tab-completion implemented for Kawa

tab-completion implemented for Kawa

[Per Bothner]

I checked in an implementation of command-completion (tab-completion)
for Kawa.  This is context-senstive - for example the following works
when using the JLine2 input-ediing library:

#|Kawa:1|# ab<TAB>s
abs              absolute-path?
#|Kawa:1|# abs<ENTER>
#<procedure abs>

Kawa wills in the 's' and gives you two options; <ENTER> choose "abs".

What makes it interesting is that it is content-dependent,
and includes bindings in the lexical scope:

#|Kawa:2|# (let ((absol 12)) abso<TAB>l
absol            absolute-path?

It works by running the compiler in tentative mode
with "(let ((absol 12)) abso<COMPLETE>" as the input.
Here <COMPLETE> is a special private-use Unicode character
that tells the Kawa name-lookup framework to look for names
starting with "abso".

This "tentative mode" could also be useful for on-the-fly
error checking, syntax coloring, use/definition cross-references.
But that is for another day.

Currently this only works with the JLine2 library.
You have to:
    ./configure --with-jline2=/path/to/jline2.jar

However, most of the logic is independent of JLine2, and
is in the file gnu/expr/CommandCompleter.java.  It should be
doable (for-example) to have an Emacs frontend call the
CommandCompleter#complete method.
-- 
	--Per Bothner
per@bothner.com   http://per.bothner.com/

Re: tab-completion implemented for Kawa

[Andrea Bernardini]

Hi Per,
This is a very nice feature!

BTW, something I miss in the Kawa REPL is some kind of history. I'm
used to the bash shell, where you can use the up arrow key to get the
previous executed commands. Currently pressing the arrow key in Kawa
inserts a weird sequence of characters. I know that other languages'
REPLs have the same problem (like Racket), however there are some that
support this feature (Python, Haskell). Do you think that could be
doable for Kawa?

All the best

Andrea

(Sorry for the double mail, Per)

On 25 February 2016 at 23:31, Andrea Bernardini <andrebask@gmail.com> wrote:
> Hi Per,
> This is a very nice feature!
>
> BTW, something I miss in the Kawa REPL is some kind of history. I'm used to
> the bash shell, where you can use the up arrow key to get the previous
> executed commands. Currently pressing the arrow key in Kawa inserts a weird
> sequence of characters. I know that other languages' REPLs have the same
> problem (like Racket), however there are some that support this feature
> (Python, Haskell). Do you think that could be doable for Kawa?
>
> All the best
>
> Andrea
>
> (Sorry for the double mail, Per)
>
>
> On 25 February 2016 at 23:27, Andrea Bernardini <andrebask@gmail.com> wrote:
>>
>> Hi Per,
>> This is a very nice feature!
>>
>> BTW, something I miss in the Kawa REPL is some kind of history. I'm used
>> to the bash shell, where you can use the up arrow key to get the previous
>> executed commands. Currently pressing the arrow key in Kawa inserts a weird
>> sequence of characters. I know that other languages' REPLs have the same
>> problem (like Racket), however there are some that support this feature
>> (Python, Haskell). Do you think that could be doable for Kawa?
>>
>> All the best
>>
>> Andrea
>>
>> On 25 February 2016 at 02:12, Per Bothner <per@bothner.com> wrote:
>>>
>>> I checked in an implementation of command-completion (tab-completion)
>>> for Kawa.  This is context-senstive - for example the following works
>>> when using the JLine2 input-ediing library:
>>>
>>> #|Kawa:1|# ab<TAB>s
>>> abs              absolute-path?
>>> #|Kawa:1|# abs<ENTER>
>>> #<procedure abs>
>>>
>>> Kawa wills in the 's' and gives you two options; <ENTER> choose "abs".
>>>
>>> What makes it interesting is that it is content-dependent,
>>> and includes bindings in the lexical scope:
>>>
>>> #|Kawa:2|# (let ((absol 12)) abso<TAB>l
>>> absol            absolute-path?
>>>
>>> It works by running the compiler in tentative mode
>>> with "(let ((absol 12)) abso<COMPLETE>" as the input.
>>> Here <COMPLETE> is a special private-use Unicode character
>>> that tells the Kawa name-lookup framework to look for names
>>> starting with "abso".
>>>
>>> This "tentative mode" could also be useful for on-the-fly
>>> error checking, syntax coloring, use/definition cross-references.
>>> But that is for another day.
>>>
>>> Currently this only works with the JLine2 library.
>>> You have to:
>>>    ./configure --with-jline2=/path/to/jline2.jar
>>>
>>> However, most of the logic is independent of JLine2, and
>>> is in the file gnu/expr/CommandCompleter.java.  It should be
>>> doable (for-example) to have an Emacs frontend call the
>>> CommandCompleter#complete method.
>>> --
>>>         --Per Bothner
>>> per@bothner.com   http://per.bothner.com/
>>
>>
>

Re: tab-completion implemented for Kawa

[Per Bothner]

On 02/25/2016 03:33 PM, Andrea Bernardini wrote:
> Hi Per,
> This is a very nice feature!

Thanks!
>
> BTW, something I miss in the Kawa REPL is some kind of history. I'm
> used to the bash shell, where you can use the up arrow key to get the
> previous executed commands. Currently pressing the arrow key in Kawa
> inserts a weird sequence of characters.

That what the arrow keys are supposed to send.  However, Java doesn't
have a portable pure-Java mechanism to interpret those key sequences.

> I know that other languages'
> REPLs have the same problem (like Racket), however there are some that
> support this feature (Python, Haskell). Do you think that could be
> doable for Kawa?

That's what the --enable-kawa-frontend configure option does.  It compiles
a C program as a front-end to Kawa.  This front-end program uses GNU readline.
There may be some portability issues with the frontend - I haven't made any
effort to port it, and it would certainly need major changes on Windows.

More portable is the JLine2 support I recently announced.  That is a 99%-pure-Java
re-implementation of GNU readline.  That gives you input editing, history,
and (now) tab-completion.

JLine2 may requires some native code on Windows; I haven't tried it under Windows yet.
It should work out-of-the-box on most Unix-like OSes, including GNU/Linux and MacOS.
(It should work on Windows too - I just don't have the instructions for doing that.)
-- 
	--Per Bothner
per@bothner.com   http://per.bothner.com/

Re: tab-completion implemented for Kawa

[Andrea Bernardini]

On Thu, 25 Feb 2016 15:51:34 -0800
Per Bothner <per@bothner.com> wrote:

> On 02/25/2016 03:33 PM, Andrea Bernardini wrote:
> > Hi Per,
> > This is a very nice feature!  
> 
> Thanks!
> >
> > BTW, something I miss in the Kawa REPL is some kind of history. I'm
> > used to the bash shell, where you can use the up arrow key to get
> > the previous executed commands. Currently pressing the arrow key in
> > Kawa inserts a weird sequence of characters.  
> 
> That what the arrow keys are supposed to send.  However, Java doesn't
> have a portable pure-Java mechanism to interpret those key sequences.
> 
> > I know that other languages'
> > REPLs have the same problem (like Racket), however there are some
> > that support this feature (Python, Haskell). Do you think that
> > could be doable for Kawa?  
> 
> That's what the --enable-kawa-frontend configure option does.  It
> compiles a C program as a front-end to Kawa.  This front-end program
> uses GNU readline. There may be some portability issues with the
> frontend - I haven't made any effort to port it, and it would
> certainly need major changes on Windows.

Interesting. I just tried on Debian, it works, thanks

> More portable is the JLine2 support I recently announced.  That is a
> 99%-pure-Java re-implementation of GNU readline.  That gives you
> input editing, history, and (now) tab-completion.

This works too, and with the tab completion indeed, great!