A comment <https://github.com/LuaJIT/LuaJIT/issues/1475#issuecomment-47...> has already been made on the issue regarding the ternary operator, recommending `if x then y else z` over `x ? y : z`. This is exactly how it's done with if-then-else expressions in Luau <https://luau.org/syntax/#if-then-else-expressions>, another language compatible with Lua, and makes it a ton easier to nest (especially with elseif) and I believe still easier to read than `y if x else z`.
Exactly. I don't understand why people think the ternary operator is needed when you can just make `if` an expression instead of a statement. Then there is no new syntax to learn and `if` just becomes more useful.
I don't really understand where is this need to compress the logic into where small chunks comes from. In result we get single line of code which has multiple statement conditions, different paths, and it's not possible to grasp in one go.
Other practical example why ternary is bad: Many code-coverage solutions break on ternary because they don't correctly see that one of the branches was missed in tests.
I think that allowing an if statement to return a value to deal with the ternary introduces a now concept to Lua and that is that the value on the final line of a block is a return value much like Ruby. This changes the logic of the entire language more than adding a ternary. I do prefer the if statement as it allows so much more emergent behaviour, but it does have more implications to consider.
I suppose, though I feel what most people in this thread are thinking of is updating the existing if statement to also work as an expression, which does have plenty of implications (not that I think they would be bad, just more of a change to the language than the feature designers were going for) including final returns. The example I took from Luau still keeps the if statement and the if-then-else expression as separate constructs. One problem is that the statement and expression versions look very similar despite having different semantics (expression version must only contain expressions in its branches, must have an `else` case, does not have `end`).
Of course there are differences between LuaJIT and Luau that I think influence their decisions on possible ternary expression features:
- Luau users are disproportionately beginners to programming that I believe would find the if-then-else expression syntax easier to learn; LuaJIT developers have a larger user base of professional devs wanting to make their code faster, and they will probably be more familiar with the `x ? y : z` style since it's used in plenty of other languages.
- Luau is a lot faster moving in its development than LuaJIT in terms of language features, the Luau team just wanted to move people to ternaries from `x and y or z` because it's easier to optimise in a normal interpreter; LuaJIT, with their JIT, I assume would be able to more easily implement optimisations for constructs like `x and y or z` despite its slight semantic differences (my assumption on why the change is being considered now rather than earlier).
I find that so much harder to read compared to if/else or case/when in ruby.
The ? is basically an attempt to use fewer if/else, at the cost of condensed if-else like structure. I always need to look at both parts after the ? whereas in a single if or elsif I don't. case/when in ruby is even better here e. g. regex check:
def foo(i)
case i
when /^cat/
handle_cats
when /^dog/
handle_dogs
(I ommitted the "end"s here to just focus on the conditional logic.)
So is LuaJIT resuming active development after a decade or so of only maintenance? Great!
A lot of these changes make sense (although some of them are a bit too TIMTOWTDI for my taste) - but perhaps LuaJIT 3 would benefit from a change of name as well? Certainly with all these changes, it would be more like a separate language than merely a JIT-compiled version of Lua.
A bunch of them are from Luau, the Roblox fork of Lua and the dialect most young programmers know. Adding them to LuaJIT will make it easier to write for both Zoomers and AI agents, who have been exposed to a lot of Luau code.
I know the LuaJIT maintainer(s?) will never add it because it's too radical a departure from Lua but I wish they would include Luau's type annotations. There are typed languages like Teal that will compile to Lua that should work (although I've had a difficult time getting Teal to work with cdefs) and you can kind of fake it by using C structs obviously but having such a feature be native to Lua itself would be nice.
That takes me back a bit. It's a perl-ism. I used to think it was a great design feature but I've come to strongly prefer "There should be one way to do it, and it should be obvious"
Looks like LuaJIT is catching up, but calling these "syntax extensions" is confusing. Is the intent to hold LuaJIT fixed against some earlier Lua version (I guess 5.1) and adopt newer syntax piecemeal?
I welcome the compound assignment operators. Playdate's version of Lua also has that extension.
I wouldn't be surprised if their plan is to keep compatible with 5.1 and adopt newer features where feasible/compatible. Luau, another language with the explicit goal of extending Lua 5.1 in a compatible way, has a section in their documentation listing all newer Lua features and detailing why they chose to or not to adopt them <https://luau.org/compatibility/>. The needs of Luau and LuaJIT are different though the reasoning is nonetheless fascinating.
> Is the intent to hold LuaJIT fixed against some earlier Lua version (I guess 5.1)
Yeah. PUC-Rio went in a direction that Mike Pall didn't want to follow. Something to do with garbage collector finalizers, if I remember correctly, which is a notoriously thorny issue in every language it exists.
I think it was more the extra layers of indirection added to function environment\effectively global variable access added in 5.2 .
The removal of scanning for changed userdata finalizer meta method in 5.2 is just a commonsense fix for bad design that made GC atomic phase run time, thats not incremental scale up with the number of GC userdata objects alive no matter if they have a finalizer or not.
LuaJIT is an involuntary fork of 5.1. It already had various extensions that conflicted with the 5.2 implementation of the same features, and Mike Pall made it clear on the mailing list he wasn't going to change how LuaJIT worked.
> For compatibility with other computer languages, the following classic Lua operators can be written in a more customary syntax:
Why though? What does changing `and` to `&&` actually achieve? Were people confused?
Changing the syntax seems very surface level. It's not actually fixing any problems, just making Lua no longer look like Lua. It's not going to help anyone write/learn Lua. It will make everything more complicated as there are now two ways to do everything.
This feels like adding braces to Python because you don't like indenting your code.
Ruby has both kinds of operators as well, and it's fine. The thing in Ruby, though, is that the English logical operators have lower precedence than the symbolic logical operators, so you can use them in place of parentheses. Sometimes that's confusing, other times it can be used to make code very readable.
In general, I would expect symbolic operators to be desirable in complex boolean expressions, because "loud punctuation" stands out among English words when reading the code.
yes, ruby inherited this from perl, though 'or' has lower precedence than 'and' in perl, and they're equal in ruby. Which sounds like something going to cause mistakes, but I yet to see 'and' and 'or' together in the same expression in ruby.
I've always found it odd that and/or in Ruby isn't just considered equal to &&/||, and I have never really used the english operators except for the usual modifiers like if and unless.
What is a practical use case where the lower precedence makes sense?
I can't imagine that making the differences between the languages more subtle would improve the performance of chatbots. Subtleties aren't their strong suit.
Tangently related but I’ve been deep in Lua recently working on a rust implementation that supports Lua 5.1-5.5 in one Rust Binary https://github.com/ianm199/omnilua.
My ultimate goal was to support LuaJIT in Rust as well but this does not make it easier.
Also, one issue I have with this repo is that, since so much of it seems to use Claude, as an actual human I struggle to read and parse any of the information.
In a mix of official and unofficial benchmarks wall clock performance is ~1.4x as fast as C Lua and the memory usage is ~1.7x.
So performance is worse to be clear but within range. There’s some performance improvements I haven’t gone for yet that would get it down to ~1.1 I think.
This is amazing! Can a program call across versions? Like could we take a Lua 5.1 codebase and upgrade only a portion of it at a time to a new Lua version?
Hmm I think in general I would not recommend doing this just because Lua does change pretty significant things from 5.1-5.5 so you could have some really hard to understand behavior if some portions are 5.1 vs 5.4 for example.
I think where it would be most helpful is converting a codebase and being able to easily run tests to ensure behavior is the same.
One of the interesting things about Lua is because they don't really maintain compatibility between major versions, there isn't a huge ecosystem, and as a result there's less friction against making your own, slightly incompatible version. When you add on the simplicity of implementing the language, it's created a really diverse set of lua-alikes. Weird (and cool) for a language to have a diverse ecosystem of implementations, but not necessarily libraries.
Fragmentation is terrible for a language. Just look at Scheme. Nobody actually uses Scheme itself, it's always some Scheme implementation like Racket, Guile, Chicken, Chez, etc.
Languages should probably protect themselves with trademarks or something.
Some of these really look like QoL improvements. I'm not convinced ternary statements are an ergonomic improvement in particular. The examples given don't make a compelling case, 'visually tidy' is not the same as readable.
I kinda have seen somewhere on internet, that the language design of lua and js(well, ecmascript to be precise) is somehow related. But can't really find the exact reference I have seen.. it was long time ago when I read this.
There's some overlap in the languages they were inspired by (eg Scheme, or the chains Modula -> Lua vs Modula -> Java -> Javascript), but as far as I'm aware, the original designs were made independently.
Now, the object systems do look similar, but that seems to be a case of convergent evolution: Javascript took direct inspiration from Self, whereas Lua's system is based on a more generic fallback mechanism for table access.
Never will I understand ternary operators. As soon as you introduce it, some chuckle heads want to use them everywhere. Worse if the syntax allows nested ternarys. I guess it keeps the language open for code golfing, but it otherwise seems like redundant syntax that at best saves a few characters.
This is the best answer in my opinion. Ternary is just sugar for an expressive if. LuaJIT seems to be focusing on adding new syntax though, maintainer might not be amenable to updating existing semantics.
I imagine that if/if-else as expressions and then necessarily their bodies as expressions would entail a much more fundamental change to the language. You then have to think of a way for the bodies to indicate a result. That, or you have to make a special case for if/else sequences where the bodies are bare expressions, in which case you've just invented the ternary operator.
Zig and Rust have addressed the problem of how the result of a block expression should be presented, but neither solution seems particularly satisfying to me.
In Rust, blocks may end with an expression, giving them a non-void result. But a block may also end in a statement, the only difference being that the statement ends in a semicolon, in which case the expression still has the void result, and I think that semicolon being the only difference makes it hard to scan at a glance where values come from.
In Zig, blocks may give non-void results by `break`ing out of them with an expression. But break normally ignores blocks and break out of loops only, so to break out of blocks you have to provide a label for it and give that when you break so as to break out of the named block and not the outer loop, e.g. `const x = label: { break :label 35; }`. That creates a problem of one of the most difficult classes in software engineering: naming things. Ideally I think `break` from a block should have its own keyword, e.g. `const x = { give 35; }`
I don't think if-expressions have to affect existing semantics. Basically, in the parser you would have two different kinds of AST nodes, one for when the `if` keyword is encountered in statement position and another for when it's encountered in expression position.
Right now, `if` in expression position is just a syntax error ("unexpected symbol")
Well, I believe there could be some complications with parsing related to the fact that Lua grammar doesn't really requires semicolons between the statements.
But other than that, yeah, detecting "if" in the expression position is pretty unambiguous. No idea why most languages went with "cond-expr ? then-expr : else-expr" bracketed syntax instead.
Surely the most likely explanation is familiary from C?
But e.g. ml-family languages (like OCaml, F#, Haskell) and Rust just have the *if* expression that has a non-void value. If your language accepts expressions as statements (most do?), then I think that should just be compatible out of the box.
Yes, but why C had that syntax? Oh, right, because it didn't use if-then[-else]-end for the conditional statement, and reusing if(cond)[-else] with prohibited braces would be awkward.
Oh, and Lua most famously does not accept expressions as statements. Which, now that I think of it, would actually evade most of the parsing complications.
No, not basically, it simply doesn't have them, Ternary means three as in it operates on 3 operands, and/or operates on 2 operands. They are also not equivalent.
x = a ? b : c # x is b, same as you would if a {x=b} else {x=c}
lua and/or
a,b,c = true, 1, 2
x = a and b or c -- x is b
a,b,c = true, false, 2
x = a and b or c -- x is c
The or is dependent on its previous operand, so b will return false and skips to c, even if you meant for it to be b. you must use an if then else. However, you can have more than a ternary, if there is no need for short-circuit evaluation, as in, any of the operand is not a function CALL like c(), and you want to remain inside an expression, then you can do this instead
select (select is a native C function, this is faster than the table creation below)
I guess for the JS case it makes sense to be able to shave a few characters for file shrinking purposes, but generally I'm more biased to code clarity and "self-explainability"
I find it most useful in languages that have non-mutable variables and you want to avoid a mutable variable or an extra function when the value comes from a simple condition.
It might have been better to publicly document and stabilise the LuaJIT bytecode, which would allow people to then come up with whatever syntax they wanted in their own custom frontends.
Looks like LuaJIT is really going to fork away from Lua this time. After these changes, it won't be a compatible Lua 5.1 implementation anymore, it will be a new language.
LuaJIT has held back the lua ecosystem for over a decade. There's no reason to not at least try to move the implementation closer to luau or puc lua, not create yet more incompatible syntax
I don't think Lua is your average ecosystem. Lua is used as an embedded interpreter. For example, Neovim doesn't want to change its configuration language's syntax just because there is a new version of Lua available.
On the contrary, we can claim that luajit has stabilized lua for implementations and for users (strengthening Lua 5.1 dominance, which makes the experience more homogenous across apps).
The syntax proposals look fine. But I don’t feel they are needed. Lua is easy to write and grok. I default to using LuaJIT, and have never had an issue with the actual code. Integration with the Lua ecosystem is the problem. Fix the compatibility issues with LuaRocks packages and PucRio. That would be the best dev ex update in my opinion.
They shouldn't add the ternary operator, it keeps `?` from being usable on it's own for safe navigation and requires the ugly `?.` operator, like `a?.[b]` or `f?.()` instead of `a?[b]` or `f?()`.
We did have a project some years ago looking at extending LuaJIT on which Tom Fransham did excellent work. Alas, the funder's priorities moved on (as is their right!), so we didn't get to finish it. It was a bit sad, as we (well, mostly Tom) had built up a real head of steam, but c'est la vie. Still, either way, I would never have claimed enough personal expertise with LuaJIT to take over maintainership!
He left for a break, returned and there was no second break or anything.
I don't want to spam it in repo, so leaving it here: he is kind of a hero doing this work and I (hopefully we) am very grateful for his contribution to this world.
Some of these things are already implemented in PUC Lua. I don't know why they are diverting from lua spec on other aspects though. Why not work together with the PUC Lua team to add some of these to both lua versions and work on bringing their functionality closer to each other
instead of further apart. You might as well just make a new language instead. New features will end up not being used in effort to keep lua scripts portable.
In effort to not pollute the github issue, and hopes that the authors read this thread, I will put some of my thoughts here. There are 3 main strengths of Lua: Embeddable, Fast, and Small(easy to learn). I worry some of these changes divert from the last, expanding the language into a more complicated language.
Here is a list of things already implemented in PUC Lua so can be considered safe to add:
● ~ a Bitwise negate
● a & b Bitwise and
● a | b Bitwise or
● a ~ b Bitwise Xor
● a << b Left-shift
● a >> b Logical right-shift
● a // b Floor divide
● break Break statement
Don't get me wrong, I love some of these quality of life changes like:
● Const keyword: changing const from `local a <const> = 42` to `const a = 42` is far better syntax. The bracketed syntax was never a good idea.
● nil-Coalescing and safe navigation are great additions as they are basically macros at the parsing stage.
● Compound assignment is also basically a macro at the parsing stage as well. Lua should already have this honestly.
● Ternary Operator: I *like* it and it will help the stumbling block of the `a and b or c` common pattern already in use. Though I think (like others have stated) the If/then/else syntax would be more inline with the language, similar to ruby and would enable far more emergent behaviour. However it does establish a new pattern that the last value in a block is a return value similar to ruby so I am conflicted about that.
● `continue` it is nicer than a goto and is helpful.
● String interpolation: I honestly don't love lua's concat operator `..` so honestly string interpolation would be a nice to have and a feature of many modern languages. However I do worry about it's effect on parsing performance, and complexity of the language.
● Underscores in numbers: *shrug*
These are great ideas for the language but I would want all lua versions to support them, not just JIT. These are things that I think are a distraction:
● The `and` `&&` and `or` `||`. This just goes in the wrong direction for lua. It is often confusing in ruby (especially because of precedence issues) but also lua is a wordy language. It has `do` `end` blocks instead of brackets. It adds ambiguity for no reason.
● Short form function syntax. Lua does not need this and I am not sure anyone asked for this. Why `a = |x| do ... end` is more helpful than just `a = function(x) ... end` is unclear and would love to hear more about why this is being considered.
● Named varargs: It may be nice, but there is no real reason to add this. If you wanted a name for your varargs you could do `local name = ...` or just use the `args` variable already available in every function.
● Switch/Match/Select Statements: An optimized if/else block works just as well and another expansion of a small language.
I would love to see all of these come to LuaJIT (and love2d to support the new version too). It’s nice that Lua is simple, the syntax changes should hopefully make Lua code even simpler to read too
> It’s nice that Lua is simple, the syntax changes should hopefully make Lua code even simpler to read too
But which Lua?
Lua as implemented by LuaJIT is a fork of the language at this point. It's not fully compatible with PUC Lua (the reference implementation) and LuaJIT does not support features from the latest Lua version.
I imagine these changes make the original Lua adepts think their training wheels have come off. The language now looks like any other. That's a good thing to me, and it will help with the adoption of the JIT, but the whole language could have been syntax modernized as a result. But.. when the work is done someone else can fork it into something independent from its Lua roots.
From that perspective the conditional operator seems defensible, where it would be feature creep otherwise, as it is generally unloved elsewhere.
Please don't, inscrutable bitwise operators are an accident of the past even in systems languages, let alone in a scripting language. I'm not against infix operators for bitwise operations, just please spell them out with keywords rather than giving them sigils.
Likewise, going from `and` and `or` to `&&` and `||` would be a dispiriting regression. This is something that Zig got right.
It's not about having to remember them, it's that you shouldn't waste these short single symbols on operations that are only rarely used.
This stuff (especially the ternary) are a step backwards. There is just no reason to waste | on a bitwise or that gets used at 1% of the frequency of the standard or. In the future you might have a better use for it (pipeline syntax, sum or union types come to mind in other languages).
I dislike basically everything about these syntax extensions.
I'm going to disagree only because one of the primary use cases for LuaJIT is interop with C and I think there's a case for making the ergonomics match.
What’s the Lua/LuaJIT story these days for bundling up all the scripts of an application into a single file? Is there a way to do the super convenient go-like thing?
There's a bunch of options from a Google search, but embedding it in a thin C program and building that with https://github.com/jart/cosmopolitan would be a pretty go-like experience, I'd think.
I personally use a hand-written C wrapper program (which is not much more than a dozen lines long), and then embed the Lua scripts using objdump. This isn't quite as easy as Go since cross-compiling C programs is often somewhat tricky, but Lua is very portable and has zero dependencies, so it's usually not too hard.
There was a Lua[JIT] fork called Idle that seems to have fallen off the face of the Earth that did exactly that: it would take a small stub program, a runtime library and all the scripts and package them into a single PE/COFF binary that would read itself when run.
Love2D does it as well:
zip -9 -r SuperGame.love .
cat love.exe SuperGame.love > SuperGame.exe
Cool to see this - ergonomic syntax will make it easier to recommend Lua. Hope the PUC team aligns with this.
Also, I love this kind of pragmatism:
> Exponentiation assignment a ^= b has been deliberately omitted to avoid a predictable pitfall: this is how xor assignment is written in most other computer languages. Also, a syntax for exponentiation assignment is rarely asked for.
A ‘defer’ for closing files or deleting temp files at the end of a script will make life more enjoyable.
Yeah, "?." as safe navigation operator even in JS where it already exists is eye-sore. They could use some other single character instead of two characters. Question mark is already doing a lot with ternaries etc.
Instead of obj?.:method?.(…) it would be like obj#:method#(…)
Replace # with your favorite extra character instead of questionmark.
Lua has a lot of useless syntax. For instance, the "then". I have been using ruby and python for many years. Lua is living in the old age here.
That's just one example of so many more. I get that lua occupies a useful niche with its focus on embedded systems, but lua is not really a well-designed language in general. JavaScript has a similar problem.
Agreed, it keeps the parser fast as well because it is a lot more clear when the boolean statement ends and the code block begins. You either need parentheses, `then` or brackets around the block to make parsing clearly defined.
A comment <https://github.com/LuaJIT/LuaJIT/issues/1475#issuecomment-47...> has already been made on the issue regarding the ternary operator, recommending `if x then y else z` over `x ? y : z`. This is exactly how it's done with if-then-else expressions in Luau <https://luau.org/syntax/#if-then-else-expressions>, another language compatible with Lua, and makes it a ton easier to nest (especially with elseif) and I believe still easier to read than `y if x else z`.
Exactly. I don't understand why people think the ternary operator is needed when you can just make `if` an expression instead of a statement. Then there is no new syntax to learn and `if` just becomes more useful.
I don't really understand where is this need to compress the logic into where small chunks comes from. In result we get single line of code which has multiple statement conditions, different paths, and it's not possible to grasp in one go.
Other practical example why ternary is bad: Many code-coverage solutions break on ternary because they don't correctly see that one of the branches was missed in tests.
Because you can deduplicate certain parts of the logic which make the whole thing less error prone, such as
If I ever want to change x, or refactor this code some other way, its a more brittle process over x=c?1:2The ternary expression also takes up much less space so there is less of an emphasis on it, this can be a stylistic tool in a programmer's toolbox
I think that allowing an if statement to return a value to deal with the ternary introduces a now concept to Lua and that is that the value on the final line of a block is a return value much like Ruby. This changes the logic of the entire language more than adding a ternary. I do prefer the if statement as it allows so much more emergent behaviour, but it does have more implications to consider.
I suppose, though I feel what most people in this thread are thinking of is updating the existing if statement to also work as an expression, which does have plenty of implications (not that I think they would be bad, just more of a change to the language than the feature designers were going for) including final returns. The example I took from Luau still keeps the if statement and the if-then-else expression as separate constructs. One problem is that the statement and expression versions look very similar despite having different semantics (expression version must only contain expressions in its branches, must have an `else` case, does not have `end`).
Of course there are differences between LuaJIT and Luau that I think influence their decisions on possible ternary expression features:
- Luau users are disproportionately beginners to programming that I believe would find the if-then-else expression syntax easier to learn; LuaJIT developers have a larger user base of professional devs wanting to make their code faster, and they will probably be more familiar with the `x ? y : z` style since it's used in plenty of other languages.
- Luau is a lot faster moving in its development than LuaJIT in terms of language features, the Luau team just wanted to move people to ternaries from `x and y or z` because it's easier to optimise in a normal interpreter; LuaJIT, with their JIT, I assume would be able to more easily implement optimisations for constructs like `x and y or z` despite its slight semantic differences (my assumption on why the change is being considered now rather than earlier).
The ternary operator is easy to nest if you put each clause on a separate line. Then it looks just like nested if-then-else.
I love the ternary operator as much as anyone. But dang if it doesn't get hard to read when there is are a few, nested even.
Does that operator compile to faster assembly that if I make the same logic with verbose `if` logic? Is that a language specific outcome?
Or they could just ask granddaddy for advice:
=)or
or what is most lua-like?I find that so much harder to read compared to if/else or case/when in ruby.
The ? is basically an attempt to use fewer if/else, at the cost of condensed if-else like structure. I always need to look at both parts after the ? whereas in a single if or elsif I don't. case/when in ruby is even better here e. g. regex check:
(I ommitted the "end"s here to just focus on the conditional logic.)Unless you mess up its associativity, like PHP until 7.4.
https://wiki.php.net/rfc/ternary_associativity
http://phpsadness.com/sad/30
So is LuaJIT resuming active development after a decade or so of only maintenance? Great!
A lot of these changes make sense (although some of them are a bit too TIMTOWTDI for my taste) - but perhaps LuaJIT 3 would benefit from a change of name as well? Certainly with all these changes, it would be more like a separate language than merely a JIT-compiled version of Lua.
A bunch of them are from Luau, the Roblox fork of Lua and the dialect most young programmers know. Adding them to LuaJIT will make it easier to write for both Zoomers and AI agents, who have been exposed to a lot of Luau code.
I know the LuaJIT maintainer(s?) will never add it because it's too radical a departure from Lua but I wish they would include Luau's type annotations. There are typed languages like Teal that will compile to Lua that should work (although I've had a difficult time getting Teal to work with cdefs) and you can kind of fake it by using C structs obviously but having such a feature be native to Lua itself would be nice.
>TIMTOWTDI
What on earth is this supposed to mean?
There Is More Than One Way To Do It.
That takes me back a bit. It's a perl-ism. I used to think it was a great design feature but I've come to strongly prefer "There should be one way to do it, and it should be obvious"
TCBOO, AKA, the Highlander Conjecture.
Interesting, thank you.
Using acronyms is one of those ways. /s
There is more than one way to do it.
> but perhaps LuaJIT 3 would benefit from a change of name as well?
> Certainly with all these changes, it would be more like a separate language than merely a JIT-compiled version of Lua.
I agree. I suggested this on the GitHub issue but got nothing but downvotes.
Lua 5.3 (2015-01-12) added the bitwise operators:
https://www.lua.org/versions.html#5.3
https://www.lua.org/manual/5.3/manual.html#3.4.2
Looks like LuaJIT is catching up, but calling these "syntax extensions" is confusing. Is the intent to hold LuaJIT fixed against some earlier Lua version (I guess 5.1) and adopt newer syntax piecemeal?
I welcome the compound assignment operators. Playdate's version of Lua also has that extension.
I wouldn't be surprised if their plan is to keep compatible with 5.1 and adopt newer features where feasible/compatible. Luau, another language with the explicit goal of extending Lua 5.1 in a compatible way, has a section in their documentation listing all newer Lua features and detailing why they chose to or not to adopt them <https://luau.org/compatibility/>. The needs of Luau and LuaJIT are different though the reasoning is nonetheless fascinating.
> Is the intent to hold LuaJIT fixed against some earlier Lua version (I guess 5.1)
Yeah. PUC-Rio went in a direction that Mike Pall didn't want to follow. Something to do with garbage collector finalizers, if I remember correctly, which is a notoriously thorny issue in every language it exists.
I think it was more the extra layers of indirection added to function environment\effectively global variable access added in 5.2 .
The removal of scanning for changed userdata finalizer meta method in 5.2 is just a commonsense fix for bad design that made GC atomic phase run time, thats not incremental scale up with the number of GC userdata objects alive no matter if they have a finalizer or not.
LuaJIT is an involuntary fork of 5.1. It already had various extensions that conflicted with the 5.2 implementation of the same features, and Mike Pall made it clear on the mailing list he wasn't going to change how LuaJIT worked.
> For compatibility with other computer languages, the following classic Lua operators can be written in a more customary syntax:
Why though? What does changing `and` to `&&` actually achieve? Were people confused?
Changing the syntax seems very surface level. It's not actually fixing any problems, just making Lua no longer look like Lua. It's not going to help anyone write/learn Lua. It will make everything more complicated as there are now two ways to do everything.
This feels like adding braces to Python because you don't like indenting your code.
> This feels like adding braces to Python because you don't like indenting your code.
Now this I can get behind...
You need Bython https://github.com/mathialo/bython
Ruby has both kinds of operators as well, and it's fine. The thing in Ruby, though, is that the English logical operators have lower precedence than the symbolic logical operators, so you can use them in place of parentheses. Sometimes that's confusing, other times it can be used to make code very readable.
In general, I would expect symbolic operators to be desirable in complex boolean expressions, because "loud punctuation" stands out among English words when reading the code.
Same in Perl, hence the good old pattern:
yes, ruby inherited this from perl, though 'or' has lower precedence than 'and' in perl, and they're equal in ruby. Which sounds like something going to cause mistakes, but I yet to see 'and' and 'or' together in the same expression in ruby.
I've always found it odd that and/or in Ruby isn't just considered equal to &&/||, and I have never really used the english operators except for the usual modifiers like if and unless.
What is a practical use case where the lower precedence makes sense?
Those two behave in the same way if you drop the parentheses:
1. statement if (condition || something)
2. (statement if condition) or something
> Why though? What does changing `and` to `&&` actually achieve? Were people confused?
Also consider AI, that has a greater training base of JavaScript than Lua. So making Lua look more like JS, should improve output and reduce mistakes.
I can't imagine that making the differences between the languages more subtle would improve the performance of chatbots. Subtleties aren't their strong suit.
No, don't consider AI.
AI has greater training on Python, which uses `and` and `or`, and it has absolutely no issue keeping that straight.
Tangently related but I’ve been deep in Lua recently working on a rust implementation that supports Lua 5.1-5.5 in one Rust Binary https://github.com/ianm199/omnilua.
My ultimate goal was to support LuaJIT in Rust as well but this does not make it easier.
Also, one issue I have with this repo is that, since so much of it seems to use Claude, as an actual human I struggle to read and parse any of the information.
For example, what’s the performance like?
Performance dashboard is linked here https://ianm199.github.io/omnilua/performance.html.
In a mix of official and unofficial benchmarks wall clock performance is ~1.4x as fast as C Lua and the memory usage is ~1.7x.
So performance is worse to be clear but within range. There’s some performance improvements I haven’t gone for yet that would get it down to ~1.1 I think.
Oh wow, seriously, I always thought Lua should have been like this. The 5.1/5.2/5.3+ split was so painful.
> My ultimate goal was to support LuaJIT in Rust as well but this does not make it easier.
I think you could stop right before the syntax extension.
Heh I was going to queue up a skill that uses the CLI to make it easy to convert scripts to older or newer versions.
If you have any Lua use cases let me know I’m looking for more real world use use cases to justify the effort here.
This is amazing! Can a program call across versions? Like could we take a Lua 5.1 codebase and upgrade only a portion of it at a time to a new Lua version?
Hmm I think in general I would not recommend doing this just because Lua does change pretty significant things from 5.1-5.5 so you could have some really hard to understand behavior if some portions are 5.1 vs 5.4 for example.
I think where it would be most helpful is converting a codebase and being able to easily run tests to ensure behavior is the same.
I created a github gist https://gist.github.com/ianm199/5ba0366376eca673142e1f0c79b4... that explains what is practical (I used AI for this to be clear feel free to skim).
Do you have a use case in mind? Would love to chat or take a look at an github issue if you create one.
One of the interesting things about Lua is because they don't really maintain compatibility between major versions, there isn't a huge ecosystem, and as a result there's less friction against making your own, slightly incompatible version. When you add on the simplicity of implementing the language, it's created a really diverse set of lua-alikes. Weird (and cool) for a language to have a diverse ecosystem of implementations, but not necessarily libraries.
Fragmentation is terrible for a language. Just look at Scheme. Nobody actually uses Scheme itself, it's always some Scheme implementation like Racket, Guile, Chicken, Chez, etc.
Languages should probably protect themselves with trademarks or something.
I see JavaScript.
Some of these really look like QoL improvements. I'm not convinced ternary statements are an ergonomic improvement in particular. The examples given don't make a compelling case, 'visually tidy' is not the same as readable.
Worse, I see C (as in ! or &&), and Perl (as in manifestly more than one way to do it).
There are real improvements though, such as ?. and ??= that help with default-nullable everything.
Ternary is very useful, but it I'd rather see it implemented idiomatically:
Structural pattern-matching could be fantastic, but no syntax is suggested.I kinda have seen somewhere on internet, that the language design of lua and js(well, ecmascript to be precise) is somehow related. But can't really find the exact reference I have seen.. it was long time ago when I read this.
There's some overlap in the languages they were inspired by (eg Scheme, or the chains Modula -> Lua vs Modula -> Java -> Javascript), but as far as I'm aware, the original designs were made independently.
Now, the object systems do look similar, but that seems to be a case of convergent evolution: Javascript took direct inspiration from Self, whereas Lua's system is based on a more generic fallback mechanism for table access.
Lua to me always felt very JavaScripty, just with a different syntax.
Lua predates JavaScript by about 2 years though.
Never will I understand ternary operators. As soon as you introduce it, some chuckle heads want to use them everywhere. Worse if the syntax allows nested ternarys. I guess it keeps the language open for code golfing, but it otherwise seems like redundant syntax that at best saves a few characters.
That’s why “if” should just be an expression
This is the best answer in my opinion. Ternary is just sugar for an expressive if. LuaJIT seems to be focusing on adding new syntax though, maintainer might not be amenable to updating existing semantics.
I imagine that if/if-else as expressions and then necessarily their bodies as expressions would entail a much more fundamental change to the language. You then have to think of a way for the bodies to indicate a result. That, or you have to make a special case for if/else sequences where the bodies are bare expressions, in which case you've just invented the ternary operator.
Zig and Rust have addressed the problem of how the result of a block expression should be presented, but neither solution seems particularly satisfying to me.
In Rust, blocks may end with an expression, giving them a non-void result. But a block may also end in a statement, the only difference being that the statement ends in a semicolon, in which case the expression still has the void result, and I think that semicolon being the only difference makes it hard to scan at a glance where values come from.
In Zig, blocks may give non-void results by `break`ing out of them with an expression. But break normally ignores blocks and break out of loops only, so to break out of blocks you have to provide a label for it and give that when you break so as to break out of the named block and not the outer loop, e.g. `const x = label: { break :label 35; }`. That creates a problem of one of the most difficult classes in software engineering: naming things. Ideally I think `break` from a block should have its own keyword, e.g. `const x = { give 35; }`
I don't think if-expressions have to affect existing semantics. Basically, in the parser you would have two different kinds of AST nodes, one for when the `if` keyword is encountered in statement position and another for when it's encountered in expression position.
Right now, `if` in expression position is just a syntax error ("unexpected symbol")
Well, I believe there could be some complications with parsing related to the fact that Lua grammar doesn't really requires semicolons between the statements.
But other than that, yeah, detecting "if" in the expression position is pretty unambiguous. No idea why most languages went with "cond-expr ? then-expr : else-expr" bracketed syntax instead.
Surely the most likely explanation is familiary from C?
But e.g. ml-family languages (like OCaml, F#, Haskell) and Rust just have the *if* expression that has a non-void value. If your language accepts expressions as statements (most do?), then I think that should just be compatible out of the box.
Yes, but why C had that syntax? Oh, right, because it didn't use if-then[-else]-end for the conditional statement, and reusing if(cond)[-else] with prohibited braces would be awkward.
Oh, and Lua most famously does not accept expressions as statements. Which, now that I think of it, would actually evade most of the parsing complications.
As long as the language supports lazy evaluation and short-circuiting through expressions, then great.
Yep. Everything should be.
Lua basically already has ternary operators anyway since "and" and "or" short circuit. I also don't see the need of adding additional syntax for it.
No, not basically, it simply doesn't have them, Ternary means three as in it operates on 3 operands, and/or operates on 2 operands. They are also not equivalent.
lua and/or The or is dependent on its previous operand, so b will return false and skips to c, even if you meant for it to be b. you must use an if then else. However, you can have more than a ternary, if there is no need for short-circuit evaluation, as in, any of the operand is not a function CALL like c(), and you want to remain inside an expression, then you can do this insteadselect (select is a native C function, this is faster than the table creation below)
table creation/selection of course, doing something like does not look so bad> The classic Lua idiom a and b or c has a pitfall when b is nil or false: then c is returned, even when a is truthy.
> E.g. true and false or 42 returns 42, whereas true ? false : 42 returns the (expected) false.
I guess for the JS case it makes sense to be able to shave a few characters for file shrinking purposes, but generally I'm more biased to code clarity and "self-explainability"
That’s what compression is for.
I find it most useful in languages that have non-mutable variables and you want to avoid a mutable variable or an extra function when the value comes from a simple condition.
In Lua (and LuaJIT) you can already use `and` and `or`:
The knuckle heads are already using them everywhere.It might have been better to publicly document and stabilise the LuaJIT bytecode, which would allow people to then come up with whatever syntax they wanted in their own custom frontends.
For others interested in alternative syntax to the Lua VM/API sometime ago I've created LJS https://github.com/mingodad/ljs and also https://github.com/mingodad/ljsjit, I've also included an utility lua2ljs program based on the Lemon parser and re2c that convert Lua scripts to LJS with line by line synchronization https://github.com/mingodad/ljs/tree/master/lua2ljs, to test it I've also translated a few non trivial projects (https://github.com/mingodad/ZeroBraneStudioLJS , https://github.com/mingodad/raptorjit-ljs, https://github.com/mingodad/snabb-ljs, https://github.com/mingodad/premake-core/tree/ljs, https://github.com/mingodad/CorsixTH-ljs).
I'm proud of it and thankfull to the Lua/Luajit projects.
Looks like LuaJIT is really going to fork away from Lua this time. After these changes, it won't be a compatible Lua 5.1 implementation anymore, it will be a new language.
So shouldn't it have a new name?
Why won't it be compatible? Any code written in Lua 5.1 will run on LuaJIT.
But not the other way around.
LuaJIT has always been a superset of 5.1.
Which is why it should be a new language. Like C and C++.
well, it doesn’t say Lua5.1-JIT
It could be opt in.
Are there any rough estimates on popularity of lua implementations? At this point it feels lua means luajit
not even close, because there are a lot of places where you can't run LuaJIT
Where can you not run LuaJIT? Genuinely curious
Wasm and platforms like iOS and Nintendo Switch (I think).
anywhere that does not allow self modifying code such as app stores.
LuaJIT is not just a JIT, it also includes high speed interpreters for x86, Arm, and more.
LuaJIT has held back the lua ecosystem for over a decade. There's no reason to not at least try to move the implementation closer to luau or puc lua, not create yet more incompatible syntax
I don't think Lua is your average ecosystem. Lua is used as an embedded interpreter. For example, Neovim doesn't want to change its configuration language's syntax just because there is a new version of Lua available.
On the contrary, we can claim that luajit has stabilized lua for implementations and for users (strengthening Lua 5.1 dominance, which makes the experience more homogenous across apps).
The syntax proposals look fine. But I don’t feel they are needed. Lua is easy to write and grok. I default to using LuaJIT, and have never had an issue with the actual code. Integration with the Lua ecosystem is the problem. Fix the compatibility issues with LuaRocks packages and PucRio. That would be the best dev ex update in my opinion.
+= and ..= are things i find i'm constantly missing in lua.
Personally im a fan of introducing ternaranary operator in lua. Everyone uses `x and y or z` as a ternanary which i find way more confusing than ?:
Lua pursues "simplicity, purity, and simplicity." So... too much syntactic sugar is unlikely
They shouldn't add the ternary operator, it keeps `?` from being usable on it's own for safe navigation and requires the ugly `?.` operator, like `a?.[b]` or `f?.()` instead of `a?[b]` or `f?()`.
Yep. This is awful:
Seems like a bad idea to actively diverge from Lua, hostile even, especially without at least a clear change of name.
I thought luajit had completely stopped feature updates
I’m confused I thought Mike Pall left luajit and Laurence Tratt took over as maintainer?
We did have a project some years ago looking at extending LuaJIT on which Tom Fransham did excellent work. Alas, the funder's priorities moved on (as is their right!), so we didn't get to finish it. It was a bit sad, as we (well, mostly Tom) had built up a real head of steam, but c'est la vie. Still, either way, I would never have claimed enough personal expertise with LuaJIT to take over maintainership!
Mike Pall is to LuaJIT as PG is to Hacker News.
Edit: meaning he can come back anytime.
He left for a break, returned and there was no second break or anything.
I don't want to spam it in repo, so leaving it here: he is kind of a hero doing this work and I (hopefully we) am very grateful for his contribution to this world.
Are there great uses of LuaJIT out there? It was such a big thing before fast JS engines IIRC.
Off the top of my head, openresty, neovim, love2d
Cloudflare does or at least did use it a lot.
https://en.wikipedia.org/wiki/Torch_(machine_learning)
http://torch.ch/
https://github.com/torch/torch7
Some of these things are already implemented in PUC Lua. I don't know why they are diverting from lua spec on other aspects though. Why not work together with the PUC Lua team to add some of these to both lua versions and work on bringing their functionality closer to each other instead of further apart. You might as well just make a new language instead. New features will end up not being used in effort to keep lua scripts portable.
In effort to not pollute the github issue, and hopes that the authors read this thread, I will put some of my thoughts here. There are 3 main strengths of Lua: Embeddable, Fast, and Small(easy to learn). I worry some of these changes divert from the last, expanding the language into a more complicated language.
Here is a list of things already implemented in PUC Lua so can be considered safe to add:
Don't get me wrong, I love some of these quality of life changes like: These are great ideas for the language but I would want all lua versions to support them, not just JIT. These are things that I think are a distraction:I would love to see all of these come to LuaJIT (and love2d to support the new version too). It’s nice that Lua is simple, the syntax changes should hopefully make Lua code even simpler to read too
> It’s nice that Lua is simple, the syntax changes should hopefully make Lua code even simpler to read too
But which Lua?
Lua as implemented by LuaJIT is a fork of the language at this point. It's not fully compatible with PUC Lua (the reference implementation) and LuaJIT does not support features from the latest Lua version.
LuaJIT of course.
Is LuaJIT still based on Lua5.1? I wonder why they haven't followed the language spec up to Lua5.5.
Because 5.2 had a bunch of breaking changes. https://www.lua.org/manual/5.2/manual.html#8
Because they don't like the changes, to put it mildly.
> local gauge = direction == "up" ? count + 10 : count - 10
local gauge = count + (direction == "up" ? 10 : -10)
I imagine these changes make the original Lua adepts think their training wheels have come off. The language now looks like any other. That's a good thing to me, and it will help with the adoption of the JIT, but the whole language could have been syntax modernized as a result. But.. when the work is done someone else can fork it into something independent from its Lua roots.
From that perspective the conditional operator seems defensible, where it would be feature creep otherwise, as it is generally unloved elsewhere.
The question is does the coding agent like it or not ?
Please don't, inscrutable bitwise operators are an accident of the past even in systems languages, let alone in a scripting language. I'm not against infix operators for bitwise operations, just please spell them out with keywords rather than giving them sigils.
Likewise, going from `and` and `or` to `&&` and `||` would be a dispiriting regression. This is something that Zig got right.
What kind of person understands and needs bitwise operators but can't easily remember & | ~ and the arrows for shift? It's very little information.
The part I'd call a hassle is the different kinds of right shift but you have that same hassle if you use keywords.
I like using the and/or keywords for logical operations. Now let's make bitwise look significantly different from that.
It's not about having to remember them, it's that you shouldn't waste these short single symbols on operations that are only rarely used.
This stuff (especially the ternary) are a step backwards. There is just no reason to waste | on a bitwise or that gets used at 1% of the frequency of the standard or. In the future you might have a better use for it (pipeline syntax, sum or union types come to mind in other languages).
I dislike basically everything about these syntax extensions.
Lua is very unlikely to want to add newer/less-common syntax with special symbols.
Also a syntax for types can repurpose most symbols without being ambiguous.
And you can overload the bitwise operators. You can configure __bor to give you pipelining right now.
You can have them in C/C++ at least [1]
[1] https://en.cppreference.com/cpp/language/operator_alternativ...
Doesn't Zig also have bitwise operators?
The btiwise operators library doesn’t go away
I'm going to disagree only because one of the primary use cases for LuaJIT is interop with C and I think there's a case for making the ergonomics match.
What’s the Lua/LuaJIT story these days for bundling up all the scripts of an application into a single file? Is there a way to do the super convenient go-like thing?
There's a bunch of options from a Google search, but embedding it in a thin C program and building that with https://github.com/jart/cosmopolitan would be a pretty go-like experience, I'd think.
I personally use a hand-written C wrapper program (which is not much more than a dozen lines long), and then embed the Lua scripts using objdump. This isn't quite as easy as Go since cross-compiling C programs is often somewhat tricky, but Lua is very portable and has zero dependencies, so it's usually not too hard.
There was a Lua[JIT] fork called Idle that seems to have fallen off the face of the Earth that did exactly that: it would take a small stub program, a runtime library and all the scripts and package them into a single PE/COFF binary that would read itself when run.
Love2D does it as well: zip -9 -r SuperGame.love . cat love.exe SuperGame.love > SuperGame.exe
This doesn't work with ELF files, though.
extensions are detachable/optional, those arent extensions but features
Cool to see this - ergonomic syntax will make it easier to recommend Lua. Hope the PUC team aligns with this.
Also, I love this kind of pragmatism:
> Exponentiation assignment a ^= b has been deliberately omitted to avoid a predictable pitfall: this is how xor assignment is written in most other computer languages. Also, a syntax for exponentiation assignment is rarely asked for.
A ‘defer’ for closing files or deleting temp files at the end of a script will make life more enjoyable.
In aggregate this looks like a godsend, but there are some examples (like foo?.:method) that looks atrocious.
Yeah, "?." as safe navigation operator even in JS where it already exists is eye-sore. They could use some other single character instead of two characters. Question mark is already doing a lot with ternaries etc.
Instead of obj?.:method?.(…) it would be like obj#:method#(…)
Replace # with your favorite extra character instead of questionmark.
Is there any reason why they're not considering a single '?' like rust? Is it a parsing issue?
So you'd have: obj?:method(…)
Mike Pall wrote in the issue that it's easier to parse. If they get rid of the ternary operator, I'll ask him again to drop the period.
What are some pragmatic embedded scripting languages of choice these days if one has to consider:
1) Ease of learning, ideally minimal deviant behaviour (eg i consider lua tables to be a new concept in itself)
2) Reasonably fast. Not as much as lua jit but even half would be good enough
3) Mature
4) Has Rust bindings
Lua. Lua tables are easy and awesome. My hobby language unites Lua tables with functions too.
Lua has a lot of useless syntax. For instance, the "then". I have been using ruby and python for many years. Lua is living in the old age here.
That's just one example of so many more. I get that lua occupies a useful niche with its focus on embedded systems, but lua is not really a well-designed language in general. JavaScript has a similar problem.
For readability, `then` allows splitting with newlines very long conditional expressions, without having to wrap the condition in parentheses:
after `if` and `elseif` the parser simply goes on until it finds `then`.This is something I don't see a lot of people do. I've tended to do
and I'm still a little irked it works so well, the only alternative would be for the language to have labeled blocks. but that might be too terseAgreed, it keeps the parser fast as well because it is a lot more clear when the boolean statement ends and the code block begins. You either need parentheses, `then` or brackets around the block to make parsing clearly defined.
Python spells "then" as ":"
In Ruby you can choose between "then" and a newline.
This is very pot calling the kettle black.
English too