Again. "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" No. This is been made abundantly clear. The syntax rules have nothing to do with parsing. So all other line of thought based on that faulty premise are flawed.
On Tue, Feb 5, 2013 at 8:07 AM, Caitlin Potter < [email protected]> wrote: > 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]> 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]] >> *Sent:* Monday, February 04, 2013 10:54 PM >> *To:* Caitlin Potter >> *Cc:* Ralph Giles; [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]> 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= >>> [email protected] [dev-media-bounces+caitlin.potter= >>> [email protected]] on behalf of Kyle Barnhart [ >>> [email protected]] >>> Sent: Monday, February 04, 2013 10:21 PM >>> To: Ralph Giles >>> Cc: [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]> 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] >>> > 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
