[7/9] Update exec_file_attach to cope with "target:" filenames

  Pedro Alves wrote:
> On 03/20/2015 04:48 PM, Gary Benson wrote:
> > This commit adds support for filenames prefixed	with "target:" to
> > exec_file_attach.  This is required to correctly follow inferior
> > exec* calls when a gdb_sysroot prefixed with "target:" is set.
> 
> Hmm, I don't see how.  Isn't this only true when target_pid_to_exec_file
> prepends the sysroot, which it doesn't yet?  I think this should move
> to that other series.

Search for gdb_sysroot in infrun.c.

> A couple bits could use more explanation (in commit log and/or
> comments):
> 
>  - Why is writing into executable files not supported with "target:" ?
>  - The skipping of gdb_realpath_keepfile.

Are these two additions ok?


Cheers,
Gary

--
http://gbenson.net/
  

On 04/01/2015 02:55 PM, Gary Benson wrote:
> Pedro Alves wrote:
>> On 03/20/2015 04:48 PM, Gary Benson wrote:
>>> This commit adds support for filenames prefixed	with "target:" to
>>> exec_file_attach.  This is required to correctly follow inferior
>>> exec* calls when a gdb_sysroot prefixed with "target:" is set.
>>
>> Hmm, I don't see how.  Isn't this only true when target_pid_to_exec_file
>> prepends the sysroot, which it doesn't yet?  I think this should move
>> to that other series.
> 
> Search for gdb_sysroot in infrun.c.

Ah, thanks.  I went from memory (which obviously failed), while
I should have looked.

> 
>> A couple bits could use more explanation (in commit log and/or
>> comments):
>>
>>  - Why is writing into executable files not supported with "target:" ?
>>  - The skipping of gdb_realpath_keepfile.
> 
> Are these two additions ok?

Yes, thanks.

BTW, looking again at the patch I noticed a spurious new line:

> +	{
> +
> +	  scratch_chan = openp (getenv ("PATH"), OPF_TRY_CWD_FIRST,

Patch is OK with those fixed.

> 
> diff --git a/gdb/exec.c b/gdb/exec.c
> index 6b6fc7d..ce61303 100644
> --- a/gdb/exec.c
> +++ b/gdb/exec.c
> @@ -192,6 +192,7 @@ exec_file_attach (const char *filename, int from_tty)
>  
>        if (load_via_target)
>  	{
> +	  /* gdb_bfd_fopen does not support "target:" filenames.  */
>  	  if (write_files)
>  	    warning (_("writing into executable files is "
>  		       "not supported for %s sysroots"),

That's a limitation that can be lifted later, by either
teaching gdb_bfd_fopen about "target:" or using an alternative
like gdb_bfd_open instead, right?

Thanks,
Pedro Alves
  

Pedro Alves wrote:
> On 04/01/2015 02:55 PM, Gary Benson wrote:
> > Pedro Alves wrote:
> > > On 03/20/2015 04:48 PM, Gary Benson wrote:
> > > > This commit adds support for filenames prefixed	with "target:" to
> > > > exec_file_attach.  This is required to correctly follow inferior
> > > > exec* calls when a gdb_sysroot prefixed with "target:" is set.
> > >
> > > Hmm, I don't see how.  Isn't this only true when target_pid_to_exec_file
> > > prepends the sysroot, which it doesn't yet?  I think this should move
> > > to that other series.
> > 
> > Search for gdb_sysroot in infrun.c.
> 
> Ah, thanks.  I went from memory (which obviously failed), while
> I should have looked.

I may make a followup patch to convert that bit to use exec_file_find
from my "Don't require 'file' for remote targets" series when that's
in.

> > diff --git a/gdb/exec.c b/gdb/exec.c
> > index 6b6fc7d..ce61303 100644
> > --- a/gdb/exec.c
> > +++ b/gdb/exec.c
> > @@ -192,6 +192,7 @@ exec_file_attach (const char *filename, int from_tty)
> >  
> >        if (load_via_target)
> >  	{
> > +	  /* gdb_bfd_fopen does not support "target:" filenames.  */
> >  	  if (write_files)
> >  	    warning (_("writing into executable files is "
> >  		       "not supported for %s sysroots"),
> 
> That's a limitation that can be lifted later, by either teaching
> gdb_bfd_fopen about "target:" or using an alternative like
> gdb_bfd_open instead, right?

Yes, it's not a permanent restriction, just stuff that's not written
yet.  The appropriate methods are in RSP I think, it's just gdb_bfd.c
that needs touching.  (It could probably be refactored some too,
maybe make gdb_bfd_open a special case of gdb_bfd_fopen and put all
the real code in the latter.) 

Cheers,
Gary