Don't suggest completion functions when 'correcting' on non-existant commands

Don't suggest completion functions when 'correcting' on non-existant commands

[Richard Hartmann]

Hi all,

bug 257336[1] made me think of something that has annoyed me
for ages, but never enough to actually send email to the list:

richih@roadwarrior ~ % figlet
zsh: correct 'figlet' to '_figlet'? [N/y/a/e]

I don't know enough about zsh internas to guesstimate if this is
an easy fix or not. Two possible solutions which come to mind:

1) Make zsh aware of what functions came in through $FPATH
and which came in through $PATH so the correction function
does not even see them when looking for other executable
commands.

2) Make the completion function do the filtering outlined in 1)


Richard

[1] http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=257336

Re: Don't suggest completion functions when 'correcting' on non-existant commands

[Frank Terbeck]

Richard Hartmann <richih.mailinglist@gmail.com>:
> bug 257336[1] made me think of something that has annoyed me
> for ages, but never enough to actually send email to the list:
> 
> richih@roadwarrior ~ % figlet
> zsh: correct 'figlet' to '_figlet'? [N/y/a/e]

Here's a possible solution for this:
<http://bewatermyfriend.org/posts/2007/12-26.11-50-38-tooltime.html>

I've thought about submitting this for inclusion... And I'd look into
it the next days, if there's interest in it.

Regards, Frank

-- 
In protocol design, perfection has been reached not when there is
nothing left to add, but when there is nothing left to take away.
                                                  -- RFC 1925

Re: Don't suggest completion functions when 'correcting' on non-existant commands

[Bart Schaefer]

[Removed debian bugs from recipients]

On Dec 31,  6:48pm, Richard Hartmann wrote:
}
} richih@roadwarrior ~ % figlet
} zsh: correct 'figlet' to '_figlet'? [N/y/a/e]
} 
} I don't know enough about zsh internas to guesstimate if this is
} an easy fix or not. Two possible solutions which come to mind:
} 
} 1) Make zsh aware of what functions came in through $FPATH
} and which came in through $PATH

No, that doesn't make sense.  Functions never "come in" through $PATH,
only external commands do; and of functions only autoloaded ones are
affected by $FPATH.

A closer approximation to what you want would be for correction to
filter out names that are bound to widgets, but even that doesn't go
far enough; it'd also have to filter out names that begin with an
underscore.  The reason for the leading-underscore convention is so
that things like that would be possible, but of course it is only a
convention.

Filtering out widgets is problematic because correction is a part of
the base shell and ZLE is added as a module.

} 2) Make the completion function do the filtering outlined in 1)

That doesn't work either because correction and completion have almost
nothing to do with one another, internally.  Completion is long since
over before correction kicks in.

What you really want is something like $fignore applied to correction.

On Dec 31,  7:12pm, Frank Terbeck wrote:
} 
} Here's a possible solution for this:
} <http://bewatermyfriend.org/posts/2007/12-26.11-50-38-tooltime.html>

That's not really a solution; it just replaces the correction prompt
with a different prompt.  It might be interesting for other reasons.

Re: Don't suggest completion functions when 'correcting' on non-existant commands

[Frank Terbeck]

Bart Schaefer <schaefer@brasslantern.com>:
[...]
> On Dec 31,  7:12pm, Frank Terbeck wrote:
> } 
> } Here's a possible solution for this:
> } <http://bewatermyfriend.org/posts/2007/12-26.11-50-38-tooltime.html>
> 
> That's not really a solution; it just replaces the correction prompt
> with a different prompt.

Well, to me it's not really another prompt, because you're not
actually prompted for anything. It's merely a warning message, that
gets displayed below the prompt. If you hit enter again, the
command line is processed normally (probably with command correction
kicking in). So, it does prevent a prompt the first time you hit
enter.

E.g.: If it's due to a not-yet-installed piece of software you can
push-input the current cmdline, aptitude-install (or whatever way your
OS does it) and then just hit enter to have the pushed command
executed - with the missing program installed.

> It might be interesting for other reasons.

What would those reasons be? ...maybe you see use-cases I didn't see,
yet. :-)

Regards, Frank

-- 
In protocol design, perfection has been reached not when there is
nothing left to add, but when there is nothing left to take away.
                                                  -- RFC 1925

Re: Don't suggest completion functions when 'correcting' on non-existant commands

[Bart Schaefer]

On Jan 2, 11:54pm, Frank Terbeck wrote:
}
} > On Dec 31,  7:12pm, Frank Terbeck wrote:
} > } 
} > } <http://bewatermyfriend.org/posts/2007/12-26.11-50-38-tooltime.html>
} > 
} > That's not really a solution; it just replaces the correction prompt
} > with a different prompt.
} 
} Well, to me it's not really another prompt, because you're not
} actually prompted for anything.

It nevertheless stops whatever is going on and makes you hit enter, which
is not that much better than answering "N" to a request to correct the
command name.

} E.g.: If it's due to a not-yet-installed piece of software you can
} push-input the current cmdline, aptitude-install (or whatever way your
} OS does it) and then just hit enter to have the pushed command
} executed - with the missing program installed.
} 
} > It might be interesting for other reasons.
} 
} What would those reasons be? ...maybe you see use-cases I didn't see,
} yet. :-)

Well, on Ubuntu under bash when a command is not found you get told what
to do to install it (e.g. what package to apt-get install).  This could
be the basis for something like that for zsh.

Re: Don't suggest completion functions when 'correcting' on non-existant commands

[Frank Terbeck]

Bart Schaefer <schaefer@brasslantern.com>:
> On Jan 2, 11:54pm, Frank Terbeck wrote:
> } > On Dec 31,  7:12pm, Frank Terbeck wrote:
> } > } 
> } > } <http://bewatermyfriend.org/posts/2007/12-26.11-50-38-tooltime.html>
> } > 
> } > That's not really a solution; it just replaces the correction prompt
> } > with a different prompt.
> } 
> } Well, to me it's not really another prompt, because you're not
> } actually prompted for anything.
> 
> It nevertheless stops whatever is going on and makes you hit enter, which
> is not that much better than answering "N" to a request to correct the
> command name.

Since the problem only occurs as soon as you hit enter, I don't see
the problem, that this workaround (yeah, let's call it that way) only
kicks in as soon as you hit enter.

If it detects the problem, it displays a warning below the prompt and
ignores the first enter completely. Nothing changes. So, at least for
me (and some other folks who I know use this, too) that's a lot less
annoying than the correction question.

Of course, a solution within zle would be way cleaner. I tried that,
before I wrote that accept-line wrapper back then. Obviously, I
failed - who would have guessed. :-)

> } E.g.: If it's due to a not-yet-installed piece of software you can
> } push-input the current cmdline, aptitude-install (or whatever way your
> } OS does it) and then just hit enter to have the pushed command
> } executed - with the missing program installed.
> } 
> } > It might be interesting for other reasons.
> } 
> } What would those reasons be? ...maybe you see use-cases I didn't see,
> } yet. :-)
> 
> Well, on Ubuntu under bash when a command is not found you get told what
> to do to install it (e.g. what package to apt-get install).  This could
> be the basis for something like that for zsh.

Yeah, that could probably be done. I'll take a look at it as soon as I
can find the time. :)

Regards, Frank

-- 
In protocol design, perfection has been reached not when there is
nothing left to add, but when there is nothing left to take away.
                                                  -- RFC 1925

Re: Don't suggest completion functions when 'correcting' on non-existant commands

[Richard Hartmann]

On Sat, Jan 3, 2009 at 12:42, Frank Terbeck <ft@bewatermyfriend.org> wrote:

> Yeah, that could probably be done. I'll take a look at it as soon as I
> can find the time. :)

Not exactly mine any more but bump nonetheless :)


Richard