[racket] Cleaner way to work with gzipped data?

From: Robby Findler (robby at eecs.northwestern.edu)
Date: Thu Aug 8 12:10:07 EDT 2013

As to the interactions with dynamic-wind and continuations: l did, as you
figured out, intend for you to use only the post-thunk in the dynamic-wind
(to close the pipes). In principle, you could use the pre-thurnk to try to
restore the pipe, but there really isn't enough information to do this
correctly in all cases.

I see that you've protected things a little bit by using the port-closed?
predicate as a guard, but if you did that to protect against possible
continuation re-entry, then probably you're better off adding something to
the pre-thunk that explicitly raises an error saying that it isn't allowed
to re-enter. Something like this:

#lang racket

(define (only-once thunk)
  (define already-in-once? #f)
   (λ ()
     (when already-in-once? (error 'only-once "no no")))
   (λ ()
     (set! already-in-once? #t)

(only-once (λ () "hi"))

(let ([saved-k #f])
  (only-once (λ () (let/cc k (set! saved-k k))))
  (saved-k 11))

On Tue, Aug 6, 2013 at 11:16 AM, JP Verkamp <racket at jverkamp.com> wrote:

> I've never actually used dynamic-wind, although it does look interesting
> / like what I need. A few questions / caveats though:
> - Should the pipe be created in the pre-thunk or before the dynamic-windentirely? The thunks don't seem to share scope, so I'm guessing the latter,
> but that seems a bit odd. I'm guessing the pre-thunk is for an entirely
> different use case though when you are actually dealing with closing and
> reopening resources are the like as control gets passed around.
> - Doesn't dynamic-wind break if the user messes with continuations during
> the value-thunk? So far as I understand, when control passes out,
> post-thunk is called and then pre-thunk on the way back in, but that
> means that when control returns the port will be closed. I don't know how
> often this will come up, but it seems to break if I nest a thread inside of
> the with-gzip call. Granted, my version did as well because of the
> close-input-port call. Is this just expected behavior?
> (And yes, it works fine in the more likely / sensible case of wrapping the
> entire with-gzip in a thread in both cases.)
> - So far as error rather than raise, raise was my original guess. But
> that added another layer of indirection to the stack trace which I didn't
> at first notice (I thought I wasn't even catching the error). It makes
> sense to have that though in the long run.
> That all being said, how does this version look?
> (define (with-gunzip thunk)
>   (define-values (pipe-from pipe-to) (make-pipe))
>   (dynamic-wind
>    void
>    (λ ()
>      (gunzip-through-ports (current-input-port) pipe-to)
>      (close-output-port pipe-to)
>      (parameterize ([current-input-port pipe-from])
>        (thunk)))
>    (λ ()
>      (unless (port-closed? pipe-to) (close-output-port pipe-to))
>      (unless (port-closed? pipe-from) (close-input-port pipe-from)))))
> On Tue, Aug 6, 2013 at 11:47 AM, Robby Findler <
> robby at eecs.northwestern.edu> wrote:
>> You might consider using dynamic-wind instead of that with-handlers. Or,
>> instead of (error 'with-gunzip ...) just do (raise exn). That way you won't
>> lose the stack information in the original exception (which is likely the
>> one a user would want).
>> Robby
>> On Tue, Aug 6, 2013 at 10:40 AM, JP Verkamp <racket at jverkamp.com> wrote:
>>> Figured it out and cleaned it up. It turns out that I was using
>>> with-handlers oddly, but reading further though the documentation it
>>> works as expected. Here's a new version (generalized to any input-port):
>>> (define (with-gunzip thunk)
>>>   (define-values (pipe-from pipe-to) (make-pipe))
>>>   (with-handlers ([exn:fail?
>>>                    (λ (err)
>>>                      (close-output-port pipe-to)
>>>                      (close-input-port pipe-from)
>>>                      (error 'with-gunzip (exn-message err)))])
>>>     (gunzip-through-ports (current-input-port) pipe-to)
>>>     (close-output-port pipe-to)
>>>     (parameterize ([current-input-port pipe-from])
>>>       (thunk))
>>>     (close-input-port pipe-from)))
>>> If anyone's interested in a more in depth write up / source code for
>>> this and with-gzip:
>>> - writeup:
>>> http://blog.jverkamp.com/2013/08/06/adventures-in-racket-gzip/
>>> - source:
>>> https://github.com/jpverkamp/small-projects/tree/master/blog/with-gzip.rkt
>>> On Mon, Aug 5, 2013 at 5:36 PM, JP Verkamp <racket at jverkamp.com> wrote:
>>>> Thanks! make-pipe isn't something that I've had to use otherwise, so I
>>>> missed the optional parameter. That does certainly seem to help.
>>>> Here's my first take of with-input-from-gzipped-file:
>>>> (define (with-input-from-gzipped-file filename thunk #:buffer-size
>>>> [buffer-size #f])
>>>>   (call-with-input-file filename
>>>>     (lambda (file-from)
>>>>       (define-values (pipe-from pipe-to) (make-pipe buffer-size))
>>>>       (thread
>>>>           (λ ()
>>>>             (gunzip-through-ports file-from pipe-to)
>>>>             (close-output-port pipe-to)))
>>>>       (current-input-port pipe-from)
>>>>       (thunk)
>>>>       (close-input-port pipe-from))))
>>>> The main thing missing is that there's no error handling (where the
>>>> pipe should still be closed). At the very least, if I try to call this on a
>>>> non-gzipped file, it breaks on the gunzip-through-ports line.
>>>> Theoretically, some variation of with-handlers should work (errorshould raise an
>>>> exn:fail?, yes?), but it doesn't seem to be helping.
>>>> Any help with that?
>>>> Alternatively, I've now found this:
>>>> http://planet.racket-lang.org/display.ss?package=gzip.plt&owner=soegaard
>>>> It seems to do exactly what I need, albeit without the call-with-*
>>>> forms, but that's easy enough to wrap. With some very basic testing, it
>>>> does seem to be buffering though, although it is a bit slower than the
>>>> above. Not enough to cause trouble though.
>>>> On Mon, Aug 5, 2013 at 4:51 PM, Ryan Culpepper <ryanc at ccs.neu.edu>wrote:
>>>>> On 08/05/2013 04:29 PM, JP Verkamp wrote:
>>>>>> Is there a nice / idiomatic way to work with gzipped data in a
>>>>>> streaming
>>>>>> manner (to avoid loading the rather large files into memory at once).
>>>>>> So
>>>>>> far as I can tell, my code isn't doing that. It hangs for a while on
>>>>>> the
>>>>>> call to gunzip-through-ports, long enough to uncompress the entire
>>>>>> file,
>>>>>> then reads are pretty quick afterwords.
>>>>>> Here's what I have thus far:
>>>>>> #lang racket
>>>>>> (require file/gunzip)
>>>>>> (define-values (pipe-from pipe-to) (make-pipe))
>>>>>> (with-input-from-file "test.rkt.gz"
>>>>>>    (lambda ()
>>>>>>      (gunzip-through-ports (current-input-port) pipe-to)
>>>>>>      (for ([line (in-lines pipe-from)])
>>>>>>        (displayln line))))
>>>>> You should probably 1) limit the size of the pipe (to stop it from
>>>>> inflating the whole file at once) and 2) put the gunzip-through-ports call
>>>>> in a separate thread. The gunzip thread will block when the pipe is full;
>>>>> when your program reads some data out of the pipe, the gunzip thread will
>>>>> be able to make some more progress. Something like this:
>>>>> (define-values (pipe-from pipe-to) (make-pipe 4000))
>>>>> (with-input-from-file "test.rkt.gz"
>>>>>   (lambda ()
>>>>>     (thread
>>>>>       (lambda ()
>>>>>         (gunzip-through-ports (current-input-port) pipe-to)
>>>>>         (close-output-port pipe-to)))
>>>>>     (for ([line (in-lines pipe-from)])
>>>>>       (displayln line))))
>>>>>  As an additional problem, that code doesn't actually work.
>>>>>> in-lines seems to be waiting for an eof-object? that
>>>>>> gunzip-through-ports isn't sending. Am I missing something? It ends up
>>>>>> just hanging after reading and printing the file.
>>>>> The docs don't say anything about closing the port, so you'll probably
>>>>> have to do that yourself. In the code above, I added a call to
>>>>> close-output-port.
>>>>> Ryan
>>> ____________________
>>>   Racket Users list:
>>>   http://lists.racket-lang.org/users
> ____________________
>   Racket Users list:
>   http://lists.racket-lang.org/users
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.racket-lang.org/users/archive/attachments/20130808/f9081efe/attachment-0001.html>

Posted on the users mailing list.