Alias completion

[hauleth]

It will be nice if we can make completion work for aliases too. I.e.

function g; git $argv; end
completion -c g --alias git

And now everything that will be completed in git <tab> will work also in g <tab>.

[zanchey]

see also #165

[zanchey]

see also #165

[stestagg]

It may be simpler to add an argument to the 'function' builtin, that allows you to specify that the function 'completes like'.

Alias can then provide this argument by default.

An example:

alias gp="git push"

would result in

function --completes-like="git push" gp; git push $argv; end

The completer then will just virtually substitute 'git push' for gp when it needs to.

This may be simpler than parsing the function body every time

[stestagg]

It may be simpler to add an argument to the 'function' builtin, that allows you to specify that the function 'completes like'.

Alias can then provide this argument by default.

An example:

alias gp="git push"

would result in

function --completes-like="git push" gp; git push $argv; end

The completer then will just virtually substitute 'git push' for gp when it needs to.

This may be simpler than parsing the function body every time

[hauleth]

@stestagg the way what it will be done is left to the creator, but I also think that manually setting completition is better than parsing body function. But using completion function to alias completion also can be useful i.e. to alias pacman completion to yaourt.

[hauleth]

@stestagg the way what it will be done is left to the creator, but I also think that manually setting completition is better than parsing body function. But using completion function to alias completion also can be useful i.e. to alias pacman completion to yaourt.

[gustafj]

An argument to the function builtin would be enough for all cases, and would have the benefit of being defined in the same place as the function which needs it.
As opposed to a argument to complete which could be defined anywhere.

The pacman example could also use this syntax (since functions has priority over commands) (but it gets a bit repetitive...) ex:
function --completes-like="pacman" yaourt; command yaourt; end

[gustafj]

An argument to the function builtin would be enough for all cases, and would have the benefit of being defined in the same place as the function which needs it.
As opposed to a argument to complete which could be defined anywhere.

The pacman example could also use this syntax (since functions has priority over commands) (but it gets a bit repetitive...) ex:
function --completes-like="pacman" yaourt; command yaourt; end

[JeanMertz]

I just found out about this "missing feature" as well.

function gco
  git checkout $argv
end

This won't auto-complete the available branches/tags, which is really a shame. To implement this myself I had to c/p the relevant parts from shared/completions/git.fish

[JeanMertz]

I just found out about this "missing feature" as well.

function gco
  git checkout $argv
end

This won't auto-complete the available branches/tags, which is really a shame. To implement this myself I had to c/p the relevant parts from shared/completions/git.fish

[gf3]

This is a big one for me, too.

[gf3]

This is a big one for me, too.

[mgsk]

For the love of all that is fishy, make it so!

[mgsk]

For the love of all that is fishy, make it so!

[maxfl]

Before alias completion is supported you can use this function to produce alias completion:

function make_completion --argument-names alias command
    echo "
    function __alias_completion_$alias
        set -l cmd (commandline -o)
        set -e cmd[1]
        complete -C\"$command \$cmd\"
    end
    " | .
    complete -c $alias -a "(__alias_completion_$alias)"
end

use it as follows:
make_completion pacs 'pacman -S'
for alias pacs for command 'pacman -S'

[maxfl]

Before alias completion is supported you can use this function to produce alias completion:

function make_completion --argument-names alias command
    echo "
    function __alias_completion_$alias
        set -l cmd (commandline -o)
        set -e cmd[1]
        complete -C\"$command \$cmd\"
    end
    " | .
    complete -c $alias -a "(__alias_completion_$alias)"
end

use it as follows:
make_completion pacs 'pacman -S'
for alias pacs for command 'pacman -S'

[maxfl]

Or another implementation, without creating intermediate function. It can be added to the 'alias' definition.

function make_completion --argument alias command
    complete -c $alias -a "(
        set -l cmd (commandline -op);
        set -e cmd[1];
        complete -C\"$command \$cmd\";
    )"
end

[maxfl]

Or another implementation, without creating intermediate function. It can be added to the 'alias' definition.

function make_completion --argument alias command
    complete -c $alias -a "(
        set -l cmd (commandline -op);
        set -e cmd[1];
        complete -C\"$command \$cmd\";
    )"
end

[mgsk]

Wonderful! This is how I've used it:

function make_completion --argument-names alias command
    echo "                                             
    function __alias_completion_$alias                 
        set -l cmd (commandline -o)                    
        set -e cmd[1]                                  
        complete -C\"$command \$cmd\"                  
    end                                                
    " | .                                              
    complete -c $alias -a "(__alias_completion_$alias)"
end                                                    

function aptin; sudo apt-get install $argv; end;

make_completion aptin 'apt-get install'

On Thu, Jan 24, 2013 at 03:06:52AM -0800, maxfl wrote:

Before alias completion is supported you can use this function to produce alias completion:

function make_completion --argument-names alias command
    echo "
    function __alias_completion_$alias
        set -l cmd (commandline -o)
        set -e cmd[1]
        complete -C\"$command \$cmd\"
    end
    " | .
    complete -c $alias -a "(__alias_completion_$alias)"
end

use it as follows:
make_completion pacs 'pacman -S'
for alias pacs for command 'pacman -S'

---

Reply to this email directly or view it on GitHub:
#393 (comment)

[mgsk]

Wonderful! This is how I've used it:

function make_completion --argument-names alias command
    echo "                                             
    function __alias_completion_$alias                 
        set -l cmd (commandline -o)                    
        set -e cmd[1]                                  
        complete -C\"$command \$cmd\"                  
    end                                                
    " | .                                              
    complete -c $alias -a "(__alias_completion_$alias)"
end                                                    

function aptin; sudo apt-get install $argv; end;

make_completion aptin 'apt-get install'

On Thu, Jan 24, 2013 at 03:06:52AM -0800, maxfl wrote:

Before alias completion is supported you can use this function to produce alias completion:

function make_completion --argument-names alias command
    echo "
    function __alias_completion_$alias
        set -l cmd (commandline -o)
        set -e cmd[1]
        complete -C\"$command \$cmd\"
    end
    " | .
    complete -c $alias -a "(__alias_completion_$alias)"
end

use it as follows:
make_completion pacs 'pacman -S'
for alias pacs for command 'pacman -S'

---

Reply to this email directly or view it on GitHub:
#393 (comment)

[hauleth]

It is almost great except that if I alias git with g and try to complete:

$ g checkout <tab>

this still doesn't work.

[hauleth]

It is almost great except that if I alias git with g and try to complete:

$ g checkout <tab>

this still doesn't work.

[maxfl]

I see. The problem is that 'commandline -o' looses the trailing space. Here is the solution.

function make_completion --argument alias command
    complete -c $alias -xa "(
        set -l cmd (commandline -pc | sed -e 's/^ *\S\+ *//' );
        complete -C\"$command \$cmd\";
    )"
end

Since now we do not split the commandline output, we have to use sed to replace the command. I've also changed '-a' to '-xa' option, because the filename completions should be added by nested 'complete' command.

[maxfl]

I see. The problem is that 'commandline -o' looses the trailing space. Here is the solution.

function make_completion --argument alias command
    complete -c $alias -xa "(
        set -l cmd (commandline -pc | sed -e 's/^ *\S\+ *//' );
        complete -C\"$command \$cmd\";
    )"
end

Since now we do not split the commandline output, we have to use sed to replace the command. I've also changed '-a' to '-xa' option, because the filename completions should be added by nested 'complete' command.

[ridiculousfish]

Does anyone have any thoughts on how commandline ought to be handled? For example, if I write alias gb 'git branch'. The git completion wants to only return completions for the relevant subcommand, which it uses builtin commandline to parse out. Since it won't be able to parse it out of the actual commandline, this will fail.

One possibility would be to make commandline lie and return something with the alias replaced, i.e.

alias gb 'git branch'
gb fi<tab>

Within the completion, commandline would return "git branch fi" instead of "gb fi".

Another possibility is to replace the ad-hoc use of commandline with true subcommand support (so you could write a completion specifically for 'git branch').

[ridiculousfish]

Does anyone have any thoughts on how commandline ought to be handled? For example, if I write alias gb 'git branch'. The git completion wants to only return completions for the relevant subcommand, which it uses builtin commandline to parse out. Since it won't be able to parse it out of the actual commandline, this will fail.

One possibility would be to make commandline lie and return something with the alias replaced, i.e.

alias gb 'git branch'
gb fi<tab>

Within the completion, commandline would return "git branch fi" instead of "gb fi".

Another possibility is to replace the ad-hoc use of commandline with true subcommand support (so you could write a completion specifically for 'git branch').

[gf3]

Maybe we should check out how ZSH does it? It's able to complete things like shell aliases as well as aliases set in .gitconfig.

[gf3]

Maybe we should check out how ZSH does it? It's able to complete things like shell aliases as well as aliases set in .gitconfig.

[maxfl]

@ridiculousfish, there is another way to look at it: it's not an issue of the 'commandline', but an issue of the 'complete':

1. commandline should be as transparent as possible. It returns the command lines as is and doesn't distinguish between commands and aliases.
2. 'complete' can have something like modes, or a possibility to use the completion for another command:
   complete -c gb -C 'git branch' # to tell fish to complete gb as if it were 'git branch'
   It's a thing that is done by that script two posts above, but script is ugly since it requires calling another commands.
3. [a bit offtopic here] complete can have some 'mode' or 'preset' option to provide a set of completions which can be reused by several commands:
   like:
   complete --preset vim -l longcommand1 -s s1 -d desc 1
   complete --preset vim -l longcommand2 -s s2 -d desc 2
   complete -c vim --use-preset vim
   complete -c gvim --use-preset vim

[maxfl]

@ridiculousfish, there is another way to look at it: it's not an issue of the 'commandline', but an issue of the 'complete':

1. commandline should be as transparent as possible. It returns the command lines as is and doesn't distinguish between commands and aliases.
2. 'complete' can have something like modes, or a possibility to use the completion for another command:
   complete -c gb -C 'git branch' # to tell fish to complete gb as if it were 'git branch'
   It's a thing that is done by that script two posts above, but script is ugly since it requires calling another commands.
3. [a bit offtopic here] complete can have some 'mode' or 'preset' option to provide a set of completions which can be reused by several commands:
   like:
   complete --preset vim -l longcommand1 -s s1 -d desc 1
   complete --preset vim -l longcommand2 -s s2 -d desc 2
   complete -c vim --use-preset vim
   complete -c gvim --use-preset vim

[dag]

However you solve this I'd like it if the solution works also for functions, not just aliases. @maxfi's suggestion in the comment above would seem to qualify, for example. Making alias special on the other hand, seems to not qualify.

[dag]

However you solve this I'd like it if the solution works also for functions, not just aliases. @maxfi's suggestion in the comment above would seem to qualify, for example. Making alias special on the other hand, seems to not qualify.

[dag]

BTW if this becomes a reality we also need a way to escape it. Consider alias git hub; you want to keep the git completions, not use the non-existent hub completions. I wonder how zsh deals with that...

[dag]

BTW if this becomes a reality we also need a way to escape it. Consider alias git hub; you want to keep the git completions, not use the non-existent hub completions. I wonder how zsh deals with that...

[Gonzih]

Just migrated to fish from zsh and that issue is now bothering me.

[Gonzih]

Just migrated to fish from zsh and that issue is now bothering me.

[sjl]

I agree that this should be a feature of complete. "Complete this command like this other one" seems like a decent way to solve this problem.

Also @maxfl's make_completion is not working for me, not exactly sure why.

[sjl]

I agree that this should be a feature of complete. "Complete this command like this other one" seems like a decent way to solve this problem.

Also @maxfl's make_completion is not working for me, not exactly sure why.

[ghost]

I think that #731 would fix this.

[ghost]

I think that #731 would fix this.

[tomxtobin]

Having aliases automatically expand as I type (as per the abbreviations feature in #731) would be a huge turn-off. I'm very sensitive to anything that changes what I'm typing out from under me.

[tomxtobin]

Having aliases automatically expand as I type (as per the abbreviations feature in #731) would be a huge turn-off. I'm very sensitive to anything that changes what I'm typing out from under me.

[dag]

Note that abbreviations only expand when you hit enter or space. Try it and you might be surprised at how non-invasive it actually is. Or maybe you're not impressed at all – people are different!

I certainly do think we need some way to "redirect" completions, regardless of aliases and abbreviations. This should probably be a feature of complete because it's the most general solution, and not all "aliases" are in the shell (for example egrep).

[dag]

Note that abbreviations only expand when you hit enter or space. Try it and you might be surprised at how non-invasive it actually is. Or maybe you're not impressed at all – people are different!

I certainly do think we need some way to "redirect" completions, regardless of aliases and abbreviations. This should probably be a feature of complete because it's the most general solution, and not all "aliases" are in the shell (for example egrep).

[stestagg]

I'm not keen on this at all, bash does something similar with tilda
expansion when pressing TAB, and it's annoying and makes editing previous
commands harder.

Thanks

Steve

On Fri, Nov 1, 2013 at 10:53 AM, Dag Odenhall notifications@github.comwrote:

Note that abbreviations only expand when you hit enter or space. Try it
and you might be surprised at how non-invasive it actually is. Or maybe
you're not impressed at all – people are different!

I certainly do think we need some way to "redirect" completions,
regardless of aliases and abbreviations. This should probably be a feature
of complete because it's the most general solution, and not all "aliases"
are in the shell (for example egrep).

—
Reply to this email directly or view it on GitHubhttps://github.com/fish-shell/fish-shell/issues/393#issuecomment-27558448
.

[stestagg]

I'm not keen on this at all, bash does something similar with tilda
expansion when pressing TAB, and it's annoying and makes editing previous
commands harder.

Thanks

Steve

On Fri, Nov 1, 2013 at 10:53 AM, Dag Odenhall notifications@github.comwrote:

Note that abbreviations only expand when you hit enter or space. Try it
and you might be surprised at how non-invasive it actually is. Or maybe
you're not impressed at all – people are different!

I certainly do think we need some way to "redirect" completions,
regardless of aliases and abbreviations. This should probably be a feature
of complete because it's the most general solution, and not all "aliases"
are in the shell (for example egrep).

—
Reply to this email directly or view it on GitHubhttps://github.com/fish-shell/fish-shell/issues/393#issuecomment-27558448
.

[Rameshv]

+1

[Rameshv]

+1

[ridiculousfish]

My thinking is that "completes like" is a property you can set in a function signature: #478

[ridiculousfish]

My thinking is that "completes like" is a property you can set in a function signature: #478

[dag]

@ridiculousfish I'm not convinced that's good enough, though; consider egrep. I guess you can make a wrapper function for that scenario, but it doesn't feel right.

[dag]

@ridiculousfish I'm not convinced that's good enough, though; consider egrep. I guess you can make a wrapper function for that scenario, but it doesn't feel right.

[xfix]

complete --command egrep --like grep
complete --command egrep --erase --short-option E --long-option extended-regexp

Just thinking. Needs better name than --like, but I like this.

[xfix]

complete --command egrep --like grep
complete --command egrep --erase --short-option E --long-option extended-regexp

Just thinking. Needs better name than --like, but I like this.

[hauleth]

For me --like looks good enough.

[hauleth]

For me --like looks good enough.

[ridiculousfish]

@dag "Function signature" is a misnomer. I should have called it command signature. You can set one on a function in its declaration, on external commands, and builtins will have them too.

[ridiculousfish]

@dag "Function signature" is a misnomer. I should have called it command signature. You can set one on a function in its declaration, on external commands, and builtins will have them too.

[xixixao]

This is a real must! For me, the tab completion is the killer feature of fish, but I need to be able to easily add it to my aliases (functions).#393 (comment) works for me.

[xixixao]

This is a real must! For me, the tab completion is the killer feature of fish, but I need to be able to easily add it to my aliases (functions).#393 (comment) works for me.

[ahti]

This is really something where complete is lacking, which also leads to more complicated completion files.

For an example see the completion for pacman, pacsrv, yaourt, ...

All of the completion files for these tools just call __fish_complete_pacman which is not even defined inside the completions folder but in the functions directory. This is a very non-generic way to deal with this problem.

If you take a look into /share/functions, you will find a bunch of functions starting with __fish_complete_, which is also a little code duplication and (imho) a very inelegant way to solve this problem.

This could be easily solved with a --like. The original tool (pacman in this case) contains the "real" completion, yaourt, pacsrv, et. al. just contain a complete yaourt --like pacman.

[ahti]

This is really something where complete is lacking, which also leads to more complicated completion files.

For an example see the completion for pacman, pacsrv, yaourt, ...

All of the completion files for these tools just call __fish_complete_pacman which is not even defined inside the completions folder but in the functions directory. This is a very non-generic way to deal with this problem.

If you take a look into /share/functions, you will find a bunch of functions starting with __fish_complete_, which is also a little code duplication and (imho) a very inelegant way to solve this problem.

This could be easily solved with a --like. The original tool (pacman in this case) contains the "real" completion, yaourt, pacsrv, et. al. just contain a complete yaourt --like pacman.

[melekes]

+1. Feature is critical for me as well.

[melekes]

+1. Feature is critical for me as well.

[tonatiuh]

+1. This is also very important to me, it bothers me to have to write the full command name.

Btw other things from fish are very cool : )

[tonatiuh]

+1. This is also very important to me, it bothers me to have to write the full command name.

Btw other things from fish are very cool : )

[toabi]

+1. That's the only thing I miss since moving from (oh-my)-zsh

[toabi]

+1. That's the only thing I miss since moving from (oh-my)-zsh

[ignlg]

+1. The --like param for complete seems perfect and flexible.

[ignlg]

+1. The --like param for complete seems perfect and flexible.

[ridiculousfish]

This is implemented as 06400b8. The commit message has more details.

In short you can now do this:

complete --command foo --wraps bar

or

function foo --wraps bar
    ...
end

Either way, foo dynamically inherits bar's completions.

I went with -w/--wraps instead of --like because function foo --like bar seemed less clear. I think wraps is also more precise: yaourt wraps pacman in the sense that it includes all of pacman's completions, but also provides its own completions in addition.

I didn't implement support for aliasing to multiple commands, i.e. compete --command gco --wraps 'git checkout' won't work. That use case seems better served by abbreviations to me. zsh allows arbitrarily powerful aliases, e.g. you can write alias haha=''and nowhaha` introduces an unclosed backtick. It's just pure textual substitution. I don't think we want to go that far so I went with something conservative.

Definitely interested in your suggestions or ideas for improvements here.

[ridiculousfish]

This is implemented as 06400b8. The commit message has more details.

In short you can now do this:

complete --command foo --wraps bar

or

function foo --wraps bar
    ...
end

Either way, foo dynamically inherits bar's completions.

I went with -w/--wraps instead of --like because function foo --like bar seemed less clear. I think wraps is also more precise: yaourt wraps pacman in the sense that it includes all of pacman's completions, but also provides its own completions in addition.

I didn't implement support for aliasing to multiple commands, i.e. compete --command gco --wraps 'git checkout' won't work. That use case seems better served by abbreviations to me. zsh allows arbitrarily powerful aliases, e.g. you can write alias haha=''and nowhaha` introduces an unclosed backtick. It's just pure textual substitution. I don't think we want to go that far so I went with something conservative.

Definitely interested in your suggestions or ideas for improvements here.

[gglanzani]

So basically there is still no way to have things like gco do completion? Because I don't think I fully understand what abbreviations means in that context (an example would be great).

[gglanzani]

So basically there is still no way to have things like gco do completion? Because I don't think I fully understand what abbreviations means in that context (an example would be great).

[terlar]

set -U fish_user_abbreviations 'gco=git checkout'

Next time you write gco+<enter> or gco+<space> you would get git checkout.

[terlar]

set -U fish_user_abbreviations 'gco=git checkout'

Next time you write gco+<enter> or gco+<space> you would get git checkout.

[stestagg]

It would be great if the abbreviations could be seamless (i.e. not expand when entered)

[stestagg]

It would be great if the abbreviations could be seamless (i.e. not expand when entered)

[terlar]

That is pretty much the point with the abbreviations though, you could just create a function wrapper if you want to keep it unexpanded. I find it much more clear and useful when it expands, barely have any aliases anymore.

[terlar]

That is pretty much the point with the abbreviations though, you could just create a function wrapper if you want to keep it unexpanded. I find it much more clear and useful when it expands, barely have any aliases anymore.

[stestagg]

But you can't then get completions wrapped if you do, so the feature looses
a lot of it's functionality

On Tue, Aug 26, 2014 at 7:10 PM, Terje Larsen notifications@github.com
wrote:

That is pretty much the point with the abbreviations though, you could
just create a function wrapper if you want to keep it unexpanded. I find it
much more clear and useful when it expands, barely have any aliases anymore.

—
Reply to this email directly or view it on GitHub
#393 (comment)
.

[stestagg]

But you can't then get completions wrapped if you do, so the feature looses
a lot of it's functionality

On Tue, Aug 26, 2014 at 7:10 PM, Terje Larsen notifications@github.com
wrote:

That is pretty much the point with the abbreviations though, you could
just create a function wrapper if you want to keep it unexpanded. I find it
much more clear and useful when it expands, barely have any aliases anymore.

—
Reply to this email directly or view it on GitHub
#393 (comment)
.

[gglanzani]

@terlar Thanks, that works. It's somewhat an undocumented feature (found more info in the issue tracker).

[gglanzani]

@terlar Thanks, that works. It's somewhat an undocumented feature (found more info in the issue tracker).

[ridiculousfish]

It's an incomplete feature - we don't have any real UI for it yet (either in fish_config or command line).

[ridiculousfish]

It's an incomplete feature - we don't have any real UI for it yet (either in fish_config or command line).

[najamelan]

For people wondering about git aliases, I can say that it works if you set your alias in .gitconfig instead of in fish configuration.

[najamelan]

For people wondering about git aliases, I can say that it works if you set your alias in .gitconfig instead of in fish configuration.

[niedzielski]

Thanks! This is very convenient! Worth building master for!

[niedzielski]

Thanks! This is very convenient! Worth building master for!

[joar]

This does not seem to solve the issue when you've abbreviated git add to ga.

Using

function --wraps 'git add' ga
    git add $argv;
end

does not help, and using

function --wraps 'git' ga
    git add $argv;
end

gives you completions for git instead of git add. Presumable since the completions for giit sub-commands are defined using conditions, e.g:

complete --no-files \
    --command git \
    --arguments '(__fish_git_add_files)' \
    --condition '__fish_git_using_command add'

(Rambling, hope it's helpful)

[joar]

This does not seem to solve the issue when you've abbreviated git add to ga.

Using

function --wraps 'git add' ga
    git add $argv;
end

does not help, and using

function --wraps 'git' ga
    git add $argv;
end

gives you completions for git instead of git add. Presumable since the completions for giit sub-commands are defined using conditions, e.g:

complete --no-files \
    --command git \
    --arguments '(__fish_git_add_files)' \
    --condition '__fish_git_using_command add'

(Rambling, hope it's helpful)

[zanchey]

Correct, abbreviations only support the token in command position - see #1976 for some discussion.

[zanchey]

Correct, abbreviations only support the token in command position - see #1976 for some discussion.