A cute bug involving aliases and _remote_files

A cute bug involving aliases and _remote_files

[Philippe Troin]

Hi,

I have the following aliases in my .zshenv:
  alias ls='command ls -Fbv'

The interesting part there is the "-b" which instructs ls to quote
unprintable characters and spaces.

But in _remote_files we do:
  _call_program files $cmd $cmd_args $host ls -d1FL -- 
which will use the alias, and when I complete a file with a space, zsh
ends up prefixing the space with 7 backslashes, as there's now an extra
round of escaping performed.

Would it make sense to use "command ls" instead of "ls" for remote
directory listings in _remote_files?  That would make sure that no
aliases nor functions are involved.
I don't need any downsides as all shells that I'm aware of have the
"command" pre-command modifier.

Example of the issue (you need to alias ls as described above for this
to show):

% mkdir test_dir
% touch test_dir/'a file'
% scp localhost:$PWD/test_dir/a
                               ^ Press TAB
===> completes to:
% scp localhost:$PWD/test_dir/a\\\\\\\ file

I have this in my .zshrc in the meantime:
     _remote_files() {
       unfunction _remote_files
       autoload +X _remote_files
       if [[ $functions[_remote_files] != *' command ls -' ]]
       then
         functions[_remote_files]=${functions[_remote_files]// ls -/ command ls -}
       fi
       _remote_files $*
     }


Yet it would be neat if this were fixed upstream.
Thanks,

Phil.

Re: A cute bug involving aliases and _remote_files

[dana]

On 25 Feb 2019, at 20:49, Philippe Troin <phil@fifi.org> wrote:
>Would it make sense to use "command ls" instead of "ls" for remote
>directory listings in _remote_files?  That would make sure that no
>aliases nor functions are involved.

That *seems* reasonable...?

Now that you mention it, GNU ls changed its default quoting behaviour somewhat
recently. I think it only quotes when (a) QUOTING_STYLE is set, (b) one of the
quoting options is given on the command line, or (c) the output is a TTY — so
it's unlikely to be an issue for _remote_files, given how it's normally used.
But maybe it'd be a good idea to force it off anyway, just in case?

dana


diff --git a/Completion/Unix/Type/_remote_files b/Completion/Unix/Type/_remote_files
index 267715a51..89cf102cf 100644
--- a/Completion/Unix/Type/_remote_files
+++ b/Completion/Unix/Type/_remote_files
@@ -59,7 +59,10 @@ if zstyle -T ":completion:${curcontext}:files" remote-access; then
   fi
 
   # remote filenames
-  remfiles=(${(M)${(f)"$(_call_program files $cmd $cmd_args $host ls -d1FL -- "$rempat" 2>/dev/null)"}%%[^/]#(|/)})
+  remfiles=(${(M)${(f)"$(
+    _call_program files $cmd $cmd_args $host \
+      QUOTING_STYLE=literal command ls -d1FL -- "$rempat" 2>/dev/null
+  )"}%%[^/]#(|/)})
 
   compset -P '*/'
   compset -S '/*' || (( ${args[(I)-/]} )) || suf='remote file'

Re: A cute bug involving aliases and _remote_files

[Peter Stephenson]

On Tue, 2019-02-26 at 03:23 -0600, dana wrote:
> On 25 Feb 2019, at 20:49, Philippe Troin <phil@fifi.org> wrote:
> > 
> > Would it make sense to use "command ls" instead of "ls" for remote
> > directory listings in _remote_files?  That would make sure that no
> > aliases nor functions are involved.
> That *seems* reasonable...?

Looks to me like that's probably a good rule of thumb any time you're
relying on a bunch of specific arguments to a command... But ls is
particularly often aliased.

pws

Re: A cute bug involving aliases and _remote_files

[Philippe Troin]

On Tue, 2019-02-26 at 03:23 -0600, dana wrote:
> On 25 Feb 2019, at 20:49, Philippe Troin <phil@fifi.org> wrote:
> > Would it make sense to use "command ls" instead of "ls" for remote
> > directory listings in _remote_files?  That would make sure that no
> > aliases nor functions are involved.
> 
> That *seems* reasonable...?
> 
> Now that you mention it, GNU ls changed its default quoting behaviour somewhat
> recently. I think it only quotes when (a) QUOTING_STYLE is set, (b)
> one of the
> quoting options is given on the command line, or (c) the output is a
> TTY — so
> it's unlikely to be an issue for _remote_files, given how it's
> normally used.
> But maybe it'd be a good idea to force it off anyway, just in case?

I'm completely unaware of how QUOTING_STYLE could possibly be getting
us into trouble, but the way you call the binary (variable=value
command ls) is bourne-shell only (bash, sh, ash, ksh), and will not
work if the remote system is using c-shells (csh, tcsh).
I'd recommend using env, as in:

+  remfiles=(${(M)${(f)"$(
+    _call_program files $cmd $cmd_args $host \
+      env QUOTING_STYLE=literal ls -d1FL -- "$rempat" > 2>/dev/null
+  )"}%%[^/]#(|/)})

That will work everywhere.
Now, I haven't seen anyone using c-shells in a very long while, but I'm
sure there are some holdouts :-)

Since you use "env" instead of calling ls directory, you don't need
command anymore, because no alias expansion will occur for ls.
I assume we don't care about global aliases, which none of the proposed
solutions would handle anyways.

Phil.

> diff --git a/Completion/Unix/Type/_remote_files
> b/Completion/Unix/Type/_remote_files
> index 267715a51..89cf102cf 100644
> --- a/Completion/Unix/Type/_remote_files
> +++ b/Completion/Unix/Type/_remote_files
> @@ -59,7 +59,10 @@ if zstyle -T ":completion:${curcontext}:files"
> remote-access; then
>    fi
>  
>    # remote filenames
> -  remfiles=(${(M)${(f)"$(_call_program files $cmd $cmd_args $host ls
> -d1FL -- "$rempat" 2>/dev/null)"}%%[^/]#(|/)})
> +  remfiles=(${(M)${(f)"$(
> +    _call_program files $cmd $cmd_args $host \
> +      QUOTING_STYLE=literal command ls -d1FL -- "$rempat"
> 2>/dev/null
> +  )"}%%[^/]#(|/)})
>  
>    compset -P '*/'
>    compset -S '/*' || (( ${args[(I)-/]} )) || suf='remote file'
> 

Re: A cute bug involving aliases and _remote_files

[dana]

On 27 Feb 2019, at 02:27, Philippe Troin <phil@fifi.org> wrote:
>I'm completely unaware of how QUOTING_STYLE could possibly be getting
>us into trouble, but the way you call the binary (variable=value
>command ls) is bourne-shell only (bash, sh, ash, ksh), and will not
>work if the remote system is using c-shells (csh, tcsh).

Oh, i forgot. It won't work in fish either. Not good enough :/

The quoting thing probably isn't a big deal, like i said — seems easiest to
just go with the original suggestion and leave it out. Unless someone feels
strongly otherwise, i think i'll do that

dana