Re: [PATCH] remote.c: Allow inferior to reply with an error

[Andrew Burgess <aburgess@redhat.com>]

AlexandraHájková@sourceware.org writes:

> From: Alexandra Hájková <ahajkova@redhat.com>
>
> When gdb communicates with the inferior with the remote
> protocol, the only possible response to the QSetWorkingDir
> packet is "OK". If the inferior will reply with anything
> else, gdb will complain about the unexpected reply and stop
> its communication with the inferior.
>
> [remote] Sending packet: $QSetWorkingDir:2f746d70#bb
> [remote] Packet received: E00
> Remote replied unexpectedly while setting the inferior's working
> directory: E00
> (gdb)
>
> But setting the inferior's working dir is not always possible due
> to various reasons and we may not want to always stop the communication.
> This patch proposes to just warn the user in a case it wasn't possible
> to set the working dir but proceed without an error.
>
> [remote] Sending packet: $QSetWorkingDir:2f746d70#bb
> remote] Packet received: E00
> warning: Remote failed to set the inferior's working directory: E00
> [remote] Sending packet: $vRun;2f7573722f62696e2f6563686f#3d
> ---
>  gdb/doc/gdb.texinfo | 3 +++
>  gdb/remote.c        | 8 +++-----

Changes like this should definitely have a test, but, in this case, I
have bigger worries, see below...

>  2 files changed, 6 insertions(+), 5 deletions(-)
>
> diff --git a/gdb/doc/gdb.texinfo b/gdb/doc/gdb.texinfo
> index dd8f8bc757c..54cab3afa9c 100644
> --- a/gdb/doc/gdb.texinfo
> +++ b/gdb/doc/gdb.texinfo
> @@ -42515,6 +42515,9 @@ Reply:
>  @table @samp
>  @item OK
>  The request succeeded.
> +
> +@item E @var{nn}
> +An error occurred.  The error number @var{nn} is given as hex digits.
>  @end table
>  
>  @item qfThreadInfo
> diff --git a/gdb/remote.c b/gdb/remote.c
> index 218bca30d04..db5b949a28b 100644
> --- a/gdb/remote.c
> +++ b/gdb/remote.c
> @@ -10421,11 +10421,9 @@ remote_target::extended_remote_set_inferior_cwd ()
>        if (packet_ok (rs->buf,
>  		     &remote_protocol_packets[PACKET_QSetWorkingDir])
>  	  != PACKET_OK)
> -	error (_("\
> -Remote replied unexpectedly while setting the inferior's working\n\
> -directory: %s"),
> -	       rs->buf.data ());
> -
> +	warning (_("\
> +Remote failed to set the inferior's working directory: %s"),
> +                 rs->buf.data ());

I don't think this should be merged.

I don't think changing this error into a warning is a good idea.
Surely, if GDB can't switch to the directory that the user expects, then
we should stop and tell the user that we can't do what they've asked,
then the user could update things and try again.

As a (maybe extreme) example, what if the user was debugging 'rm -f *',
but first changed from their $HOME directory to '/tmp/test-dir/'.  If
the requested directory doesn't exist then we're going to just carry on
and run the test in $HOME .... that seems like a bad idea to me.

Thanks,
Andrew