should
> generally be quite small, so I think readability is more important than
> optimization in this case.

Ah, ok.  I thought it could be done trivially, but I can't easily 
build-test on FreeBSD right now.

I agree with you.

Simon