During a WHATWG spec meeting this morning we agreed to not throw for malformed canvas filters. There are still a few editorial changes to go through on that PR, but other than that everything is settled for this launch.
On Friday, October 29, 2021 at 1:53:20 PM UTC-4 dom...@chromium.org wrote: > The usual definition of "breaking change" that we use when shipping > features is not "if web developers always do valid things, then the change > we are making avoids breakages". We can't rely on web developers to write > valid code all the time... see e.g. stats about what percent of the web is > valid HTML. > > For example, while people might not write "asdf", they might write > "colourMatrix" (spot the typo), or similar. > > On Fri, Oct 29, 2021 at 1:50 PM Fernando Serboncini <fs...@chromium.org> > wrote: > >> yes, of course if you pass "asdf" one would throw and the other wouldn't. >> There will be a potential behavior difference. >> >> My point is nobody would write "asdf" as a filter, because "asdf" is not >> a spec'ed or implemented filter anywhere. There are no possible mismatches >> at this point. If CanvasFilter exists, the set of filters is well defined >> everywhere. >> >> On Fri, Oct 29, 2021 at 1:45 PM Domenic Denicola <dom...@chromium.org> >> wrote: >> >>> I don't think that's true. If someone passes "asdf" to the CanvasFilter >>> constructor, changing from not-throwing to throwing would be a breaking >>> change. We'd need to add use counters, etc. to find out if anyone is >>> passing such invalid values. >>> >>> Whereas, if we start off with throwing, we can always remove the >>> throwing behavior later, with no use counters required. >>> >>> On Fri, Oct 29, 2021 at 1:43 PM Fernando Serboncini <fs...@chromium.org> >>> wrote: >>> >>>> Since the current implementation is not throwing, we could reasonably >>>> launch as-is and the add throwing if it is agreed to do so after. >>>> The added risk is the exact same risk that we are arguing to begin with >>>> (i.e., code that uses not-implemented filters without try/catch would >>>> break). But since there are no "not-implemented filters" yet, it may not >>>> be >>>> a problem. >>>> >>>> On Fri, Oct 29, 2021 at 1:34 PM Domenic Denicola <dom...@chromium.org> >>>> wrote: >>>> >>>>> Yep. Reasonable people can disagree on the tradeoffs here. The >>>>> question is whether this is something we want to deadlock over with other >>>>> implementers. >>>>> >>>>> On Fri, Oct 29, 2021 at 12:28 PM Mike Taylor <mike...@chromium.org> >>>>> wrote: >>>>> >>>>>> On 10/29/21 1:23 AM, Yoav Weiss wrote: >>>>>> >>>>>> Hey Domenic! :) >>>>>> >>>>>> On Thu, Oct 28, 2021 at 11:00 PM Domenic Denicola < >>>>>> dom...@chromium.org> wrote: >>>>>> >>>>>>> >>>>>>> FWIW the two HTML editors on the thread (myself and Anne, with our >>>>>>> HTML editor hats on), as well as Mozilla (via Anne with his Mozilla hat >>>>>>> on), prefer the throwing behavior. I think in most cases to overcome >>>>>>> that >>>>>>> position we would need some really strong reasons why the Chromium >>>>>>> project >>>>>>> believes the non-throwing behavior is better. It's not clear to me how >>>>>>> strong Chromium's position is on this issue, and whether it's worth >>>>>>> delaying the feature over. (Or indeed, delaying all the features, since >>>>>>> the >>>>>>> plan seems to be to bundle them together?) >>>>>>> >>>>>> >>>>>> My concerns with the throwing behavior are similar to the ones we >>>>>> have discussed >>>>>> <https://github.com/w3c/mediasession/issues/228#issuecomment-886455386> >>>>>> in the context of MediaSession actions. >>>>>> If we go with the throwing behavior, every future addition of filters >>>>>> would have a significant interop risk, in case adopting developers won't >>>>>> use try/catch properly. If they do that and they are not testing in >>>>>> not-yet-supporting browsers, their apps are likely to break entirely in >>>>>> those browsers. >>>>>> If we go with a silent failure + feature detection approach, >>>>>> developers using the feature without properly detecting it may not have >>>>>> the >>>>>> desired visual effects they are going for, but won't have unrelated >>>>>> parts >>>>>> of their app break. >>>>>> >>>>>> From my perspective (with my API owner hat on), less risk is better, >>>>>> and the second approach seems less risky to me. >>>>>> >>>>>> I agree with Yoav here (sorry, I don't own any hats). Not throwing >>>>>> will likely result in fewer broken pages in less-well-tested browsers >>>>>> that >>>>>> haven't implemented the APIs yet. And +1 for devtools warnings to help >>>>>> developers figure out "silent" failures. >>>>>> >>>>>> (I also wonder if requiring try/catch won't trip up new developers >>>>>> trying to use it inside Promises, who don't yet know about >>>>>> `then()/catch()` >>>>>> patterns). >>>>>> >>>>> -- You received this message because you are subscribed to the Google Groups "blink-dev" group. To unsubscribe from this group and stop receiving emails from it, send an email to blink-dev+unsubscr...@chromium.org. To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/0612b019-71ee-41e9-aa78-aba0712d0a0dn%40chromium.org.
