Re: Shell-style programming in Kawa

[Per Bothner <per at bothner dot com>]

This message is about setting/getting the current path.

On 01/27/2014 05:06 PM, Jamison Hope wrote:

> > Similarly, a Scheme interface to the Path.currentPath stuff, like
> >
> > (chdir "..") => Path.setCurrentPath(Path.currentPath().resolve("..").getCanonical());
> > (cwd) => Path.currentPath();
> >
> > could be nice.  Scsh has them, along with a function version of with-cwd that
> > takes a thunk as its second argument.

I implemented (current-path) asa parameter.

Thus instead of cwd you have currentpath; instead of (chdir "..")
you can use a setter: (set! (current-path) "..") or
(current-path "..").

> (define-syntax with-cwd
>    (syntax-rules ()
>      ((with-cwd wd e ...)
>       (let ((old-path (cwd)))
>         (try-finally
>          (begin (chdir wd) e ...)
>          (chdir old-path))))))
>
> (define (with-cwd* wd thunk)
>    (let ((old-path (cwd)))
>      (try-finally
>       (begin (chdir wd) (thunk))
>       (chdir old-path))))

We could add these as syntactic sugar, but since
parameterize now works I think that is good enough:

(parameterize ((current-path (path "..")))
   &sh{echo pwd is `pwd`})

I'm still pondering whether the initial (current-path) - i.e. the
value of Path.useDirPath should be "." or the expanded user.dir.
Certainly it seems weird and uninformative for (current-path) to
just print ".".  But perhaps that would be ok if we added a
(path-absolute PATH) function.


> The resolve-uri call in chdir depends upon the patch to Path.java I just sent.
> Without that, this works instead: (resolve-uri (path (->String dir)) (cwd)).
> I tried just capitalizing the s in string, not wrapping it in (path ...),
> but that gives me a spurious warning:
>
> > warning - no known slot 'getCanonical' in java.lang.Object
>
>
> It seems like the ->string version ends up calling Path#resolve(Path),
> while the ->String version (without the wrap) calls Path#resolve(String),
> which is abstract.  Does the compiler ignore the declared return type
> of an abstract method when doing its type inferencing?

I believe so. Perhaps we need to do more "target typing" in the Java 8 
world.


> [Would the compiler be able to inline away the lambda if with-cwd were defined
> as
>
> (define-syntax with-cwd
>    (syntax-rules ()
>      ((with-cwd wd e ...) (with-cwd* wd (lambda () e ...))))))
>
> or does that require a corresponding Procedure.validateApplyKey property on
> with-cwd*?

The latter.  There is currently no mechanism to automatically inline
previously-compiled functions.  It would certainly be handy.

Can those be implemented in/for functions written in Scheme?]

Yes.  See with-exception-handler for an example where the hook is
done in Scheme. Though the actual validateApplyWithExceptionHandler
method is written in Java, I can't think of a reason it couldn't
be written in Scheme.  It would help to have a small library of
helper functions - and macros - for munging Expression trees.
--
	--Per Bothner
per@bothner.com   http://per.bothner.com/