wasm support

[databasedav]

sorry if this is the wrong place but i literally could not find a single place online mentioning epub and wasm together

will future epub specifications include wasm support for scripting?

[databasedav]

correction, it is mentioned here https://www.w3.org/2020/12/epub3-wg-charter.html

- Guides on improving alignment with the Open Web Platform, such as:
    - Which Web APIs that are suitable for EPUB usage.should be available in EPUB?
    - Usage of [WASM](https://www.w3.org/TR/wasm-core-1/) in EPUB content.
    - Relationship of CORS requirement, externally hosted resources, and EPUB content.

[iherman]

As far as I can see, the current standards talk about scripting with very general terms (https://www.w3.org/TR/epub-rs-33/#sec-scripted-content). The current standards are pretty silent as for the language of scripts. Of course, today it is JavaScript but who knows whether something comes up tomorrow.

What this boils down to, I believe, is that this is entirely a question of individual EPUB Reading System implementation. Some of them do not accept any scripting, some of them do it partially, and I would expect some may accept WASM already today. This does not require any update on the EPUB standards themselves.

@bduga @davemanhall @mattgarrish @wareid ?

[mattgarrish]

What this boils down to, I believe, is that this is entirely a question of individual EPUB Reading System implementation.

Not entirely, I wouldn't think. To the best of my knowledge, you'll always have an externally compiled .wasm file to import and it's not currently a core media type. Only data blocks are exempt. That'll make it a nuisance to author absent an update to the CMT list.

[mattgarrish]

I was just double-checking and we do have an exemption for script data files, but that's not quite what this is. The files might sneak through epubcheck that way since it doesn't check code fetches or their purpose, but it wouldn't be in the spirit of the specification.

[iherman]

Ah... I did not think of that. I wonder whether it is worth adding this to the discussions at TPAC. @wareid w.d.y.t.?

[iherman]

Should we raise an issue separately with the proposal of adding .wasm to the list of core media types, or this one suffices? I tend to believe it is better to have it explicitly, but it may be an overkill.

Alternatively, we can raise an explicit PR that does that, referring to this issue.

FWIW: the official media type is application/wasm, the registration is https://www.iana.org/assignments/media-types/application/wasm, see also https://www.w3.org/TR/wasm-core-1/#conventions

[bduga]

Do we need to make a change? Sorry, I don't know a lot about wasm, but my understanding is that you load it via JavaScript - that is, there is no direct reference to the module from HTML. I think that case is covered in https://www.w3.org/TR/epub-33/#sec-exempt-resources. Or can it be directly referenced from a script tag in HTML today?

…

On Fri, Sep 13, 2024, 6:49 AM Ivan Herman ***@***.***> wrote: Should we raise an issue separately with the proposal of adding .wasm to the list of core media types, or this one suffices? I tend to believe it is better to have it explicitly, but it may be an overkill. Alternatively, we can raise an explicit PR that does that, referring to this issue. FWIW: the official media type is application/wasm, the registration is https://www.iana.org/assignments/media-types/application/wasm, see also https://www.w3.org/TR/wasm-core-1/#conventions — Reply to this email directly, view it on GitHub <#2649 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AA246ZCOYZP6BDZF2XCK3LTZWLUQDAVCNFSM6AAAAABOC53ZJWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGNBZGAYTCOJZGE> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[mattgarrish]

I think that case is covered in https://www.w3.org/TR/epub-33/#sec-exempt-resources

That's what has me a bit confused. If you fetched another js script and wrote a script tag, that exception wouldn't apply. But in this case the WebAssembly api would be called on the fetched compiled code, so nothing is written into the html. The idea for that exemption was data sets being fed into a script, but maybe compiled code fits in, too.

It might help to clarify that the primary exemption is for any kind of resource used by scripts, and cite both data sets and web assembly code as examples, rather than focus so much on data sets.

[iherman]

@bduga, that section includes

This exemption allows EPUB creators to include resources in the EPUB container that are not for use by EPUB reading systems.

One reading of this is that wasm files are not exempt resources, because they are used, albeit indirectly, by the reading system. At the minimum, this ambiguity must be clarified.

[iherman]

Don't we have the same issue with javascript modules? They are only referred to indirectly from your the script referred to (or contained by) the <script> element, but you may still want them to travel with your epub.

The answer is probably that javascript modules are covered by the fact that they are "protected" by the fact of being core media types, but it would be more "natural" to consider wasm files the same way.

[bduga]

@bduga, that section includes

This exemption allows EPUB creators to include resources in the EPUB container that are not for use by EPUB reading systems.

One reading of this is that wasm files are not exempt resources, because they are used, albeit indirectly, by the reading system. At the minimum, this ambiguity must be clarified.

But the next sentence is, emphasis mine:

The primary case for this exemption is to allow data files to travel with an EPUB publication, whether for scripts to use in their constituent EPUB content documents or for external applications to use

That is the exact use here - wasm are data files used by javascript in EPUB content documents, and seems to confirm that files used by JS are not considered used by the Reading System. Though, I agree this wording could use some tweaking.

[bduga]

The answer is probably that javascript modules are covered by the fact that they are "protected" by the fact of being core media types, but it would be more "natural" to consider wasm files the same way.

I agree that we should (probably) treat them the same way, but perhaps the correct thing to do is stop dictating what script types are allowed and just make anything that is referenced by the script tag as exempt resources. Do we really want to be in the business of adding media types every time browsers start adopting new scripting languages?

[iherman]

I agree that we should (probably) treat them the same way, but perhaps the correct thing to do is stop dictating what script types are allowed and just make anything that is referenced by the script tag as exempt resources. Do we really want to be in the business of adding media types every time browsers start adopting new scripting languages?

I fully agree. Who knows what the community may come up with in the coming years? Direct usage of TypeScript instead of javascript? Or Go/Swift/Rust/whatever?

[iherman]

ust make anything that is referenced by the script tag

referenced directly or indirectly right? Which may open the floodgates...

[bduga]

Yes, but I think the floodgates are already open. The exmption for data files was intended in part to allow data for script files that could then be displayed (via script) in the EPUB. I think the intention was for instance a large table that might be sorted or filtered via a script, but of course it could also be an image in a non-core type that the script knows how to decode and display. Or some type of portable document format that could be displayed in a viewer inside of an EPUB content document with no fallback required.

…

On Fri, Sep 13, 2024 at 9:23 PM Ivan Herman ***@***.***> wrote: ust make anything that is referenced by the script tag referenced directly or indirectly right? Which may open the floodgates... — Reply to this email directly, view it on GitHub <#2649 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AA246ZGZQP6UC2EJNLX5XBLZWO24LAVCNFSM6AAAAABOC53ZJWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGNJQHAZDOMRTGE> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[iherman]

Yep. But the floodgates I was referring to are different. As you said, the reason for exemption was for "passive" data that scripts may use. Adding wasm to the exemption means adding a, fundamentally, executable content.

But anyway, I do not think we really contradict one another. Clearly, we may have to think through again how scripts and data should be/could be handled. I vaguely remember that we had some discussion back then in the WG about handling exemption data, and I believe there were voices around who did not want to put any type of fences around those data. The coming up of wasm shows that whatever explicit fences we draw up, eventually it may become too restrictive...

[mattgarrish]

Or some type of portable document format that could be displayed in a viewer inside of an EPUB content document with no fallback required.

I think you're going too far with this interpretation. The input can be a non-CMT, but the result of using it has to be a supported format. Per 6.3.2.5:

EPUB creators MUST ensure that scripts only generate core media type resources or fragments thereof.

I tend to agree with Ivan that wasm and js modules weren't what we were considering with the exemption, but probably should have been. Why does the script code that calls the module have to be a CMT because it's referenced from a script element but the additional scripting resources it relies on don't have to be? It's a bit arbitrary.

If we make it explicit that resources referenced from a script tag are exempt, as well as any resources that a script can import/fetch/whatever, then the spec will be a lot better for it.

[bduga]

I think you're going too far with this interpretation. The input can be a non-CMT, but the result of using it has to be a supported format. Per 6.3.2.5:

EPUB creators MUST ensure that scripts only generate core media type resources or fragments thereof.

I don't think I am. There is nothing in that statement to say I can't have a script that acts as a renderer of any content type, it just can't produce them (and presumably directly load). But any type of file can be loaded and displayed via JS with no fallback required, so long as it is only referenced by the script (or scripts).

If we make it explicit that resources referenced from a script tag are exempt, as well as any resources that a script can import/fetch/whatever, then the spec will be a lot better for it.

I think we are all in agreement here. I don't know how big an issue this is today, though. The original question to us was when will we support wasm, and the answer is now so long as it is loaded via javascript. It would be nice to allow direct reference of non-JS resources via the script tag, but I don't know how common that is (I think it requires a polyfill for wasm).

[mattgarrish]

But any type of file can be loaded and displayed via JS

Could you give a specific example of how this would work? A script presumably has to write some html to render the content and that has to be a core media type per the definition, or provide a fallback. Even if you were to use canvas for display you're still converting whatever your source was into the image to paint on it. You're not rendering the original foreign resource.

[bduga]

A script doesn't have to create an HTML document to render it's content, it can directly manipulate the DOM or paint using a canvas. That is, it reads the source document, extracts the relevant data, and uses the platform's drawing primitives to display it. For example, PDF.js doesn't take PDF and turn it into an HTML document and load it, it directly renders PDF into a canvas element. This is different than something that takes a PDF, generates an HTML document, and then loads the HTML in an iframe.

…

On Sat, Sep 14, 2024 at 8:02 AM Matt Garrish ***@***.***> wrote: But any type of file can be loaded and displayed via JS Could you give a specific example of how this would work? A script presumably has to write some html to render the content and that has to be a core media type per the definition, or provide a fallback. Even if you were to use canvas for display you're still converting whatever your source was into the image to paint on it. You're not rendering the original foreign resource. — Reply to this email directly, view it on GitHub <#2649 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AA246ZGVNRDX3D5LHIO6NPDZWRFXHAVCNFSM6AAAAABOC53ZJWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGNJRGAZDENBUGU> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[GeorgeKerscher]

Hi, What would this do to the Accessibility DOM the screen readers access? It seems that would also need to be modified to make it available to Assistive Technology. Best George

[mattgarrish]

For example, PDF.js doesn't take PDF and turn it into an HTML document and load it, it directly renders PDF into a canvas element.

Right, that's what I mentioned above. That's not what I'd call rendering the foreign resource without any fallback, though. That's turning it into something native, which is what the scripting requirement enforces. You draw to the canvas element which as a part of html is a CMT.

What it sounded to me like you were saying is that so long as you used some scripting and called what you made a viewer then you bypassed the CMT requirements. So inject an iframe with some buttons and call it a PDF viewer and you can load the PDF straight into the iframe and bypass otherwise needing a fallback.

[mattgarrish]

What would this do to the Accessibility DOM the screen readers access?

I don't know the inner working of pdf.js, but from what I've read they do preserve a tree and any aria roles they can if the pdf is tagged. No idea how that turns out in screen readers, though.

[pchampin]

This was discussed during the pmwg meeting on 13 February 2025.

View the transcript

w3c/epub-specs#2649 (wasm)

mgarrish: The question was do we allow webassembly?
… it isn't explicitly referenced from the script tag
… we had a discussion around it, I think all we need to do is add a comment that it is ok
… We could add it to the script tag, or just exempt that tag entirely
… Main question is do we want add something to reading systems?
… It will work if they support the API, so not sure what we would need to add

ivan: I think there are no problems, but there was a question around security
… My feeling is it is scripting, and so no, but maybe we should point at the security section

mgarrish: I am not sure what else there is outside of JS
… we should ask our security person

duga: Do we want to explicitly call out webassembly, or just make it vague?

mgarrish: I think we can leave it vague, and use webasssembly as an example

Hadrien: When it comes to webassembly it opens some doors
… but our scripting has always been spotty
… so this pushes the issue further
… the more things like this we add, the more of a balance there is with reading systems. Just something to keep in mind

shiestyle: Can the reading system distinguish between JS and webassembly?

mgarrish: Short of inspecting the file, no, but doesn't need to
… we would have to flag it

duga: No, there isn't really a way of detecting it, but it doesn't really matter

ivan: What is the next step

mgarrish: I think we just need to change the exemption text a little
… real question is the script tag itself

shiestyle: I will ask Voyager for their opinion

shiestyle + toshiakikoike: Discussion in Japanese

MasakazuKitahara: adds to discussion in Japanese

shiestyle: The feeling is there is a risk, and they are not seeing any need for it in Japan

ivan: I am curious what the risk is
… The only thing is adding some speed, so it is not clear what the risk is

shiestyle: Well, the security risk. We already have JS, so maybe there is a little more risk?

ivan: Yes, but we can't block JS, it is already allowed
… so my feeling the only thing this brings is performance

mgarrish: We aren't forcing support, so it is fine to block from a RS perspective.
… It would be very hard to ban this
… it is already going through via an exemption

shiestyle: relaying conversation in Japanese

toshiakikoike: continues in Japanese

shiestyle: toshiakikoike says JS is allowed, so webassembly can be allowed, so they understand
… so they understand, this is just their feeling

---