Re: master 1196e3f: (tramp-open-connection-setup-interactive-shell): Send -onlcr as well.

Re: master 1196e3f: (tramp-open-connection-setup-interactive-shell): Send -onlcr as well.

[Michael Albinus]

Noah Friedman <friedman+savannah@splode.com> writes:

Hi Noah,

> diff --git a/lisp/net/tramp-sh.el b/lisp/net/tramp-sh.el
> index 433b2ba..7d24b54 100644
> --- a/lisp/net/tramp-sh.el
> +++ b/lisp/net/tramp-sh.el
> @@ -4153,7 +4153,7 @@ process to set up.  VEC specifies the connection."
>  
>      ;; Disable tab and echo expansion.
>      (tramp-message vec 5 "Setting up remote shell environment")
> -    (tramp-send-command vec "stty tab0 -inlcr -echo kill '^U' erase '^H'" t)
> +    (tramp-send-command vec "stty tab0 -inlcr -onlcr -echo kill '^U' erase '^H'" t)
>      ;; Check whether the echo has really been disabled.  Some
>      ;; implementations, like busybox of embedded GNU/Linux, don't
>      ;; support disabling.

Could you pls explain the reason for this change? The default setting is
"onlcr", which is also used when you apply "stty sane".

Is there a problem you have detected with Tramp? The current settings
are used for 10+ years, and nobody did report yet.

Thanks, and best regards, Michael.

Re: master 1196e3f: (tramp-open-connection-setup-interactive-shell): Send -onlcr as well.

[Noah Friedman]

>Could you pls explain the reason for this change? The default setting is
>"onlcr", which is also used when you apply "stty sane".

That's exactly the problem.  Tramp runs an "interactive" shell on the
remote end; specifically, it sets "PS1".  That means that if your remote
shell's .profile or .bashrc thinks you're interactive and runs "stty sane"
as part of the usual login process, then the onlcr flag gets set and all of
the output lines end with ^M.  Subsequently, after trying to search for
commands to run on the remote end, tramp tries to send commands like
"/usr/bin/ls^M -args" and concludes that there is no valid ls on the remote
system.

Evidence from my .bash_history file on the remote end once I disabled my
.bashrc entirely:

	exec env ENV='' HISTFILE=/dev/null PROMPT_COMMAND='' PS1=\#\$\  PS2='' PS3='' /bin/sh

I've known for many years to put a guard in my .bashrc along the lines of

	case $PS1 in '' ) return 0 ;; esac

before doing anything that would confuse batch processing, but tramp was
specifically defeating that.

This is not just some perverse behavior on my part.  This is standard idiom
for shell initialization.  Actually, tramp is bound to cause all kinds of
other trouble besides by doing this, but at the very least, forcing -onlcr
will avoid screwing up absolutely every line of output.

And funny you should say the current settings have been used for years,
because for years tramp hasn't worked for me and I finally got tired of
trying it once in a while and seeing that it still didn't work.  I've been
punting and setting up sshfs connections manually instead, but I was in a
mood last week and decided to debug it.

In <87egh0feec.fsf@gmx.de>, Michael Albinus writes:
>Noah Friedman <friedman+savannah@splode.com> writes:
>
>Hi Noah,
>
>> diff --git a/lisp/net/tramp-sh.el b/lisp/net/tramp-sh.el
>> index 433b2ba..7d24b54 100644
>> --- a/lisp/net/tramp-sh.el
>> +++ b/lisp/net/tramp-sh.el
>> @@ -4153,7 +4153,7 @@ process to set up.  VEC specifies the connection."
>>  
>>      ;; Disable tab and echo expansion.
>>      (tramp-message vec 5 "Setting up remote shell environment")
>> -    (tramp-send-command vec "stty tab0 -inlcr -echo kill '^U' erase '^H'" t)
>> +    (tramp-send-command vec "stty tab0 -inlcr -onlcr -echo kill '^U' erase '^H'" t)
>>      ;; Check whether the echo has really been disabled.  Some
>>      ;; implementations, like busybox of embedded GNU/Linux, don't
>>      ;; support disabling.
>
>Could you pls explain the reason for this change? The default setting is
>"onlcr", which is also used when you apply "stty sane".
>
>Is there a problem you have detected with Tramp? The current settings
>are used for 10+ years, and nobody did report yet.
>
>Thanks, and best regards, Michael.
>

Re: master 1196e3f: (tramp-open-connection-setup-interactive-shell): Send -onlcr as well.

[Michael Albinus]

Noah Friedman <friedman@splode.com> writes:

> That's exactly the problem.  Tramp runs an "interactive" shell on the
> remote end; specifically, it sets "PS1".  That means that if your remote
> shell's .profile or .bashrc thinks you're interactive and runs "stty sane"
> as part of the usual login process, then the onlcr flag gets set and all of
> the output lines end with ^M.  Subsequently, after trying to search for
> commands to run on the remote end, tramp tries to send commands like
> "/usr/bin/ls^M -args" and concludes that there is no valid ls on the remote
> system.

I see. Strange enough, that nobody else did report ever.

> I've known for many years to put a guard in my .bashrc along the lines of
>
> 	case $PS1 in '' ) return 0 ;; esac
>
> before doing anything that would confuse batch processing, but tramp was
> specifically defeating that.

Tramp sets TERM=dumb; you could test for it.

> And funny you should say the current settings have been used for years,
> because for years tramp hasn't worked for me and I finally got tired of
> trying it once in a while and seeing that it still didn't work.  I've been
> punting and setting up sshfs connections manually instead, but I was in a
> mood last week and decided to debug it.

sshfs has its merits, but when it comes to running processes on the
remote side, you need Tramp.

Well, I've pushed your patch also into the Tramp git repo. Let's see
whether people will kill us because of this. I plan to release Tramp
2.2.13 end of the year (if I survive), until then there's enough time to
get reports from people using either Emacs git master or Tramp git
master.

Thanks for all your work, and best regards, Michael.