CloudFlare starts discussion about LuaJIT project governance | Hacker News Reader

CloudFlare starts discussion about LuaJIT project governance

freelists.org

89 pointsbratao52 comments

My impression is that LuaJIT is too perfect/complex to the GitHub collaborative model work, where developers uses the software, identify a bug/missing feature for his use case, dig in code for a day/week, and do a push. In LuaJIT the stakes are much higher, and a small change can have big repercussions. Relying on occasional open-source contributors seems like a direct path to abandonware.

I disagree, it lowers the barrier of entry for people to document code, clean up documentation, or just review the code and report issues (e.g. security related). Even if LuaJIT doesn't see a lot of random contributors, it makes it easier to review commits and what not.

I think it would be ok if you have a CI system that runs benchmarks. If a PR comes in you let the CI run and tell you if there are any regressions.

Good benchmarking systems are few and far between, and so many unexpected factors can cause noise, before you even write your first performance test. I'd love a Travis CI for performance tests, but it's a seriously hard problem.

How so? The benefits of GitHub benefit regular committers over fly-by PRs.

So true. Some (increasingly rare) code does not translate into modern trends. actually it's more like craftmanship and LuaJIT is a craft. Just like Gigablast open source search engine.

Fear not. First of all, occasional contributors are perfectly compatible with LuaJIT. Take a look at the Git log. Many of the commits are bugfixes that don't impact the overall "perfection". It's not a crystalline structure:

http://repo.or.cz/w/luajit-2.0.git/shortlog

You can even get a list of those commits that are contributed, mostly by the exact process that you describe as being inappropriate:

http://repo.or.cz/w/luajit-2.0.git?a=search&st=commit&s=Than...

Also, I don't think CloudFlare have the intention of entirely relying on sporadic contributions. As they said earlier this month, "we will be investing significantly in its development and are interested in hearing from people who would like to work full time on LuaJIT as employees of CloudFlare." (I'd be curious to know how that's moving along!)

Is it possible (as in, easy), to use something like gerrit/rietveld on top of (or next to) github?

I am imagining a bot what will take GH pull requests, submit the patch to gerrit for review, and answer to the user something like "thank you for your contribution, this project uses an external platform for code review, a discussion was automatically created to review your changes, please follow this link to participate".

It looks like it's already done. There's a Gerrit plugin in-tree for doing a workflow like you described [1].

[1] https://gerrit.googlesource.com/plugins/github/+/master/READ...

Another option is to use a code review tool similar to Gerrit / Rietveld but built from the ground up for integration with GitHub. Recent entrants include https://review.ninja, https://gitcolony.com, and https://reviewable.io (disclosure: this last one's mine).

Maybe the project can take inspiration from projects that use run tests for each pull request, before merging.

I don't know much about Lua and LuaJIT, so let me ask a naive question:

If you would start out with a JavaScript JIT (like V8) what things would you have to add (i.e. things that are not required to JIT JavaScript) besides the obvious modifications in the parser?

One point I can think of is support for efficient compilation of co-routines.

The reason people use LuaJIT instead of v8 is because LuaJIT is faster than v8 (on some code) and is smaller and is more easily embeddable. Or at least that is my impression, I have no personal experience with it.

For me the main reason was the FFI, even if back in the day it did not support re-entrance - e.g. something in the "C" land has to call back "lua" land.

But the JIT in luajit is simply too impressive to skip it over. I was able to quickly prototype things with it, running almost at "C" speed, and some times even faster.

There's a in-depth discussion of those issues with respect to both LuaJIT and Javascript here:

http://lambda-the-ultimate.org/node/3851

Lua has goto (and JS does not), which means not all cfg's are well structured. That could impact the compiler in certain ways.

Lua also has finalizers, which effect the GC's design. The Lua/C api also makes it impossible for the GC to move objects, which means pretty much every current JS VM's GC is out.

Lua 5.3 has 64 bit integers, which would effect most JS VM's in a significant way (but to my knowledge LuaJIT doesn't support 5.3 so...)

There are other issues too, but these are just off the top of my head.

LuaJIT supports the 5.1 Lua standard, so goto doesn't exist.

Yeah, finalizers and weak key value stores would be an issue in certain versions of JS.

LuaJIT supports goto out of the box, as well as many other Lua 5.2 features, some behind à compiler flag.

I stand corrected. Thanks!

> Lua also has finalizers, which effect the GC's design.

V8 has a weak callback mechanism, which (while not exposed to JS) allows reentering JS from inside a weak callback - which means you can emulate Lua's __gc on top of this mechanism.

> The Lua/C api also makes it impossible for the GC to move objects, which means pretty much every current JS VM's GC is out.

If we disregard lua_topointer then Lua/C API only leaks internal pointers for strings (lua_tostring), userdata (lua_newuserdata, lua_touserdata) and threads (lua_newthread,lua_tothread) - everything else is manipulated using lua_State's stack.

This means VM only has to take care with regards to these objects. Userdata and threads can be just allocated outside of movable part of the heap and strings can be "externalized" (i.e. they payload relocated into the immovable space) on first access via lua_tostring. Coincidentally last thing is something that V8 supports[1] (though of course externalization is not a cheap operation as it requires copying).

> Lua 5.3 has 64 bit integers, which would effect most JS VM's in a significant way

Yeah, that's certainly a whole ton of work, but most of this work would be pretty technical.

JS engines might actually get int64/uint64 value types in the future (at some point there was an ES7 proposal - but currently it does not seem to be on track for inclusion).

[1] https://github.com/v8/v8-git-mirror/blob/master/include/v8.h...

LuaJIT does not support 64 bit integers. It uses double NaN tagging for storing object references. That's the basic principle of LuaJIT design and one of the most important source of its superior performance. In this matter, it is very similar to JS VMs.

Support for 64 bit integers would require to abandon this model and completely redesign the LuaJIT VM. Mike opinion about that was very negative.

I believe that this was one of the reasons he decided to abandon the project: he was disappointed by Lua creators decision to introduce 64 bit ints and the fact that LuaJIT can't be made Lua 5.3 compatible without rebuilding it from scratch (but that's only my personal impression).

> If you would start out with a JavaScript JIT (like V8) what things would you have to add (i.e. things that are not required to JIT JavaScript) besides the obvious modifications in the parser?

This is such a tempting thought -- that JavaScript and Lua are similar enough languages that an engine for one could be retargeted to the other with some parser changes and a few new features.

In practice it doesn't work out that way. Here is the story of Tessel, who originally set out to do it the other way around (implement JavaScript on LuaJIT) but reversed course after two years: https://tessel.io/blog/112888410737/moving-faster-with-iojs

"We believed that over time we could cover all of the corner cases of JavaScript with our own runtime, but perhaps we should have taken a cue from the “Wat talk” that it was going to be an uphill battle. While the semantics between JavaScript and Lua are very similar, they are also just slightly different in many ways (For example, the comparisons between null vs undefined vs falsy), and capturing all of those idiosyncrasies has proved frustrating for both the users and developers of the Runtime. [...] I still do believe it’s possible to make a nearly compatible runtime, but it’s going to take much more work than we expected and that resource investment would be an unwise business decision."

I'd think that the javascript JIT would need to be gutted and extensively re-engineered. There's a reason why each language has their own jit - in order to speed up specifics parts of the language.

While certain things are similar, the specific optimizations I'm sure follows the spec of the language so closely that it's not readily transferable to other languages.

The general optimization strategies, like tracing, etc are techniques that can be ported, but if you start with a highly optimized jit for javascript, you're gonna have to rewrite large portions - so much that it would as much works as rewriting luajit from scratch.

I was exchited by what I was hearing about Lua and LuaJIT, speed and quality of code wise. And then I tried the language.

While it's poweful in its "meta-programming" facilities and has some nicities, it has many uneeded bizarro decisions.

This alone was enough to make me leave it:

http://stackoverflow.com/questions/2705793/how-to-get-number...

If you have an array in Lua, which is a table used as a contiguous list of items, this is how you get the number of items:

    num_items = #my_array

Your qualm, apparently, is that this # is not built for all use cases of tables.

In C++, what operator tells you how many instance variables a class has? There is none, because that wouldn't make sense in many use cases. Lua tables, offering prototype-based inheritance, experience a similar conceptual ambiguity in the case where you want to take the length of an object.

Lua has made a design decision that the built-in length operator only works on arrays = tables-used-as-a-list and on strings -- cases without ambiguity. If you want the length operator to make sense for a hash map usage, it's a few extra lines of work for you. It's consistent with the langauge being small and flexible, a theme which Lua embraces with elegance.

>Your qualm, apparently, is that this # is not built for all use cases of tables.

My qualms are multiple:

(a) this is an operator when it doesn't need to -- a function would do --,

(b) this only works for contiguous list of items.

(c) Lua, like PHP and JS, has the same type for "contiguous list of items" and hash tables, not even providing a builtin just for the first and/or the latter (e.g. ES6 now does with "Map").

I can understand the powerful metaprogramming capabilities of tables. I also understand that they are not needed, and are even a hidrance, in the tons of cases where all you want is a simple hashmap or a simple vector that wont change under your feet.

>* In C++, what operator tells you how many instance variables a class has?*

Nobody said C++ was well designed ever, too.

>it has many uneeded bizarro decisions

I somewhat agree, though it hasn't stopped me from using it extensively. The example that you give isn't a very good one, it makes more and more sense as you get to understand the language.

However: 1-indexed arrays; and words as block delimiters ("end end end end...") are things that make me regularly take a step back and think, this would be a bit nicer if the language did things the normal way. Note that 1-indexed arrays are especially annoying in LuaJIT because the FFI naturally encourages you to manipulate C arrays, often resulting in a mix of 0-indexed and 1-indexed arrays. Ouch.

I would have titled that as "CloudFlare starts discussion about LuaJIT project governance". We don't have a solution or prescription. We're happy to help Mike in the transition in any way possible and are taking baby steps as we do so.

Ok, we changed the title to that.

The qualms I have with this dialogue are the same as before, because CloudFlare has little to no idea how they are going to handle this. JGC tweeted me about how 'Oh, we get so much benefit from LuaJIT being FOSS" but here we have CloudFlare walling LuaJIT into it's own entity on GitHub where I predict commit bits will be few and far between.

More than that, I don't think there has been enough narrative between Mike and the 'new LuaJIT crew' (CF) to determine how the project should be structured. In this thread, agentzh had a fantastic idea to vet somebody through Mike, someone the community knows can be trusted and also is somewhat familiar with the LuaJIT internals, and that person could serve as a canary between the project and CF.

I write a fair bit of Lua for game scripting, and I've even made a few bucks here and there helping folks with their custom plugin ideas etc, but I've never touched C before. Well, when the previous announcement was made, I immediately Amazon'd some C books, which I plan to devour in my free time. At which point I'll be learning Rust, and reimplementing LuaJIT in Rust, and hopefully convince Mozilla to host the git, such that it will be protected from FOSS corruption.

My worst fear is CF taking this project into the shadows, developing it closed-source (which they absolutely have a right to do) and not sharing their insights with the community.

I think everybody with any kind of invested interest in LuaJIT needs to be gearing up right now, such that we can do our parts to keep this project alive.

JGC tweeted me about how 'Oh, we get so much benefit from LuaJIT being FOSS"

And we do. We paid Mike Pall to work on open source LuaJIT, we've contributed to NGINX, hired people to exclusively work on open source projects. Here's the harsh economic reality: it is simply better business for us to spend a relatively small amount of money on open source support to get what we need from fantastic projects like LuaJIT than to try to develop this stuff ourselves.

My worst fear is CF taking this project into the shadows, developing it closed-source (which they absolutely have a right to do) and not sharing their insights with the community.

How do we "have the right to do" that? Whatever makes you think us trying to closed source this would have any benefit to us? How is the Github account (of which Mike Pall is an owner) us walling it off?

More than that, I don't think there has been enough narrative between Mike and the 'new LuaJIT crew' (CF) to determine how the project should be structured.

I predict that if I hadn't sent an email to the list soliciting ideas and input and had announced a new structure you would have complained that everything had been done in the shadows.

> Well, when the previous announcement was made, I immediately Amazon'd some C books, which I plan to devour in my free time. At which point I'll be learning Rust, and reimplementing LuaJIT in Rust, and hopefully convince Mozilla to host the git, such that it will be protected from FOSS corruption.

See you in 15 years.

Vanilla Lua is ~25k of C these days. That's what I'm going to dive into first. I've worked on bigger projects LOC-wise.

The point isn't the LOC. If you've never even touched C, you're not just going to have to learn that, you're going to have to learn how to write an optimizing compiler (because frankly if you've never touched C I'm skeptical you have any experience in this). And not just that: you're going to have to learn how to write the world's most optimizing trace-based JIT compiler for a dynamic programming language.

Mike spent 10 years designing LuaJIT and it is in a league of its own, not paralleled by anything else. Do not expect this inane project of yours to be solved by looking at 25,000 lines of C code. Especially if, point of fact, you do not even know C. Expect it to be 'solved' after a decade of research and hard work at minimum.

I'm not sure what paranoid reality you live in where you think this is feasible, or even desireable given your original post (frankly even though a port isn't needed Rust would be an awful choice for a 'port' due to the fact it's simply not got as good availability, the compilers and tools are less mature), but when I said see you in 15 years, it wasn't a joke - it was a conservative estimate.

Reply on news.ycombinator.com