Flexibility is important because it is the only way for us to allow the common use cases (and beyond) to be implemented.
The algorithm proposed in the draft will not permit other use cases. Not only that, but it will only permit the use case that it's designed for under certain constraints (small files transferred in entirety before playback, or near the beginning of playback). Our flexibility means that we can say "okay, we want to be "standards compliant", so let's remove items that are not correct enough, as the spec says we should". That's simple to implement, and that's really the only business rule that particularly matters. So we an do this at a higher level in the error callback or cue callback, rather than in the parser code itself. We can even add helper functions to perform these checks, to make things easier for the client. What this flexibility means is that we are able to return output like the parser algorithm would, or we can validate syntax, or we can transcode between formats, or we can implement some "correction" tools to "fix" bad input, to some degree, or any number of other use cases. But the one that I'm really interested in is the use of WebVTT for live streams, which is something that is frankly not possible at all under the algorithm suggested in the specification. So basically, the notion that "the parsing section is the only important part of the spec" is mistaken, and in fact, that parsing section is fairly mistaken in and of itself, because of the serious limitations it imposes. Maybe it will be improved in the future, and I hope that it is. But in the mean time, because we need to support other use cases, or different ways of presenting that same browser use case, we need to take the spec as a whole into consideration, rather than just the "parsing section". ________________________________________ From: dev-media-bounces+caitlin.potter=senecacollege...@lists.mozilla.org [dev-media-bounces+caitlin.potter=senecacollege...@lists.mozilla.org] on behalf of Caitlin Potter [[email protected]] Sent: Tuesday, February 05, 2013 8:07 AM To: Kyle Barnhart Cc: [email protected] Subject: RE: WebVTT Parser Standards Compliance You (once again) misunderstand what I've said, Kyle. I don't know how to put it in any simpler terms. The "parsing" section of the specification (that is just ONE single aspect of a very long draft, which relates to an algorithm used when displaying track elements in the browser) The only important thing to do with that parsing algorithm is to ensure that we can output the same thing. Those extra constraints are great and all (although, if we're honest, they're really not very well thought out), but a parsing algorithm is defined by the underlying grammar. The "parsing" section contains some rules, which are easily followed. But forcing the library to follow them all the time means removing some other use cases for the library, and that's frankly just silly. And we can't realistically expect the algorithm they lay out to work in a browser, anyway. The other implementations we've looked at certainly don't. The "parsing section" lays out some rules to follow, so we follow those rules (optionally, where possible). If you think we should implement a second parser for rendering only, that follows the algorithm outlined in the parsing section to the letter, then feel free to contribute code. But I think it will be somewhat difficult to implement that in a re-entrant fashion with small blocks of data at a time, if it's a large file. ________________________________ From: Kyle Barnhart [[email protected]] Sent: Tuesday, February 05, 2013 12:24 AM To: Caitlin Potter Cc: [email protected] Subject: Re: WebVTT Parser Standards Compliance Your first point about standards has already been resolved. There is only one standard by which a WebVTT parser may be judged, and that is the parser rules in the specifications. Any cues the parser outputs which does not meet the standard set in the parsing rules of the specification is by definition non-standards compliant. (see posts by Robert O'Callahan, L. David Baron, and quotes by Glenn Maynard, Simon Pieters, Velmont, Ian Hickson, and Ms2ger) Your statement about flexibility is a good one. Usually flexibility is a good thing, but when implementing a standard it is actually very bad. I only touched on this briefly, so let me explain more fully. The purpose of a standard (such as WebVTT) is to ensure consistent behavior wherever the standardized format is used. This is done so that developers using the format only need to write one version of the object and can expect it to work the same wherever the format is used. This way they do not need to make multiple versions of the object for each setting, nor do they need to write complex code to make sure it behaves the same across settings. Take how HTML renders in browsers for an example. Developers often have to write complex code so that their website displays correctly in different browsers such as IE6. Fortunately this has improved over the last few years. Another example is an image format. If implementations allowed for flexibility where deviation from the standard was allowed, then whoever made the image could not expect it to display the same in different programs. In some programs it may not display at all. This is why image formats have such standards, and is the same reason we cannot deviate from the specifications for WebVTT. This is why I say there should only be one possible output of cues from the parser, which is standards compliant. Thank you, Kyle Barnhart Helpful Links: http://www.w3.org/standards/about.html http://www.webstandards.org/learn/faq/ On Mon, Feb 4, 2013 at 11:18 PM, Caitlin Potter <[email protected]<mailto:[email protected]>> wrote: The parser does (or, lets say, "should") output standards compliant cues. Everything is there: cue ID, start time, end time, the cue settings listed in the draft, and a tree of markup elements from the cue text. The thing that is "not standards-compliant" with regard to the unit tests, is that the test FileParser implementation does not bother to implement higher level rules laid out in the parser section of the spec. This is because we want it to catch as many syntax errors as possible for the input, and read in as much data as possible for the input. This does not mean that the spec is being "ignored", what this means is that this code is designed to be flexible, so that it can be used for different applications with different needs. As you've seen on the github repository, I've created some issues that should help us improve how easy it is to obtain the behaviour laid out in the parsing section. We can likely have that done by the next release, if we decide to do it in the manner that I've proposed. Flexible does not mean "ignore the standard". Nobody on this project has suggested that we ignore or abandon the specification. Not a single person. Flexible means, give clients of the library the power to operate in ways that make sense for the usage they want. The spec doesn't instruct us to return syntax errors, or to be able to parse not-quite-valid input. But because it's not expensive to do these things, and some applications will want to sometimes, then it just doesn't make sense to leave those things out. ________________________________ From: Kyle Barnhart [[email protected]<mailto:[email protected]>] Sent: Monday, February 04, 2013 10:54 PM To: Caitlin Potter Cc: Ralph Giles; [email protected]<mailto:[email protected]> Subject: Re: WebVTT Parser Standards Compliance I have already agreed that I need to break up the patches and be more clear. But that has nothing to do with whether the parser should be allowed to output non-standard compliant cues. Nor does it have anything to do with if there should be one set of tests or two. On Mon, Feb 4, 2013 at 10:46 PM, Caitlin Potter <[email protected]<mailto:[email protected]>> wrote: We agree that there are errors in these tests (however I did not see that one mentioned in your list of changes, which is a good reason to make much smaller, more focused issues rather than massive pull requests) But your changes have done a lot more than just this. You've removed tests claiming that they're duplicates without demonstrating that they're duplicates (eg, claiming that "4 digits" and "greater than 999" are the same thing, with the same code to step through, when they are in fact not). But even if you do demonstrate that the code is the same, for things like that I think it's a good idea to keep those tests in place for now, because they're not actually the same thing. Removing tests, changing the expected number of cues (sometimes correct, other times not so much), removing the comments regarding the syntax rules (which are essentially a guide through the parser code) and replacing them with the "relevant" section from the parser spec... Things like this, I don't agree with. There are a huge number of changed files in your pull requests, I haven't been over all of them, but a few of these things have stood out. It will be a lot simpler if we can avoid these massive patches (I'm guilty of this too) so that it's easier for other people to provide an input on exactly what is correct and what isn't. But nevermind the "big patch" issue for now, the thing is that we all agree that tests for the rules outlined in the "parsing" section are needed. The issue is that some of us believe these tests are different from the other tests (in that they will require a different FileParser implementation). They can still use the same data files, and even sit in the same folder. But some code changes are really necessary to make them test the things you want them to test. ________________________________________ From: dev-media-bounces+caitlin.potter=senecacollege...@lists.mozilla.org<mailto:[email protected]> [dev-media-bounces+caitlin.potter=senecacollege...@lists.mozilla.org<mailto:[email protected]>] on behalf of Kyle Barnhart [[email protected]<mailto:[email protected]>] Sent: Monday, February 04, 2013 10:21 PM To: Ralph Giles Cc: [email protected]<mailto:[email protected]> Subject: Re: WebVTT Parser Standards Compliance There are tests for every valid and invalid input we could think of. Let me show an example of the changes I've made. 00:01.780 --> 00:02.300 That is a timing statement for a cue. The whitespace between the timestamp and the arrow is required by the syntax rules but not required by the parsing rules. So for tests where the whitespace is missing, I've changed the expected number of cues from 0 to 1. 00:02.0005 The milliseconds are only allowed to have 3 digits. In the current tests change it 00:02.005. This is not allowed by either syntax or parsing rules, and by parsing rules the cue should be discarded. So I changed expected cues from 1 to 0. These are only two of the many changes that had to be made to make the tests correct according to the parsing rules. The other big change I made is to reference the parsing rules being tested in the comments instead of the syntax rules, since the syntax rules don't apply to a parser and the parsing rules do. Otherwise I've made no changes to the intent of any test, and I have added many missing tests, and removed duplicate test. I have not removed any debug error checks and have added many missing ones. In all the modified tests are more thorough and make sure the parser is correctly outputs cues that are standards compliant. What Caitlin is now arguing is that the parsing library should have two settings, one outputs non-standard cues, and one outputs standard cues, and there is a set of tests for each. However I can see no possible reason ever to output non-standard cues. In fact is it bad, dangerous, and a whole lot of unnecessary work. The purpose of a standard is to make sure WebVTT behaves the same in all settings. Outputting non-compliant cues directly violates the standard. Allowing it can only serve to make it more difficult to work with WebVTT, and developers will not know how their file will behave from one application to the next. Therefore it is far less work and far better to have one set of standard compliant tests, and fix the parser to those standards. And it is better to do it now, then to go back and re-engineer the thing later. Thanks, Kyle Barnhart On Mon, Feb 4, 2013 at 8:53 PM, Ralph Giles <[email protected]<mailto:[email protected]>> wrote: > On 13-02-04 5:20 PM, Caitlin Potter wrote: > > > The issue here is that Kyle insists on rewriting unit tests that are > concerned with the "syntax specification", rather than adding new tests > that are concerned with the "parser specification". > > Like Chris, I'm a confused what the contended issue is. To be clear, are > we looking at the version of the webvtt spec at > https://dev.w3.org/html5/webvtt/ ? > > This has several sections which seem relevant: > > * 3.1 "Syntax" described the WebVTT text format and includes a few > parser descriptions, such as the one for timestamps. > > * 3.2 "Parsing" describes the top-level parsing algorithm for WebVTT > text files. > > * 3.3 "Cue text parsing" describes a parsing algorithm for the contents > of each cue, building a particular dom-ish data structure. > > Is one of these sections the "syntax specification" and another the > "parser specification"? If so, where do they disagree? Can you give a > specific example where one is more permissive or restrictive than another? > > The point of having a specific parser algorithm documented in the spec > is to achieve uniform implementation. If everyone implements the > algorithm (or its equivalent) the handling of edge cases will be more > consistent than if everyone implements a parser from scratch based on > some incomplete syntax description. So we should be implementing the > parser the spec describes. If the spec is internally inconsistent we > should file spec bugs and get the text fixed. > > Nevertheless, the current code doesn't pass--or even run to completion > on--a number of the current tests, so it's difficult to tell what works > and what doesn't. I think fixing that should be the highest priority for > those working on the parser. Without tests we don't know where we stand, or > > Kyle, Caitlin's suggestion that you provide a separate set of parser > tests seems reasonable to me if she wants the current set for code > coverage or additional features. The test sets can always be merged > later if there's consensus that's appropriate. In the meantime you won't > get in each other's way. > > - Ralph > _______________________________________________ > dev-media mailing list > [email protected]<mailto:[email protected]> > https://lists.mozilla.org/listinfo/dev-media > _______________________________________________ dev-media mailing list [email protected]<mailto:[email protected]> https://lists.mozilla.org/listinfo/dev-media _______________________________________________ dev-media mailing list [email protected] https://lists.mozilla.org/listinfo/dev-media _______________________________________________ dev-media mailing list [email protected] https://lists.mozilla.org/listinfo/dev-media
