Candidate for 0xc AngryBots #14
Conversation
This adds 0xc changes, except block signatures. Changes: - converted loops to not use their exit label - updated indirect_call operand order to put the callee last - inserted a gratuitous dependence on stack-machine rules (see the very first set_local) - inserted a gratuitous use of branch-to-function scope - optimized away unnecessary `blocks` - new "flat" text format
Changes:
- Convert to block signatures.
- Convert to section opcodes.
- Minimally update Release/AngryBots.js to work with the new JS API.
- Include a binary-encoded wasm file in 0xc format.
I made up a syntax for block signatures in the flat text format, taking
care to allow for adding arguments and multiple results in the future.
For example:
```
block $label28 () : (i32)
```
Is a `block` no arguments and one return type which is i32.
However, I'm not attached to this particular syntax, and it can easily
be converted to something else if desired.
Use the block signature syntax implemented here: WebAssembly/spec#336
|
The .wast file is now updated to use the block signature syntax of WebAssembly/spec#336. |
|
This won't run until we add polyfill support, right? |
Update the binary file to anticipate the changes described here: WebAssembly/design#800
|
The index.html file uses |
|
Makes sense. Let me know when you're finished updating the binary and I'll lgtm |
|
I've finished all the updates I'm aware of at this time. The binary runs in my development tree of Firefox, which has all the anticipated 0xc binary format changes I'm currently aware of implemented. |
|
Got it. Weren't we going to switch over to the |
Add a path to use the new JS API with Module and Instance, and use it instead of the old Wasm.instantiatemodule when wasm.experimentalVersion is not less than 0xc.
|
I've now updated the Angrybots.js script to switch to the new JS API, keeping the old Wasm.instantiateModule path in place when Wasm.experimentalVersion is less than 0xc. |
| var instance; | ||
| instance = Wasm.instantiateModule(binary, info).exports; | ||
| mergeMemory(instance.memory.buffer); | ||
| if (Wasm.experimentalVersion < 0xc) { |
There was a problem hiding this comment.
perhaps if (Wasm.experimentalVersion < 0xc || typeof WebAssembly == "undefined") would be more robust (in case some engines haven't yet switched over)?
|
LGTM modulo minor nit |
|
Nit fixed. |
|
I think we might have jumped the gun on landing this, as we haven't yet tried out these bits in V8. |
|
@titzer Sorry, I didn't check yet because we haven't revved |
|
Well, almost: we were about to land our 0xc patch which, after landing, would make updating this binary more of a hassle. @titzer Any ETA on V8 confirmation of the binary? |
|
Right, I think this was merged too quickly: we should be more disciplined about releasing binaries which we may just break in the future. No big deal, but let's not to that again. |
|
We should probably discuss how long we want browsers to support 0xC (and thus only demo bits) until the toolchain and everything else is also generating 0xC. I think we need a day or two to verify this binary and the other spec tests. |
|
Sorry folks. Thought this was less of an issue given that the codepath won't be hit until we all rev |
|
Looks like I need an LGTM on the new revert PR |
|
@titzer That's a good point. Having the binary verified will be a good first step, though, in demonstrating we're all on the same page. |
|
Sorry for the churn. @sunfishcode would you mind rebasing your changes against the rebase and re-opening? |
This is the same change as https://github.com/WebAssembly/webassembly.github.io/pull/14. Note that it includes the changes described in WebAssembly/design#740 however it doesn't have any user-defined or names sections, so it's unaffected by the details of how those are handled.
|
Rebased and reopened in #16. |
Update to the remaining anticipated 0xc features.
Additional Changes:
I made up a syntax for block signatures in the flat text format, taking
care to allow for adding arguments and multiple results in the future.
For example:
Is a
blockno arguments and one return type which is i32.However, I'm not attached to this particular syntax, and it can easily
be converted to something else if desired.