[whatwg] <object> element feedback

[Ian Hickson]

On Sat, 7 May 2005, Christian Biesinger wrote:
>
> OK, another <object> fallback question: Consider an object that has no 
> attributes that would allow selecting a plugin/content handler. For 
> simplicity, consider an object without any attributes, say:
>
>   <object>Foo</object>
> 
> What should be displayed? Should the user agent just fall back to the 
> contents of the object? (Presumably "empty string" as fallback content 
> does not affect the answer to this question either :-) )

The spec now defines this.


On Wed, 25 Jan 2006, Hallvord Reiar Michaelsen Steen wrote:
> 
> On encountering an OBJECT, the UA must check type and/or classid.
> 
> If the type attribute identifies a file type the UA handles 
> internally, check if the OBJECT has a data attribute. Render contents 
> of this attribute if found, proceed to fallback contents if not.
> 
> If type/classid identifies a known plugin, the UA must initialise the 
> plugin passing on information about the attributes and the name/value 
> pairs of any PARAM descendants that are not children of nested OBJECT 
> tags.
> 
> If type/classid does not identify a known plugin, and there is a data 
> attribute, do a HEAD request on the resource identified by the data 
> attribute. If the server sends a content-type that identifies a known 
> plugin, initialise plugin.
> 
> If there is no data attribute, the resource can not load or the 
> content-type of the resource is unknown, proceed to render fallback 
> contents.

The above is somewhat what the spec says, though it is more closely 
aligned with what browsers (Mozilla in particular in this case) seem to 
do.


> Issues: if there is a known content-type and a data: attribute, 
> should the UA check if the content-type sent from server also 
> identifies the plugin? In other words, do we want to say that the 
> type attribute on the OBJECT tag is only a hint and that the actual 
> HTTP content-type header is the one that counts?

Right now the spec uses the Content-Type header if it is present.


> I'm not sure if "only a hint" should be speced. It sounds quite risky to 
> meddle with the logic for embedding contents - could break many sites, 
> particularly if we are supposed to start relying on the famously 
> malconfigured HTTP content-type responses.

What should it be instead?


> Since there are use cases for plugins that do not load data at all, the 
> UA should initialise the plugin even without data attribute if the type 
> is known. That sort of negates the point about type being a hint because 
> we can't initialise one plugin, look at the content-type of files that 
> plugin decides to request and say "whops, we started the wrong plugin, 
> let's do this again". If the spec goes the "hint" route, it really needs 
> to make that apply only to OBJECTs with a data attribute, and to 
> minimise problems with malconfigured servers it should only take place 
> if the UA doesn't know the specified type attribute IMO.

Right now if type="" is present but data="" isn't then the respective 
plugin is fired, otherwise the plugin to fire is taken from the 
Content-Type header.


> Another issue is of course if and how one should map classid to plugins.

Indeed. I don't know how to do that. I suppose it is UA-specific.


On Tue, 31 Jan 2006, Shadow2531 wrote:
>
> The current methods of "The server Content-Type rules all" and "If 
> there's no data attribute, then fail" are not working out. There are 
> cases where a data attribute is not needed and there are cases where a 
> type attribute should be the priority. The type attribute is really used 
> as "Here's what plug-in/handler I want to use".

For the no-data="" case, the spec now handles it.


> There also needs to be some specifics for the classid attribute.  If the 
> classid contains an unsupported naming scheme, should the UA really have 
> to just fail or should it go on and try to use the type and data 
> attributes?

It appears it should fail. Do browsers do otherwise?


> How should data uris be handled?  Is it just up to the plug-in to handle 
> the passed data uri or is the UA allowed to create a file from the data 
> URI and pass the file path to the plug-in/hander (if it can do it 
> safely)?

Isn't this up to the plugin API?


> What should happen when a UA encounters a codebase attribute that has a 
> path to a cab file instead of a base URI? Should the UA always use the 
> codebase to resolve URIs even if it has junk in it?

Right now codebase="" is just ignored by the spec. What should it do?


> First, the UA should check for a classid attribute.
>
> If present and the the UA has a handler associated with the specified 
> naming scheme, then the UA should pass all arguments to the handler in a 
> manner that is specific to the handler.

That's what the spec says, indeed.


> If the classid attribute is not present or the naming scheme is not
> supported or the handler returns an error code, then the UA must check
> for the presence of a type attribute.

If classid="" isn't supported, it just shows fallback.


> If it exists, the UA must check to see if it has a handler associated
> with the Content-Type specified in the type attribute.
>
> If the Content-Type is associated with a handler, the UA must pass all
> arguments to the handler.

The resource is actually downloaded by the UA in this case, as far as I 
can tell. Is that wrong?


> If that fails and there is a data attribute, check the Content-Type
> sent by the server for the file and determine if there's a handler
> associated with the type.

The spec honours the Content-Type header over all else if it is present 
and classid="" isn't.


> If there is a handler associated with the type, pass all arguments to
> the handler.
>
> If there is not, the UA *should* compare the file extension to a list
> of known Content-Types associated with that extension.
>
> If a handler is found, pass all arguments to the handler.

I do not want to add any extension-sniffing to the spec, it is too 
dangerous, highly unreliable, and as far as I can tell isn't necessary.


> ( URIs to files should be passed to the handler and the handler should 
> download the files as necessary. The Ua should not download the file on 
> its own. Only the handler should make that request to the UA.)

Really?


> Once the handler gets the arguments, it *must* detmermine whether it has 
> the necessary and valid arguments to initialize. If it does not, it 
> should return a failure code. If it does, it should try to initialize 
> and if successful, return a success code; otherwise, return a failure 
> code. The handler should return a failure code in every situation where 
> there's no need to initialize in the first place.

This is out of our scope.


> The UA must check for this return code to determine whether to display 
> fallback content.

It's not clear to me how to tell if a plugin failed.


On Wed, 1 Nov 2006, Christoph Päper wrote:
> 
> I think it would be helpful to /explicitly/ allow content types (alias 
> media types) in |type| of |object| to omit the subtype, e.g.:
> 
>   <object type="video" data="foo.mpv"/>
>   <object type="audio" data="foo.mpa"/>
>   <object type="image" data="foo.png"/>       ~= <img src="foo.png">
>   <object type="application" data="foo.swf"/> ~= <embed src="foo.swf"/>
>   <object type="text" data="foo.txt"/>        ~= <iframe src="foo.txt"/>
> 
> Maybe this is all the support for this element type that should be 
> required from conforming implementations. Furthermore |width| and 
> |height| should be required for freely scalable formats, but OTOH not 
> apply to 'audio' types (i.e. always equal zero), and exclude the space 
> required for an optional inline GUI.
> 
> I never understood, by the way, why videos and Flash-like content 
> shouldn't work within |img|. (Parameters can be specified in URIs.)
> 
> I could also envision an HTML5 where |alt| was optional for (or even 
> removed from) |img|, which in return was only allowed to be used for 
> optional, decorative images (and perhaps likewise |embed|). Every 
> illustration conveying meaning was then to be embedded using |object| 
> (including descriptive content, but nesting |object|s would be 
> discouraged, although allowed) or more sophisticated methods. This 
> wouldn't keep many correctly authored existing pages conformant, though.

I don't really see how to handle this feedback. We can't really make the 
changes that would change how <object> works today.


On Tue, 7 Nov 2006, Lachlan Hunt wrote:
>
>   Should the data attribute be required for <object>?  There is at least 
> one plugin that doesn't need it.  XStandard embeds itself without using 
> it, like this:
> 
> <object type="application/x-xstandard" id="editor1" width="100%" height="400">
>   <param name="Value" value="Hello World!" />
> </object>
> 
> I'm not sure whether that's a correct use of the object element or not, 
> and I do think it would make more sense for XStandard to replace a 
> textarea, but there must be implementation issues with that or 
> something.
> 
> http://xstandard.com/page.asp?p=A4372B00-8D7F-4166-977C-64E5C4E3708E&ss=C2B75B64-1544-429D-ACDA-07D17E35FB56

It's allowed.


On Tue, 7 Nov 2006, Shadow2531 wrote:
> 
> Reason 1:
> <object classid="java:MyJavaClass"></object> is used for Java. (
> Although I'd love to have <object type="application/java"
> data="file.class"></object> work, but ...)

This is not allowed. Java should use the standard mechanism, as you point 
out.


> Reason 2:
> You might want to load a plugin like this: <object
> type="application/x-mplayer2"></object> and later use scripting to
> tell the plugin to get the resource. If the data attribute is
> required, you'd be forced to load a resource that you didn't want to.

Indeed.


> Reason 3:
> The tcl plugin <http://www.tcl.tk/software/plugin/> suports inline
> scripts via a script param and things like the following need to work
> without the data attribute.
> 
> <object type="application/x-tcl">
>   <param name="script" value="script content">
> </object>

Indeed.


On Sat, 23 Dec 2006, Rohan Prabhu wrote:
>
> I've been writing a spec called 'XObject' for the past week.. and i'm 
> going on a 2-day vacation from tomorrow.. just got the idea of sharing 
> this with you. I wrapped it up in a small site-like thingy in the short 
> time i had.. so the site is not really good... but the content is there 
> as much as i have worked on. It is still a work in progress. Do read the 
> foreword there, I've already warned about some factual inaccuracies... 
> :)
> 
> Here is the link to it: http://xobject.tritiumx.com

It's not clear to me what this spec (which btw is now no longer available) 
was supposed to address.


On Sat, 7 Apr 2007, Henri Sivonen wrote:
>
> HTML5 should probably make the Java applet embedding patterns documented by
> Sun conforming or at least make the <applet> case conforming as it is the
> cross-browser syntax:
> http://java.sun.com/j2se/1.5.0/docs/guide/plugin/developer_guide/using_tags.html

Java needs to get over itself and stop being such a drama queen and just 
use the standard embedding methods that everyone else uses.


On Sat, 7 Apr 2007, Michael A. Puls II wrote:
> 
> Currently, the only way to embed an applet that's allowed by the spec 
> is:
> 
> <object type="application/x-java-applet">
>    <param name="code" value="MyJavaClass">
> <object>
> 
> (That works fine for Opera and FF at least.)

Indeed.


> There's also <embed type="application/x-java-applet" 
> code="MyJavaClass">, which is currently not allowed because src is 
> required.

Indeed.


> There's also <object classid="java:MyJavaClass> (which is used as a 
> non-deprecated example at 
> <http://www.w3.org/TR/html4/struct/objects.html#h-13.4>). This is 
> currently not allowed because there's no classid allowed, which has to 
> be present to be compatible with IE.

Right.


> And, then, there's of course APPLET, which needs to be defined (no 
> rush), including how fallback content works when java support isn't 
> present and the alt attribute is present. (As in, does alt get used over 
> fallback content or the other way around or somewhere in between.)

There's a note in the spec about this.


> I also think it might be necessary to clarify (or at least hint to)
> what mime type should trigger java. application/x-java-vm and
> application/x-java-applet etc. are provided by the Sun Java plug-ins,
> but not all browsers use those plug-ins. (application/java is used in
> W3C examples.)

The spec has an example with x-java-applet now.


> On a side, if codebase is allowed on the object element, it will have 3 
> different uses.
> 
> 1. base URI for resolving (kind of like <base href="">)
> 2. URI to a .cab file (for activeX stuff)
> 3. For java, it's specifically a URI to the directory the .class file
> is in unless you're using current IE in which case, it's #2 and a
> codebase param is used instead.

o_O

I'd rather just not support it and make it non-conforming... Do any 
browsers actually support case 1? Case 2 seems incompatible with case 1.


> In the case of #3 (for browsers besides current IE), where both a 
> codebase attribute and a codebase param are present, one of them would 
> have to override the other. It might be necessary to define which one 
> and how etc.  (Going to something specific like this might be out of 
> scope for plug-ins etc. in general, but Java handling might be an 
> exception.)
> 
> With all the different OBJECT situations for Java, APPLET will be a huge 
> relief.

How about just not supporting Java?


On Mon, 22 Oct 2007, Vlad Alexander (xhtml.com) wrote:
>
> I noticed that the latest HTML 5 draft states that the "name" and 
> "codebase" attributes are not allowed on the "object" element.
> 
> 1. Plug-ins, such as XStandard, use the "name" attribute for submitting 
> content to the server without the need for JavaScript. This makes for an 
> accessible solution. XStandard has been doing this for 4 years in some 
> browsers. IE, Firefox and Opera support this feature. This attribute 
> serves a vital role for plug-ins.

name="" is now supported, though not for this purpose.


> 2. The "codebase" attribute is the only way to auto-install or update 
> plug-ins in IE. This is an important feature that should not be removed 
> without providing an alternative solution.

I take it you don't like the idea of HTML+JS being the alternative 
solution?

Auto-installing unprotected binary code on an enduser machine seems 
unbelievably unwise.


> 3. The HTML 5 draft states that the "embed" element is used for plug-in 
> content. The "object" element is a better mechanism for loading plug-ins 
> and is supported by all browsers. The HTML 5 spec should state the 
> "embed" or "object" elements can be used for loading plug-ins.

I'm not sure which bit you're referring to here.


On Thu, 25 Oct 2007, Michael A. Puls II wrote:
> 
> Yes, I agree. The name attribute has to be supported (at least when the 
> object is inside <form>. (All the form stuff hasn't been dealt with yet 
> though.)

It's not clear to me how this is supposed to work. Does the NPAPI define a 
way to submit form data?


> > 2. The "codebase" attribute is the only way to auto-install or update 
> > plug-ins in IE. This is an important feature that should not be 
> > removed without providing an alternative solution.
> 
> If we do define it, it's going to be fun because what it's for depends
> on how it's used.
> 
> It can be a URI to fetch a new version (for IE).
> It could be the directory that java class files are in (in browsers
> besides IE when using the Sun java plugin).
> It could be a base URI used for resolving relative URIs (in browsers
> besides IE for possibly native stuff and some plugins).
> 
> For the last 2, there would need to be rules to follow  to determine
> whether it's an IE upgrade URI or a base URI so browsers besides IE
> don't use it as a base path if it's not really a base path. But,
> there's not much need to use a codebase attribute in FF, Opera and
> Safari (not even for java).

I have no idea what to do here.


On Tue, 1 Jan 2008, Jeff Walden wrote:
>
> http://biesi.damowmow.com/object/011.html
> 
> Firefox 2 shows FAIL.
> Firefox 3 currently shows PASS.
> WebKit nightly shows a broken image icon.
> Opera 9.25 shows PASS.

WebKit now shows PASS, so this seems like a non-issue. (Spec agrees.)


> http://www.whatwg.org/specs/web-apps/current-work/multipage/section-embedded0.html#the-object
> 
> For the moment ignoring that WebKit's using @type instead of the 
> server-sent type of the <object> producing the PASS/FAIL/icon (contrary 
> to current HTML5 and HTTP RFC 2616), it seems that WebKit isn't 
> displaying fallback content when the primary content contains "errors" 
> -- malformed images, in this case -- where recent Firefox and Opera are.
> 
> The spec doesn't currently address falling back for malformed content, 
> i.e. malformed images, invalid XML (maybe?), content a plugin handler 
> declines to handle (?, dunno whether this is expressible in NPAPI or 
> ActiveX or whether that question's too implementation-dependent), etc.  
> I tend to think it should since a broken-image icon's not particularly 
> useful, but I don't care strongly one way or the other.

Fixed for everything but XML.

-- 
Ian Hickson               U+1047E                )\._.,--....,'``.    fL
http://ln.hixie.ch/       U+263A                /,   _.. \   _\  ;`._ ,.
Things that are impossible just take longer.   `._.-(,_..'--(,_..'`-.;.'