Re: perl local decoding leads to empty buffers

[David Smith]

Indeed my workaround was incomplete. There were two other small issues
related to handling decode commands. I've attached a patch that works
completely for me. Please take a look.

Cheers,
David


On Sat, Apr 20, 2013 at 1:07 PM, David Smith <address@hidden> wrote:
> Hi!
>
> I'm trying to use tramp to edit files via ssh from an openwrt box.
> When I attempt find-file on a tramp filename for that box, tramp
> appears to work, no errors are reported, but the resulting buffer
> contains no contents for files that have contents. I'm using tramp
> 2.2.3-24.1 bundled with emacs 24.1 on ubuntu 12.10.
>
> I debugged the problem a bit and have a workaround but I doubt it is
> the best solution. Can you please help me understand what the right
> fix should be? I'm happy to test.
>
> What I noticed is that from tramp-sh-handle-file-local-copy, the
> remote-encoding and local-decoding commands are gotten from
> tramp-get-inline-coding, then passed to
> tramp-call-local-coding-command. The decoding command looks like this
> initially:
>
>   perl -e 'binmode STDIN; binmode STDOUT; print unpack(q{u*}, join q{}, <>)'
>
> And when returned from tramp-get-inline-coding, a >%s is appended, producing
>
>   perl -e 'binmode STDIN; binmode STDOUT; print unpack(q{u*}, join q{}, <>)' 
> >%s
>
> tramp-call-local-coding-command take three args: cmd, input, and
> output. tramp-sh-handle-file-local-copy invokes
> tramp-call-local-coding-command with the perl command with >%s
> appended for cmd, the temporary file containing the output of the
> remote-encoding perl command as input, and a second temporary file as
> output. tramp-call-local-coding-command also interprets any %s in cmd
> as a place to put the input filename. The bug manifests when
> tramp-call-local-coding-command interpolates the %s in the cmd to the
> first temporary file containing the input, making perl overwrite the
> encoded contents, and giving me my empty result buffer for output.
>
> My workaround is to modify tramp-get-inline-coding such that it does
> not add >%s to the local decoding command. I've attached a patch. It
> seems deceptively easy and I am probably not realizing some case where
> this is wrong. Please take a look.
>
> Cheers,
> David

tramp-sh.el.patch
Description: Binary data