Is it acceptable for invalid XHTML? - xhtml

I've noticed a lot of sites, SO included, use XHTML as their mark-up language and then fail to adhere to the spec. Just browsing the source for SO there are missing closing tags for paragraphs, invalid elements, etc.
So should tools (and developers) use the XHTML doctype if they are going to produce invalid mark up? And should browsers be more firm in their acceptance of poor mark-up?
And before anyone shouts hypocrite, my blog has one piece of invalid mark-up involving the captha (or it did the last time I checked) which involves styling the noscript tag.

There are many reasons to use valid markup. My favorite is that it allows you to use validation as a form of regression testing, preventing the markup equivalent of "delta rot" from leading to real rendering problems once the errors reach some critical mass. And really, it's just plain sloppy to allow "lazy" errors like typos and mis-nested/unclosed tags to accumulate. Valid markup is one way to identify passionate programmers.
There's also the issue of debugging: valid markup also gives you a stable baseline from which to work on the inevitable cross-browser compatibility woes. No web developer who values his time should begin debugging browser compatibility problems without first ensuring that the markup is at least syntactically valid—and any other invalid markup should have a good reason for being there.
(Incidentally, stackoverflow.com fails both these tests, and suggestions to fix the problems were declined.)
All of that said, to answer your specific question, it's probably not worthwhile to use one of the XHTML doctypes unless you plan to produce valid (or at least well-formed) markup. XHTML's primary advantages are derived from the fact that XHTML is XML, allowing it to be processed and transformed by tools and technologies that work with XML. If you don't plan to make your XHTML well-formed XML, then there's little point in choosing that doctype. The latest HTML 4 spec will probably do everything you need, and it's much more forgiving.

We should always try to make it validate according to standards. We'll be sure that the website will display and work fine on current browsers AND future browsers.

I don't think that, if you specify a doctype, there is any reason not to adhere to this doctype.
Using XHTML makes automated error detection easy, every change can be automatically checked for invalid markup. This prevents errors, especially when using automatically generated content. It is really easy for a web developer using a templating engine (JSP, ASP.NET StringTemplate, etcetera) to copy/paste one closing tag too little or too many. When this is your only error, it can be detected and fixed immediately. I once worked for a site that had 165 validation errors per page, of which 2 or 3 were actual bugs. These were hard to find in the clutter of other errors. Automatic validation would have prevented these errors at the source.
Needless to say, choosing a standard and sticking to it can never benefit interoperability with other systems (screen scrapers, screen readers, search engines) and I have never come across a situation where a valid semantic XHTML with CSS solution wasn't possible for all major browsers.
Obviously, when working with complex systems, it's not always possible to stick to your doctype, but this is mostly a result of improper communication between the different teams developing different parts of these systems, or, most likely, legacy systems. In the last case it's probably better to isolate these cases and change your doctype accordingly.
It's good to be pragmatic and not adhere to XHTML just because someone said so, regardless of costs, but with current knowledge about CSS and browsers, testing and validation tools, most of the time the benefits are much greater than the costs.

You can say that I have an OCD on XHTML validity. I find that most of the problems with the code not being valid comes from programmers not knowing the difference between HTML and XHTML. I've been writing 100% valid XHTML and CSS or a while now and have never had any major rendering problems with other browsers. If you keep everything valid, and don't try anything too exotic css wise, you will save yourself a ton of time in fixes.

I wouldn't use XHTML at all just to save myself the philosophical stress. It's not like any browsers are treating it like XHTML anyway.
Browsers will reject poor mark-up if the page is sent as application/xhtml+xml, but they rarely are. This is fine.
I would be more concerned about things like inline use of CSS and JavaScript with Stack Overflow, just because they make maintenance harder.

Though I believe in striving for valid XHTML and CSS, it's often hard to do for a number of reasons.
First, some of the content could be loaded via AJAX. Sometimes, fragments are not properly inserted into the existing DOM.
The HTML that you are viewing may not have all been produced in the same document. For example, the page could be made of up components, or templates, and then thrown together right before the browser renders it. This isn't an excuse, but you can't assume that the HTML you're seeing was hand coded all at once.
What if some of the code generated by Markdown is invalid? You can't blame Stack Overflow for not producing valid code.
Lastly, the purpose of the DOCTYPE is not to simply say "Hey, I'm using valid code" but it's also to give the browser a heads up what you're trying to do so that it can at least come close to correctly parsing that information.
I don't think that most developers specify a DOCTYPE and then explicitly fail to adhere to it.

while I agree with the sentiment of "if it renders fine then don't worry about it" statement, however it's good for follow a standard, even though it may not be fully supported right now. you can still use Table for layout, but it's not good for a reason.

No, you should not use XHTML if you can't guarantee well-formedness, and in practice you can't guarantee it if you don't use XML serializer to generate markup. Read about producing XML.
Well-formedness is the thing that differentiates XHTML from HTML. XHTML with "just one" markup error ceases to be XHTML. It has to be perfect every time.
If "XHTML" site appears to work with some errors, it's because browsers ignore the DOCTYPE and interpret page as HTML.
See XHTML proxy that forces interpretation of pages as XHTML. Most of the time they fail miserably. This is one of the reason why future of XHTML is uncertain and why development of HTML has been resumed.

It depends. I had that issue with my blog where a YouTube video caused invalid XHTML, but it rendered fine. On the other hand, I have a "Valid XHTML" link, and a combination of a "Valid XHTML" claim and invalid XHTML is not professional.
As SO does not claim to be valid, I think it's acceptable, but personally if I were Jeff i would be bothered and try to fix it even if it looks good in modern browsers, but some people rather just move on and actually get things done instead of fixing non-existent bugs.

So long as it works in IE, FF, Safari, (insert other browser here) you should be okay. Validation isn't as important as having it render correctly in multiple browsers. Just because it is valid, doesn't mean it'll work in IE properly, for instance.
Run Google Analytics or similar on your site and see what kind of browsers your users are using and then judge which browsers you need to support the most and worry about the less important ones when you have the spare time to do so.

I say, if it renders OK, then it doesn't matter if it's pixel perfect.
It takes a while to get a site up and running the way you want it, going back and making changes is going to change the way the page renders slightly, then you have to fix those problems.
Now, I'm not saying you should built sloppy web pages, but I see no reason to fix what ain't broke. Browsers aren't going to drop support for error correction anytime in the near future.

I don't understand why everyone get caught up trying to make their websites fit the standard when some browsers sill have problems properly rendering standard code. I've been in web design for something like 10 years and I stopped double codding (read: hacking css), and changing stupid stuff just so I could put a button on my site.
I believe that using a < div> will cause you to be invalid regardless, and it get a bit harder to do any major JavaScript/AJAX without it.

There are so many standards and they are so badly "enforced" or supported that I don't think it matters. Don't get me wrong, I think there should be standards but because they are not enforced, nobody follows them and it's a massive downward spiral.

For 99.999% of the sites out there, it really won't matter. The only time I've had it matter, I ran the HTML input through HTMLTidy to XHTML-ize it, and then ran my processing on it.
Pretty much, it's the old programmer's axiom: trust no input.

Related

How important are HTML/CSS validation errors like <span>s inside <p>s? [duplicate]

This question already has answers here:
Closed 11 years ago.
Possible Duplicate:
What's the point of valid CSS/HTML?
This is a over-asked question, but I'm going to ask anyway.
How important is site/code validation?
I'm not referring to code errors that would break a site layout, like forgetting to close a tag, or using an unsupported tag. Just general warnings and tag usage errors.
I used to make sure my sites were validated 100%. No questions asked. Lately, I have not taken that approach however, especially with the use of HTML5. I feel that it is more important to make the site function as needed, even if it means putting a <p> tag within a <span> tag. Also, some of the CSS validation rules are pointless I feel. They are more considered with saving every possible byte then code readability.
Does it really matter? Looking for thoughts/concerns/opinions/links to articles. Thanks!
I do it for Peace of Mind.
When code is valid, it's more likely to work with a compliant browser. In theory, just because I test it and it works in various browsers today, does not mean it will work in newer browsers tomorrow. Standards Compliance solves that issue because if it follows the "Standard" today, it will follow the "Standard" tomorrow... of course, assuming that new browsers continue to support the Doctype you've selected to standardize against.
Why Validate - W3C
From W3C itself, here's a "Why Validate?" page that is extremely helpful: http://validator.w3.org/docs/why.html
Some of the main points:
Validation as a debugging tool
Validation as a future-proof quality
check
Validation eases maintenance
Validation helps teach good practices
Validation is a sign of
professionalism
Edit
The biggest reason I think is that when you don't validate, you're catering to broken browsers instead of designing for the browsers that do it right. Chad recently quipped that "if you use a degraded browser you deserve a degraded experience", and I think he's absolutely right. Design perfectly valid markup for standards-compliant browsers, and if absolutely necessary use IE6-specific stylesheets and the like for broken browsers. This way, you don't break the standards-compliancy of the good browsers, but you can "cheat" for the broken browsers.
Edit 2
The best exception, I think, is when you're moving forward instead of backwards. When you move backward, use separate stylesheets that only load for broken browsers - that way the good browsers aren't affected. But when you're moving forward (e.g, using CSS animations with vendor-branded tags, etc) you know that someday those will become standards (well, most of the time).
HTML5 is the culmination of our loose and sloppy HTML periods then overly-anal and pedantic semantic XML compliant xhtml period.
Which led us to an overall "validation is really important, but let's be pragmatic about it."
As stated, valid html is easier to debug, so there's the big plus right there. But there's plenty of room for exceptions as well. As long as you aren't using laziness as an excuse to allow for invalid markup when needed, it's probably a pragmatic decision.

What type of XHTML and CSS validations errors are safe to avoid?

What type of XHTML and CSS validations errors can be avoided? which would not harmful today and tomorrow (if we do not touch xhtml, css )?
I mean errors which will not create any problem on future upgrade of browser, css and html version? they just show as an error today?
I think one thing I know is Vendor extensions. Are there any other errors/warnings which will not create any bad effect for user and developer?
If I'm making a site and i get many errors should i try to give time to solve every error? if i will try to solve all error then i will have to use javascript on some instances in place of css
The XHTML and CSS validators will validate against the corresponding specifications of the W3C standards. Ignoring these mean that your page(s) are deviating from those standards.
Web browsers aim to implement these standards, so ignoring a warning is likely to cause issues on at least some browsers. Therefore, you cannot ignore any warning that the validators give.
Also, having XHTML and CSS conformant web pages is not guaranteed to work on all browsers and be compatible with them as the browsers may implement something differently or incorrectly.
Having conformant pages is still a good thing, as most browsers are (for the most part) conformant and having more conformant pages helps put the ownership on the browser implementers. That is, you (as a web page author) need only concern yourself with being standard compliant. If a browser can't handle that, the issue is with the browser, not the web page author.
If you want to be compatible with a large number of browsers, start with the valid conformant page and then add the minimum needed to get it working on other non-conformant browsers. Doing it this way is a lot easier than starting with a non-conformant page and trying to make that work on most browsers.
You should try to avoid all parse errors. If in doubt, try the validator.w3.org and use the html tidy function to clean up the code.
Each browser will render and parse XHTML and css differently. Even if it works now it might not work tomorrow.
The only safe answer is "none". The best guarantee you have for future compatibility wth all browsers is stick to the standard and have fully validated xhtml and css.

What are cons if we do not care about validation of XHTML and CSS?

What are cons if we do not care about validation of XHTML and CSS? Errors other than CSS 3 and vendor specific properties
In terms of development time(How valid XHTML and CSS code save time to find problems?),
Code debugging (How we can track then problem quickly?),
Cross browser compatibility (How it helps us to achieve cross browser compatibility?),
Website maintainability (How it would be helpful to maintain and update for someone else?),
Future changes in website (How it would be helpful to make any changes in design if client can ask in future?),
SEO ranking (How it can affect our site's search engine ranking?)
Accessibility (Does validity of code increase accessibility of site?)
I have to explain a client's Secretary,Code validation is not just Fashion, it is beneficial for his site. I'm not just advocating of this to make more money. it's not useful only for developer it mainly beneficial for his website.
There's the obvious point that if your markup is valid, the odds of it being rendered as you want it to be by a wide variety of browsers are improved.
But separate from that, sometimes you spend valuable development time tracking down bugs (usually ones that seem specific to a given browser) only to find that the reason for the bug is that your markup is invalid and different browsers are handling the invalid markup in different ways. Validating (whether it's XHTML or HTML) saves you time tracking down those sorts of problems. There was an example here just yesterday, in fact. The OP thought he was having a weird Firefox-specific jQuery problem. In fact, he just had invalid markup, and fixing the markup fixed his problem.
So I'm thinking that you tell the client that validation saves time, and therefore money.
Note that this is an argument for validating, not for proclaiming validity (via icons and such).
I found some very good answers here
http://validator.w3.org/docs/why.html
http://ianpouncey.com/weblog/2010/01/web-accessibility-myths/
Using markup improperly -- not
according to specification -- hinders
accessibility. Misusing markup for a
presentation effect (e.g., using a
table for layout or a header to change
the font size) makes it difficult for
users with specialized software to
understand the organization of the
page or to navigate through it.
Furthermore, using presentation markup
rather than structural markup to
convey structure (e.g., constructing
what looks like a table of data with
an HTML PRE element) makes it
difficult to render a page
intelligibly to other devices (refer
to the description of difference
between content, structure, and
presentation).
http://www.w3.org/TR/WAI-WEBCONTENT/#gl-structure-presentation

Is it practical to build a web site using strict XHTML and relying on CSS 100% for visual style?

I tend to take the academic approach all too often and adhere to strict principles in my development when the reality is that I could have finished the project sooner had I been a little less cautious. I'm looking to find the right amount of practicality.
I want to take the "Zen" approach to designing a site which (in my words) says "Use HTML strictly for content structure, and let the CSS magic do the rest". How practical is this in reality? One of the issues I run into is that I want to develop (make functional) the site first, then come back in and design it later. I know structure-wise how I want the site to flow, but I haven't even begun playing with the CSS layout, graphics, or any of the other designy stuff. What is the right approach here?
It's absolutely practical, and provides infinite benefit. In fact it's exactly what CSS and the separation of content and layout is designed for.
The right approach given the above, is to let different teams get on with the different tasks at hand. That requires (perhaps) an initial graphic design which can be quite rough, and a documented and collaboratively agreed set of naming conventions for things like "#viewport", ".user" etc..
The markup team will usually be backend driven and will usually lead the design team slightly, but they should and must remain flexible enough to change markup where required, or put that in the control of the designers.
This last is just my $.02, but where one person is both roles, again I think you lead yourself with the markup/backend first and then iteratively go to a design stage, then markup, then design, as required.
The approach you want to follow is the right one. Just two things:
If you use a validator for css or html, don't pretend that all your html or css pass the test. Obviously the ideal goal is that everything validates, but at a first stage I think is better not spend a lot of time in validation issues. And remember that no one validator is perfect, and the good way to use it at the beginning is to guide you in the right direction and avoid big mistakes (i.e. put the same id twice in one page, or put block elements inside inline ones...). Then, when the application is at a good stage, you can make your css and html perfect and valid.
Don't design the interface at the end. I think that the interface of the application can give you good directions in how develop your back-end too. So, at your place, I would design the interface first, with html and css, and then I'd start to add functionality to it.
(Sorry for my english, spelling corrections are welcome.)
It can be very practical and you will be suprised how clean your HTML looks. I like using a CSS reset file to help get started, I personally like the YUI reset. Another Zen item to consider is the use of unobtrusive JavaScript. This further separates the different layers of your code. JavaScript libraries, like jquery, prototype and dojo can help with this.
It can be done, and I think your site (and your web design skills, not least) will be much better for it. But it also has a certain learning curve. It requires a more thorough understanding of the XHTML/CSS specs than many people have.
Making sure your HTML can be validated is just the beginning.
Oh, and make sure all browsers run in strict mode when rendering the page.
Of course, you will require workarounds for IE support, but that can be done with several methods.
First, IE supports conditional comments, allowing you to include special CSS stylesheets just to fix IE bugs, which should get you most of the way, without affecting your compliant standard-version of the page.
For some things, you may need a bit of javascript as well, but it shouldn't be necessary for most common functionality.
There are reasons explained in http://www.webdevout.net/articles/beware-of-xhtml against using XHTML today. To summarize, XHTML is not supported unless you serve it as such and if you target older browsers (any IE version is old considering most of its features are implemented when they were still immature and not changed substantially for a while) you have no choice but serve it as HTML.
Unless you don't require features that being XML provides (like SVG, MathML), stick with HTML. You won't have any serious advantage over HTML, be any more semantic, have better CSS support (even less). But you get wider compatibility and your layout will be more predictable (for example table cells can inherit from first cell in the row in HTML, no such thing in XML, not even sure XHTML has any exceptions somewhere).
Validators won't help writing XHTML any more than HTML. Even annoy, if you use a strict one, leaving you wondering what is all the fuss about / in the br tag if you lie and say it is HTML. (Firefox view source shows it bright red if you serve XHTML as HTML). I am sure you can find more examples.
Sure, you can do that, but be prepared that it WILL NOT render under IE. On a recent web project, the majority of our front-end defects were fixing stuff in IE that already worked fine in Firefox. Maybe this will change in IE8, but I doubt it. In some cases we even had to write some javascript that would be executed on IE only to work around things that couldn't be done with just CSS.
while it sounds good in theory you cant create the layout for a site 100% with css. You still need to use some markup so that you have something to apply the css to. That said, you can come fairly close to ideal using this method. I'm constantly amazed at how little markup a true css guru actually requires.
closer the "zen" approach that you are really looking for is xslt. it works by your app generating xml data and then the xslt transforms that xml into html/css. this requires learning xslt and adds another layer of complication to the process of generating a page, but adds the separation you are looking for. In an ideal world the theory is that a programmer only has to worry about generating xml data and then a designer can generate the visuals using that data, however it rarely works that way as xslt is more technical than most designers can handle. Most of the time the programmer ends up generating the xslt which somewhat defeats the purpose.
One approach that works for me is to structure the HTML first, then add some minimal CSS in a tag in the same file (just enough to create the right layout etc). Then once you're happy with the structure, you can pull the CSS out into separate files and / or completely rework the CSS. This strikes the right balance for me - it's still a lightweight process, but it avoids the potential headache of finding and replacing inline CSS.
In theoria yes, in practice, browser differences may force you to add a bit of javascript to deal with the differences.
Now... Benefits of something is different from practicality of doing it. Are you guys forgetting IE or even the pain-in-the-whatever client who wants the impossible done?
I am tempted to say you have to make some exception to the strict DTD that you are using to make it work in a reasonable set of browsers and please your stakeholder for the website/web-app.
I am a standards freak and no one would be more happy than me if it was possible to build a website that doesn't violate even 1 DTD rule. But after 4 years, I just haven't been able to do it for practical purposes.
Sure if I am the one coming up with the requirements for the website I am going to develop, it might be possible, but I have to bend the business rules to accommodate that. Believe me, that's the only way it is possible.

Is it worth the development time to output valid HTML?

Developing websites are time-consuming. To improve productivity, I would code a prototype to show to our clients. I don't worry about making the prototype comform to the standard. Most of the time, our clients would approve the prototype and give an unreasonable deadline. I usually end up using the prototype in production (hey, the prototype works. No need to make my job harder.)
I could refactor the code to output valid HTML. But is it worth the effort to output valid HTML?
It is only worth the effort if it gives you a practical benefit. Sticking to standards might make it easier to build a website that works across most browsers. Then again, if you're happy with how a website displays on the browsers you care about (maybe one, maybe all), then going through hoops to make it pass validation is a waste of time.
Also, the difference in SEO between an all-valid html website and a mostly-valid html website is negligible.
So always look for the practical benefit, there are some in some situations, but don't do it just for the sake of it.
Yes. It's hard enough trying to deal with how different browsers will render valid HTML, never mind trying to predict what they'll do with invalid code. Same goes for search engines - enough problems in the HTML may lead to the site not being indexed properly or at all.
I guess the real answer is "it depends on what is invalid about the HTML". If the invalid parts relate to accessibility issues, you might even find your customer has legal problems if they use the site on a commercial basis.
Probably not if you have a non-complying site to begin with and are short on time.
However, and you won't believe me because I didn't believe others to begin with, but it is easier to make a site compliant from the start - it saves you headaches in terms of browser compatibility, CSS behaviour and even JavaScript behavior and it is typically less markup to maintain.
Site compliance (at least to Transitional) is pretty easy.
Producing compliant HTML is similar to ensuring that you have no warnings during a compilation - the warnings are there for a reason, you may not realise what that reason is, but ignore the warnings and, before you know where you are, there as so many, you can't spot the one that's relevant to the problem that you're trying to fix.
If you use Firefox to view your web pages, you'll get a helpful green tick or red cross in the bottom right hand corner, quickly showin you whether you've complied or not. Clicking on a red cross will show you all of the places where you goofed.
Some of the warnings/errors may seem a bit pedantic, but fix them and you'll benefit in many ways.
Your page is much more likely to work with a wider range of browsers.
Accessibility compliance will be easier (You'll have 'alt' attributes on your images, for example)
If you choose XHTML as a standard, your markup will be more likely to be useful in an AJAX environment.
Failure to do this results in unpredictability.
One of the biggest problems with web browsers is that they have perpetuated bad habits (And still do, in some cases) by silently correcting certain markup problems, such as failure to close table cells and/or rows. This single fact has resulted in thousands of web pages that are not compliant but 'work', lulling their developers into a false sense of security.
When you consider how many things there are that can go wrong with a website, being lazy when it comes to compliance is just adding more problems to your workload.
EDIT: having read your original post again, I notice that you say you don't bother with compliance when working on a prototype, then you go on to say that you usually use the prototype in production - this means that it's not strictly a prototype, but a candidate.
The normal situation in such circumstances is that once the customer accepts a candidate, no time is allocated for bug fixing or tidying up, thus strengthening the argument for making the markup compliant in the first place.
If you won't be given time later, do it now.
If you are given time later, then you had the time to do it anyway.
If you want your sight to be accessible to people with and without disabilities, as well as external systems, then yes, you should definitely make sure you output valid HTML.
It's easy to test your HTML with automatic validators.
I'll add to what Mike Edwards said about legal ramifications and remind you that you have a moral obligation too :)
Why not write the prototype in valid (X)HTML in the first place? I've never found that to be more of an effort than using invalid HTML. Producing valid XHTML should be a trivial task. (On the other hand, producing semantically meaningful XHTML might be more taxing.)
In short, I see no advantage whatsoever in using invalid HTML for prototypes.
I honestly dont know why it is extra effort to do standards based HTML. It's not as if it's hard and you should be doing it as a matter of professionalism.
If you paid someone to build you a house and he cut corners out of laziness, that you didnt notice at the time, but in 10 years cracks appeared in your walls, would you be happy?
Valid HTML just to be able to have a badge on your site - no.
Having "valid HTML" in the sense of "HTML that works on every major browser or browser engine" - yes.
Absolutely. Invalid code can cause all sorts of weird behaviors, and errors which don't obscure those that do when you get a validation report.
Case in point:
A yellow background was spilling out of a list of messages and over the heading for the next list of messages - but only in Internet Explorer.
Why? The background was applied to a list item, but the person who wrote the page had written it as a single list with a heading in the middle. Headings are not allowed between list items and different browsers attempted to recover from it in different ways. Internet Explorer ended the list item (with the background colour) when it saw the start of the following item (after the heading), while other browsers ended it when they saw the end tag for the first list item.
It was the only validity error on the page, so it took only a couple of minutes to track down the problem and fix it.
Because, if you stick to standards, your work will be compatible in the future. User Agents will strive for standard compliance and their quirks non-compliance mode will always be subject to change. This is the way is supposed to be.
Unless you're into that whole IE8 broken standards perpetuation thing that they want to enable by default. -- that's another argument.
Webkit, Gecko, Presto? (is that opera's engine?), and the others will always become more compliant with every release.
Unless your html work is in a IE embedded browser control, then there's really no reason to output valid html as long as it renders.
In my opinion the key criterion is "fit for purpose" - If your clients want something for a small/internal market (and don't care if that alienates potential customers who have disabilities or use less-common browsers) then that's their choice.
At the same time I think it's our (as developers) responsibility to make sure they know the implications of their decisions - Some organisations will be bound by legislative requirements that websites be useable by screen readers, which typically means standards-compliant HTML.
i believe making valid html outputs wont hurt your development time that much if you've trained yourself to code valid html from the start. for one, its not that hard to know which tags are not allowed within an elementand the required attributes in a tag are sometimes the ones you'd really need anyway - i believe these are the main errors that makes your html invalid, so why not just learn them as early as now if you plan to stay on the web for long?plus outputting valid html can help boost your sites ranking
There are two rules for writing websites:
The site must work for your users.
The site must work for your users.
To meet the first rule, you have to code such that your site renders correctly when using Internet Explorer. Unless you have the freedom to alter your site design to use only those features that IE renders correctly, this means writing invalid HTML.
To meet the second rule, you have to code such that your site renders correctly when using screen-readers and braille screens. Although some newer screen readers can work with IE-targeted sites, in general this means writing valid HTML.
If you're working on a small project, or you're part of a large team, you can code a site that outputs IE-targeted HTML for IE, and valid HTML otherwise. But if you're taking on a medium-to-large project on your own, you have to decide which rule you're going to follow and which one you're going to ignore.
UPDATE:
This is getting voted down by users who think you can always get away with valid HTML in IE. That may be true if you have the flexibility to change your design to get around IE's shortcomings, but if a client has given you a design and you have to get it working, you may have to resort to invalid HTML. It's sad, but it's true, whatever they might think.

Resources