Re: [libmicrohttpd] MHD_queue_response failing

[Christian Grothoff]

Yes, the access handler being called 3+ times is normal for an upload/POST:

1st time, you have the chance to override 100 CONTINUE (i.e. with a 
40x-message to prevent the browser from doing an upload at all). So here 
queueing a reply is OK, but it will prevent the upload.

2nd-N-1th time, you get the upload data (in fact, the upload data may be 
given to you many times, so this can be called repeatedly with more 
data, depending on the upload size). Queueing here is NOT ok, 
upload-size is non-zero.

N-th time (often 3rd for small data), you are supposed to generate the 
'normal' response to the browser, that the upload is done is indicated 
by upload-size being (again) zero.

As I said, slightly goofy API (and pretty much the only thing I'd likely 
do differently if backwards-compatibility didn't matter, but it is not 
so bad that I'd break compatibility to fix it).

Happy hacking!

Christian

On 09/18/2011 01:17 PM, Peter Ross wrote:
> On Sun, Sep 18, 2011 at 08:03:00AM +0200, Christian Grothoff wrote:
> > Dear Peter,
> >
> > I suspect you're simply queueing your response too early.  The API
> > may not be 100%-intuitive here: if you have an upload (and did not
> > replace the 100-CONTINUE with an error), you need to wait for MHD to
> > call you with an '*upload_size' of zero before giving your response.
> >  From what you write about the state we're at, your code does not do
> > this.
>
> Thanks for the speedy response. Your suggestion works! Note that in my
> example, AccessHandlder is called *three* times per POST request. The
> first time upload_data_size is zero, second time its non-zero, and then
> finally its zero again.
>
> -- Peter
> (A907 E02F A6E5 0CD2 34CD 20D2 6760 79C5 AC40 DD6B)