[drakma-devel] File uploading: adding streams and functions to the list of parameters

[Edi Weitz]

On Fri, 8 Jun 2007 11:19:43 -0400, "Andrei Stebakov" <lispercat at gmail.com> wrote:

> Recently I've been thinking how to make http-request upload a file,
> but not from the hard drive but from a function which can produce
> (or write to a) stream.
> Let's say I want to create an image with cl-gd but instead of
> writing the image to the file and then creating the http-request
> with a pathname of the file I can pass a
> (lambda (stream)
>    (cl-gd-create-image stream your param1 param2))
> in place of the pathname parameter. Or a stream which already has
> the file data.
> I was trying to do it using :content :continuation but found it too
> difficult, maybe I missed other ways.

You could have passed a function designator to :CONTENT.  That's a bit
easier than to use :CONTINUATION, but you'd still have to create the
form-data wrappers yourself.

> Anyway, I created a patch with very minimal changes to the
> make-form-data-function function which allows to do so. Also I found
> that I set *header-stream* to not nil, I was missing the debug
> output produced by the character (character only) stream from
> make-form-data-function which basically shows all the parameters we
> are sending. I added the primitive support for this based on
> make-broadcast-stream.  Here I attach the patch for your kind
> consideration.

The general idea is fine with me.  As for the patch I have a few
things I don't like:

1. Now that the value can be more than a pathname, the variable
   shouldn't be called PATHNAME anymore.

2. The docstring for HTTP-REQUEST and the HTML documentation need to
   be updated as well.  See also

     http://weitz.de/patches.html

3. The patch contains TAB characters.

4. Where you have *STANDARD-OUTPUT*, you probably want to have
   *HEADER-STREAM*.

5. But I generally don't really like this part anyway.  The stream is
   called *HEADER-STREAM* because the /headers/ are supposed to be
   sent there.  Sending /parts/ of the form-data there as well might
   be useful in a few cases but it seems like a fortuitous choice to
   me.

Thanks,
Edi.