- From: Eduard Pascual <herenvardo@gmail.com>
- Date: Wed, 7 Apr 2010 01:02:40 +0200
- To: gesteehr@googlemail.com
- Cc: public-html-comments@w3.org
On Tue, Apr 6, 2010 at 5:00 PM, Georg <gesteehr@googlemail.com> wrote: > Hello, > > I’ve a new idea for the HTML 5 specification. > > If you want to present HTML-code in a browser, you have to write < > instead of <. That's not entirely true, see below. > My Idea is, to include a type-attribute in the code-tag. > > If the type is “html” or “xml” webmaster don’t have to write <. They can > simply write <. > > Than the browser should show <. > > If you have questions for my idea, you can ask me. The idea itself is not bad; but it has important drawbacks: First, the code added there would need to be parsed, so if a <code> is included there it will be paired with the corresponding </code>. This then makes the page itself very brittle: any mistake in the code intended to display may have very nasty side-effects (most prominently, too many or too few </code>'s would miserably break the page down to a nonsensical mess). In addition, it may create a serious security vulnerability on sites that allow users to provide content (discussion boards and blogs accepting comments are among the best known examples, but there are many others; for example, the archives of this very mailing list will render the content I'm writing now as HTML). Imagine that, on a phpBB-style board, a malicious user posts a message containing "[code=html]</code><script>..." ([code]...[/code] is phpBB's own tag to introduce a code snippet on the post). It may seem natural to turn the BBtag into a <code type="html"> element; but it would be closed by the initial "</code>" and then the <script> would be executed... it may check for phpBB login cookies and harvest user's names and passwords; and that's just a possible attack from the top of my head, so imagine what someone actually evil might come up with. Of course, that's just a matter of input sanitizing; but is there any real need to open a new gate for injection-based attacks? Also, note that with your proposal sanitizing is quite more than trivial: a code snippet may include a legitimate </code> if it also included matching <code>'s; so rather than just escaping or breaking instances of a specific text, you need to parse the whole input to infer the structure. And, btw, poor design or simple bugs on such a parser can open the gate to DoS attacks. In any case, on XHTML documents, your problem is already addressed by a feature from XML itself: <[CDATA[ ... ]]>. This is far easier to sanitize (you just need to ensure that the input doesn't include the "]]>" sequence), thus being more usable on user-provided content. The only drawback of it is that "non-X" HTML doesn't support it (except within MathML and SVG content; but that doesn't address the use case). So, the solution you propose is quite broken, but your use case (presenting HTML code samples within HTML pages in a way that is saner to author and maintain) is a quite good one. Because of that, I'd like to propose allowing <[CDATA[ ... ]]> generally. I'm sure there has been some reason it isn't that allowed yet; but it may be worth reviewing the reasons for that in the arise of a use case. The main drawback I can think of is compatibility; but it may be fine to wait for older browsers to die off before relying on this; and I don't expect for legacy content to break with a change like this (after all, anyone who ever types "<[CDATA[" most probably knows that the leading "<" should be escaped as "<" for it to render as text. Also, trying to look through the list archive for discussions on the topic has been quite fruitless (searching for "CDATA" yields too many unrelated results; while searching for "<[CDATA[" to refine the search yields an ugly error); so if this has been discussed before I hope some of the "veterans" in the list can at least point to the discussions or summarize the issues. Regards, Eduard Pascual
Received on Tuesday, 6 April 2010 23:03:28 UTC