search

google / pepper.js

Pepper applications as Native Client or JavaScript
http://trypepperjs.appspot.com
BSD 3-Clause "New" or "Revised" License
323 stars 53 forks source link

what's the timeline? #5

Open sstirlin opened 10 years ago

sstirlin commented 10 years ago

pepper.js doesn't seem to be getting the love and attention that it deserves from the Native Client team (1 developer, no commits for 2 months).

This project needs to folded into the main Native Client SDK and polished. It needs to become part of the "standard way". A pnacl solution by itself is a tough sell. A pnacl + fallback to emscripten solution is far easier to justify

ncbray commented 10 years ago

When working on pepper.js, it became clear that the biggest stumbling block was threading and concurrency. For the past few months I've been trying to figure out threads in JS. That's been a lot of research and prototyping, but none of it shows up in the repo. There is love and attention here, but right now it's all forward looking "how the heck could this be standardized" sort of stuff.

As for folding it into the SDK, I've been gating on "one major user". If pepper.js gets folded into the SDK, it adds another configuration for the SDK folks to deal with. To take that step, I need to show (to myself, if no one else) that this has legs. There's an internal product at Google that may fit the bill, but I've been focusing of the threading stuff to the exclusion of all else. Your prod has reminded me that I should be pursuing this, too. Thanks. I'll see what I can make happen in April.

sstirlin commented 10 years ago

Excellent! I think developers (or, at least, myself) want to develop in Pepper without worrying about whether pnacl or asm.js will win.

Probably the best that you can do is document a threading pattern based on unique_ptr or move semantics (something like Mozilla Rust). Secretly the data is copied to/from web workers behind the scenes.

I'm not saying to write your own threading library, but rather to hack the data movement into the Boost/C++11 library behind the scenes. Then publish a pattern that declares that shared state and mutexes will be ignored and should not be used.

Frankly, C++ needs a documented pattern like this anyway. I'd just follow the Rust model (they seem to have thought it through)

fungos commented 10 years ago

+1 - I'm keeping an eye here waiting for news :)

eligrey commented 10 years ago

@ncbray Re: threading: You can use my Core Estimator to get an estimate of the number of CPU cores for browsers that don't support NaCl sysconf(_SC_NPROCESSORS_ONLN).

As for performance: It scales logarithmically with the number of cores, so your future 128-core workstation shouldn't take over 8-15 seconds or so to fully estimate (depending on the speed of the individual cores).

I'm also looking into integrating the NaCl method so Chrome users don't have to run the timing attack.

DiamondLovesYou commented 10 years ago

@sstirlin Speaking of Rust... (shameless self-promotion). Beware: it is a WIP and barely works as is.

eligrey commented 10 years ago

Update regarding core count: this value has been standardized as navigator.hardwareConcurrency and is shipping soon in WebKit and Blink.

sstirlin commented 10 years ago

@DiamondLovesYou: NICE! Do you have any of the Pepper API ported yet?

DiamondLovesYou commented 10 years ago

@sstirlin Parts of it!. Here's the (admittedly rather sparse) documentation.

danielrh commented 10 years ago

When I try to build the library rustc compiles and some of the libs compile, but eventually /home/daniel/Downloads/pnacl-rust/bin/rustc -C cross-path=/home/daniel/Downloads/nacl_sdk/nacl_sdk/pepper_canary -C nacl-flavor=pnacl --target=le32-unknown-nacl -L /home/daniel/dev/rust-ppapi/deps/rust-http --sysroot=/home/daniel/Downloads/pnacl-rust --cfg ndebug -O --cfg ndebug --dep-info build/openssl.d --out-dir build lib.rs lib.rs:1:1: 1:1 error: can't find crate for std lib.rs:1 #![feature(struct_variant, macro_rules)] ^ error: aborting due to previous error make[1]: * [build/libopenssl-06cd96fd-0.0.so] Error 101 make[1]: Leaving directory `/home/daniel/dev/rust-ppapi/deps/rust-openssl' make: * [deps/openssl.stamp] Error 2

Is that a known problem? any idea how to fix it? rust-nacl is a cool idea: I wonder if it could be made to work with pepper.js and emscripten as well!

On Wed, May 21, 2014 at 11:38 PM, Richard Diamond notifications@github.comwrote:

@sstirlin https://github.com/sstirlin Parts of it!https://github.com/DiamondLovesYou/rust-ppapi. Here's http://diamondlovesyou.github.io/rust-ppapi/ppapi/index.html the (admittedly rather sparse) documentation.

— Reply to this email directly or view it on GitHubhttps://github.com/google/pepper.js/issues/5#issuecomment-43853310 .