W3C home > Mailing lists > Public > public-fx@w3.org > July to September 2013

Re: [filter-effects] url() as pass-through on invalid reference?

From: Dirk Schulze <dschulze@adobe.com>
Date: Tue, 27 Aug 2013 00:53:32 -0700
To: Stephen White <senorblanco@chromium.org>
CC: Boris Zbarsky <bzbarsky@mit.edu>, "www-style@w3.org" <www-style@w3.org>, "public-fx@w3.org" <public-fx@w3.org>
Message-ID: <453A16A0-C48A-45C5-8B73-EEF4C0C75A2A@adobe.com>
On Jul 25, 2013, at 2:13 PM, Stephen White <senorblanco@chromium.org> wrote:

> One related case is that even a valid filter with a url() could be pending a network load. If we do a "best effort" render (render the valid parts of the chain), then a mix of shorthand filters mixed with a url() filter, e.g.
> 
> filter: blur(4px) url(foo.svg#bar) contrast(0.5);
> 
> will render progressively. I.e., it'll render the blur and contrast, and then re-render once the network fetch is done. This is likely how it works in Chrome/Safari, but I'm not sure it's the best behavior.

Even if not pleasant, I think this is the better behavior. Imagine you have two SVG Filter references in a chain

	filter: url(#invalid) url(#valid);

Given that #invalid does not reference an SVG Filter and #valid does, then the first reference would produce a 'null filter' which is the input of the second filter reference. This would proceed as normal but with a transparent black image a SourceGraphic. I think in this case you would prefer the unfiltered object as input of the second filter reference instead.

Therefore, I suggest changing the current behavior and make invalid filter reference a pass trough instead of a null filter.

Greetings,
Dirk


> 
> Stephen
> 
> On Sun, Jul 14, 2013 at 10:20 AM, Dirk Schulze <dschulze@adobe.com> wrote:
> 
> On Jul 12, 2013, at 2:31 PM, Boris Zbarsky <bzbarsky@MIT.EDU> wrote:
> 
> > On 7/12/13 5:25 PM, Robert O'Callahan wrote:
> >> I feel pretty strongly that the SVG behavior is not appropriate. I think
> >> any sort of invalid 'filter' value, including a <filter> element with a
> >> child element of unknown type, should cause the filtered element to be
> >> rendered normally (i.e. 'filter' treated as 'none'). Otherwise I think
> >> introducing new <fe> SVG element types does not get useful fallback.
> >
> > OK, should that be raised as an issue on SVG, then?  I feel like we
> > should have consistency between the case when there is only one url() in
> > 'filter' and the case when there are multiple things including a url()
> 
> I forgot to paste the link to the Filter Effects specification that allows multiple declarations on url() [1]. Sorry for that.
> 
> 1) SVG 1.1 was not clear about what happens if the specified URL is not referencing an existing <filter> element. Implementations treat this case as if 'none' was specified.
> 2) For SVG 1.1: if one of the primitives is in an error state (which used to happen quite often in earlier versions of the spec.) then nothing gets rendered at all.
> 
> I agree with you Boris that we should not treat both cases differently. However, I am more in favor of just not applying the filter then stop rendering of the filtered element entirely (suggestion from roc).
> 
> With Filter Effects we have a new situation since we allow multiple filter functions or url()s. The reason why it changes the way to look at the topic is the following:
> 
> A) Treat errors as "filter: none" and refuse the whole filter chain
> 
> might be described similarly with
> 
> B) Treat failing filter function as pass-through and still apply the other filter functions.
> 
> Both, A) and B) just make a difference on applying multiple filter functions if we agree on the behavior of 1).
> 
> Greetings,
> Dirk
> 
> [1] http://dev.w3.org/fxtf/filters/#FilterProperty
> 
> 
> CCed public-fx again.
> 
> >
> > -Boris
> >
> 
> 
> 
Received on Tuesday, 27 August 2013 07:54:01 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 19:49:46 UTC