[sane-devel] SANE2 standard completion

[m. allan noah]

On Fri, Mar 28, 2008 at 2:28 PM, Rene Rebe <rene at exactcode.de> wrote:
>
>
>  On 28.03.2008, at 19:19, m. allan noah wrote:
>
>
> > On Fri, Mar 28, 2008 at 2:14 PM, Rene Rebe <rene at exactcode.de> wrote:
> >
> > >
> > > On 28.03.2008, at 19:02, Alessandro Zummo wrote:
> > >
> > >
> > > > On Fri, 28 Mar 2008 15:49:17 +0100
> > > > Rene Rebe <rene at exactcode.de> wrote:
> > > >
> > > >
> > > > >
> > > > > Right now the only thing I miss are marks for infra-red frames, and
> > > > > maybe
> > > > > a ability to pass duplex data without buffering the rear side, which
> > > > > becomes
> > > > > even more of a problem with the next generation, higher end devices
> I
> > > > > got
> > > > >
> > > >
> > > > I'm interested in this.. you receive the data of two pages at
> > > > the same time? It's interleaved in some way?
> > > >
> > >
> > > I speak about the duplex devices here, which have 2 CCDs (or
> > > similar and scan both sides at the same time (long supported by
> > > the Avision backend, btw. :-)
> > >
> > > Avision developed several ways to transfer the duplex data (aside
> > > from the few, trivial) paper flipping devices with just a single CCD),
> > > for some they buffer the page in-memory inside the scanner, for newer
> > > ones there are several interlaced variants.
> > >
> > > For the interlaced variants I so far buffer the rear stripes in the
> > > backend
> > > and return the backed data every second frame.
> > >
> > > Probably the fujitsu backend does something similar.
> > >
> > >
> >
> > yes- fujitsu is just the same, grep fujitsu.c for COLOR_INTERLACE_*,
> > you will see several variants.
> >
> > my (as yet unreleased) backend for the big kodak machines does not
> > have this problem, because the machines have tons of ram, and buffer
> > the images themselves, and there is no way to stop it :)
> >
>
>  WIth "no way to stop", you mean no way to stop the processing?

no- i mean no way to tell scanner that you want it to interlace front and back.

>  Because
>  that is also a point that could get improvements in SANE, the backend
>  should these days better push the data to the frontend, like begin page,
>  data, end page. As the bigger machines tend to deliver data so fast they
>  only honor a cancel with n++ pages in memory or on the network wire ...
>  (or not at all).

this is true of the bigger machines, it might be 20 or 30 pages ahead
of the wire when the cancel call comes in, especially if you are not
using a compressed image type.

for your other points, i dont really see how push is any better than
blocking pull...

allan
-- 
"The truth is an offense, but not a sin"