- From: Arthur Clifford <art@artspad.net>
- Date: Tue, 6 Apr 2010 19:06:20 -0700
- To: <public-html-comments@w3.org>
Well, from the general perspective of importance, this is pretty much a non issue because things can be escaped and everything does it. If this were a discussion about optimizing html, then I'd say that the need to escape and unescape text before and after every request is generally a waste of processing ticks if there is a way to avoid it. Granted, these days there's more than enough processing ticks to spare. But at some point everybody counted on their fingers, that doesn't mean you don't consider a calculator. Then, there's the situation where eventually someone will say, hey I want to include html that has sample xml in my html cdata tag but my sample xml has cdata in it and so it has a closing cdata tag. At which point you are back to the same problem of having to escape and unescape syntax. Whereas with content length you could include very complex xml/html/javascript/whatever and it wouldn't matter because the parser would jump to the end of the content and know whatever it got is a blob of text to display. I wouldn't mind an actual cdata tag over the <![CDATA[ hacky tag. Something along the lines of: <cdata contentLength="..." mime="text/javascript">...</cdata> You know, an official tag you can do thigns like apply styles to, provide an id for, and have javascript access to the content of. However it beign cdata, the mime type would not tell a user agent to render the text as something other than text, it would just give the user agent options for formatting the text. If not content length, then perhaps a delimiter attribute: <cdata delimiter="***" >Some really awesome marked up text ***</cdata> The problem with deducing the end of a data chunk, and not knowing what you have in it, is that the user agent has to deduce it and whatever marker you are using to deduce the end has to be escaped or not included. If the parser knows how long the data chunk is, it doesn't have to deduce anything and thus you have full freedom. The delimiter technique would give the parser something to read ahead for that could be a sequence you can simply not include in the data. As far as mime typing, the more the user agent knows about what it renders, even if it is to be treated as plain text, the more flexibility it has to do something interesting. In terms of processing, every language has its strlen or String.length() equivalent and it should be a lot less processing to include that info in a tag attribute than it is to do a search and replace on every special character that might choke the html parser .. twice (during storage and during display)! Art Arthur Clifford -----Original Message----- From: Eduard Pascual [mailto:herenvardo@gmail.com] Sent: Tuesday, April 06, 2010 6:35 PM To: art@artspad.net Cc: public-html-comments@w3.org Subject: Re: HTML 5 On Wed, Apr 7, 2010 at 2:10 AM, Arthur Clifford <art@artspad.net> wrote: > If this is something that would be for code added more programmatically IMO, that's the least important use-case: as far as I know, most programming/scripting languages used on the web (both client- and server- side) have some facility that turns the escaping taskt into something trivial (like PHP's htmlspecialchars() function). Even when such a facility is not available, the task is quite simple with raw text-based replacing utilities (just remember to do "&"=>"&" first, so you don't re-escape the &'s that are inserted as part of the other escapes). The CDATA approach is a more general solution. > what > about something as simple as a content length attribute on the <code> or > <pre> tag that would allow you to specify how many bytes of data are between > the beginning and close tag? This could easily become more painful than actually escaping the content if you ever need to manually edit it; so you are reducing the use-case even further to only those cases where the content is only handled programatically. Not a very good "solution" if it only solves a probably small fraction of the problems. > Then you wouldn't need to worry about special > characters at all. The User agent would read in x number of characters, > would know the next thing it should hit is </ and if it doesn't then it can > throw a validation error. Then a single byte is lost due to a bad connection and, instead of just missing a character, the whole page breaks miserably. This doesn't suit very well in HTML5's "make sure even the craziest tag-soup renders as close to the author's intent as possible" philosophy. An UA could try to play smart on these cases, but it risks to messing things up. What about if the lost byte was part of the length attribute? Part of the snippet would be parsed as actual page code, which leads to injection issues again (it's not an intentional attack, but code being injected by pure randomness on a browser isn't something vendors will be happy to implement into their UAs). > The tags could also have a mime attribute so that > the user agent when rendering code/preformatted text could color code > syntax. With that W3 could work toward a code hinting standard and CSS for > code hinting for HTML6. While I think this is interesting, it happens to be entirely unrelated with the problem/use-case being discussed and with the potential solution you proposed. I would invite you to branch it to a new thread, but the discussion would essentially boil down to the fact that, on HTML's part, syntax highlighting could already be applied with data-* attributes plus CSS attribute selectors; so there is no point on adding further hooks for as long as CSS doesn't provide anything for syntax highlighting. If you still want to discuss this, then go ahead and start a new thread on the topic. (From my own experience, I can assure you that trying to discuss two separate topics on the same thread can easily become nearly insane). > I've never been fond of the cdata tag syntax If there is any technical issue with the syntax itself (your personal dislike may be an issue, but if it's "personal" then it's not "technical"), HTML5 could define a different syntax for the same task (after all, <![CDATA[ ... ]]> is not part of HTML, it's only part of XML, that's why it's readily available for XHTML). From the top of my head, things like "<< ... >>", "<[[ ... ]]>", "<! ... !>" could work (actually, the last would quite clash with HTML's legacy inheritance from SGML, such as the <!-- --> comment syntax; but these are just examples). > and have felt that xml elements > should have a contentLength attribute. If you have use-cases with specific requirements that are addressed by such a feature, go ahead and start a new thread to discuss the idea. If you can't materialize that "feeling", however, I wouldn't put much hopes on it being taken into account by the editor. > > Of course I also think there should be a BEXML (Binary Enabled XML) standard > that allows traditional XML markup with the addition of a BData tag that > includes a content length attribute. In thinking about that, it dawned on me > that you could have data with any characters you want with the combination > of content length and mime type. That's another feature suggestion without a hint on the use-cases or requirements. Same comments as above. However, that one sounds interesting; so if you have use-cases to warrant it some discussion be assured I'm going to be closely following the topic :P > But I digress. Anything read in for code or pre text should be treated as > read and display with no processing. It could be read in and displayed as > text, but it WOULD be nice not to have to worry about escaping characters > when saving or returning data. Also, if someone put php or other server-side > code in that didn't result in output that is the same length after the php > processing then the content length would be wrong and the page would not, or > should not, load correctly. I already mentioned it on my reply to the OP, but the key issue here is: if you ignore the markup while reading the contents of a given element, how do you know where the element is? If you are watching only for the specific "</code>" tag, how do you differentiate that one from a pair "<code> ... </code>" within the snippet? > > Art > > Arthur Clifford
Received on Wednesday, 7 April 2010 02:10:48 UTC