Lil: A Scripting Language | Hacker News Reader

Lil: A Scripting Language

beyondloom.com

104 pointsrazetime28 comments

The 'soft, spongy' type system is nice to see. That kind of strong dynamism is a good choice for the application: it will do the wrong thing sometimes, but will throw fewer show-stopping errors, which are really disheartening for casuals.

I do regard casting not-numbers to 0 as a flaw. I've had plenty of bugs where I wished `nil` would automatically cast to to `""` in string position, or the empty list/dict: but I have never, once, wanted a missing number to cast to 0.

The number zero doesn't have the "semantics of emptiness" in programming. It shouldn't be treated as false, either, unless we're dealing with a "raw" language, in which case, one should have to at least cast to boolean to get the truthiness of anything.

Personally, I would promote this to an error, because the other option, 'NaN', is itself a source of confusion. But at least NaN-poisoning the calculation will eventually inform the user that "htree" isn't a number.

This is really task/domain-dependent. It's convenient in shell scripts and AWK, but in bigger programs it really isn't good.

One of the things I love about Python is that its type system is actually quite strong, in that implicit conversion is very rare in the standard library and somewhat frowned upon in general. Having "falsy" and "truthy" values that are not actually `bool` instances is maybe the only big exception.

Lua hits my personal sweet spot for robustness vs. precision, Python is good enough to be worth criticizing, but hey, pleasant enough.

Lisps are a good example of using emptiness well, I wouldn't want an empty Lua table to be false, but you can't do classic Lisp programming without a falsy empty list, I wouldn't have it any other way. Large Lisp programs have their problems but it isn't the empty list causing those.

I wouldn't want to use Lil for the things I use Lua for. I think it's a great match for what Decker is doing here.

Definitely. I would argue even in shell scripts it's not always great. Just thinking about all those times rm -rf "~/${configdir}" accidentally wiped someone's home directory because that unset variable was implicitly turned into an empty string, etc.

I would not want to use a language with a spongy type system for any code I'm distributing to other people But for quick one-off scripts that only I will use, it is quite convenient.

   test ! "${configdir}"||rm -rf "~/${configdir}"

   [ "${configdir}" ]&&rm -rf "~/${configdir}"

   test x"${configdir}" = x""||rm -rf "~/${configdir}"

   [ x"${configdir}" != x"" ]&&rm -rf "~/${configdir}"

This assumes NetBSD ash or Debian ash which is what I use. I write scripts for me, not for others and often end up using "quick one-off" scripts for years. After decades of using UNIX, I have never wiped a root directory accidentally (but there is still time). YMMV.

Sure, and in Zsh you can do `(( $+configdir ))` if you want to differentiate "unset" from "empty". But the fact that you need this check at all demonstrates the existence of a problem.

You can also set the "-u" option (in Zsh, `setopt no_unset`) which at least will cause your script to crash with an error if you happen to forget this check. But then you have to remember to set that option as well!

Lua has a similar-ish problem with undefined variables being `nil`, but for the most part Lua is also rather strictly typed, and that will typically end up causing some kind of error deep inside a function call, rather than doing something dangerous silently.

While what you say about Lua is true, undefined variables can be promoted to errors. An `__index` function can throw an error if a slot looked up on a table is nil, and the environment is just a table.

That's Lua for you. The semantic model is dead simple and easy to understand, and even though roughly no one wants typos to become variables, the semantics would have to be less minimal to accomplish this.

So instead, there's a mechanism, a more than adequate one, but one does have to use it.

Also

  test "${#configdir}" -lt 1||rm -rf "~/${configdir}"

  [ "${#configdir}" -gt 0 ]&&rm -rf "~/${configdir}"

To clarify for others who only follow the link titles, this is not the Tcl-inspired scripting language LIL[1] nor the other Tcl-based, C-clothing-wearing scripting language Little[2].

[1] http://runtimeterror.com/tech/lil/

[2] https://www.little-lang.org/, https://news.ycombinator.com/item?id=26204218

Amusingly enough, like the linked article's Lil, i used my LIL ~11 years ago to make a HyperCard-inspired painting+card-based database hybrid[0][1] :-P. But mine was something i only played around with for a bit but ultimately abandoned.

Since the language is Tcl-like, i had some commands like "set", "get", etc to ignore arguments like "of", "the", etc for a HyperTalk-ish flavor :-P (obviously in practice the syntax is actually very different and that can be seen in the calculator video[0] below when it comes to expressions).

[0] https://www.youtube.com/watch?v=rshZHDDruAE (making a calculator, shows more LIL)

[1] https://www.youtube.com/watch?v=_8CYosAIIJw (making a telephone book, shows more painting)

This looks like a really interesting set of language features. I do wonder why they decided to invent their own entirely new language though, instead of building on Tcl or Lua or a Basic dialect. Or even embedding Python. I can understand not wanting to use Scheme due to S-expressions being intimidating.

Did they consider them harder to learn in layers? Did they feel like some language features were lacking, or were unnecessary/undesirable? Were there problems with syntax that prevented Decker scripts from looking the way they wanted them to look?

On their github [0] they mention that they like making compilers, and on the about page on their site [1] they say they like tinkering with new languages. I think they just had a cool idea for a project and wanted to build a language to base it off of with features that were convenient for their goal, and a syntax they enjoyed.

[0]:https://github.com/JohnEarnest

[1]: https://beyondloom.com/about/index.html

Nice. The embedded querying is neat.

I see the query language goes: `select ... from`

One thing I like about query languages that start with the "from" class is that then it's easier in REPLs to provide suggestions for what can be selected.

You type, say, `from people` and you can get a suggestion for `name`.

Having written an sql parser, I very very strongly agree.

I'm curious to know which part of the parsing becomes easier with Select moved later in the statements?

Typical argument for the "From table select column" are due to better matching evaluation logic and improving type-ahead suggestions in IDE's, but I haven't yet heard the argument that it also makes parsing simpler.

Good question, my answer was misleading. It changes virtually nothing in the parse or in the resulting AST.

Writing the parser as a process has made me (even) more familiar with SQL syntax and semantics and understand it better, which has only increased my disgust[1] of the whole matted thorny snarl that it is. I was asking for a dose of rationality for Lil, not a syntactic infection inherited from a misdesigned language. HTH.

(BTW the designers of SQL knew well that it had other problems and admitted them, eg. search for "A Critique of ANSI SQL Isolation Levels").

[1] yeah, that's the right word.

I've had the pleasure of alpha-testing several builds of lil and decker. It's nice to see that it's out in the wild now.

Some of the little applications I made in an hour or two each (with live help from the creator of Lil):

- a plotting program that tweens between two different plots when you move a slider

- a frame based traditional animation program

- a (color!) paint program with reflections for drawing little mandalas

It's actually quite a nice tool for rapidly prototyping all kinds of applications.

Also, I'm not sure if it's obvious from the given link, but there are implementations of both lil and decker in both C and JavaScript, so it's an incredibly portable little system!

Wow, the GUI calculator demo in ~2 minutes is pretty impressive: https://beyondloom.com/decker/

Recent and related:

Decker, a platform that builds on the legacy of Hypercard and classic macOS - https://news.ycombinator.com/item?id=33377964 - Oct 2022 (85 comments)

> Lil has a soft, spongy, dynamic type system in which values do their best to convert to a more relevant type as the need arises.

Noooooo, this is a terrible idea if you want this to be used at any non-trivial scale

This is a very domain-specific language, not designed to be used at a nontrivial scale. I think it's a good idea in this context.

Unfortunately, the fact that a system is not designed to be used at a nontrivial scale is no guarantee that it won't be used that way. For instance, there are a lot of Excel worksheets doing things I am sure the designers of Excel never expected Excel to have to do.

I mean, Excel does have LET and LAMBDA now...

What's the solution to this problem? On one extreme end you can clearly state its intended use in documentation and allow undefined behavior, and on the other end you prevent the user from using the tool in any undefined way.

I'm generally positive about restricting undefined behavior at a lower level (syntax, undefined field accesses, etc), but at a higher level (writing large scale applications, using it outside of intended domains, etc) it's not very clear how and even if you should restrict unintended usage.

The more room you have for "undefined behavior" at a higher level, the more people can explore using a tool in various ways you never thought about. This can in turn be valuable feedback on possible directions to expand your tool in.

it's not undefined behaviour, rather it's trying to be helpful by coercing stuff where it can. Unless I'm missing something, it's perfectly defined.

Triviality is in the eye of the beholder, but I consider this the correct choice for the application. I've already registered my one objection, but don't mind expanding on why I think this is a good idea.

Lil is designed for an interesting, if a bit retro, reboot of HyperCard. The fantastic thing about HyperCard was the ability for a user to just... change stuff.

In a HyperCard system, users become developers whenever they want to.

They're going to make mistakes. A lot of mistakes. Mistakes you and I, as developers, won't understand.

What should the runtime do? Not break. "Attempt to index a nil value" is a cruel thing to tell someone who is trying to create a dictionary.

What this does: "oh, you're trying to index this value? Ok, it's an empty dictionary now". "Oh, you're trying to sort it into a drop-down list? Empty string, nothing happens".

HyperCard scale isn't dozens of developers working on a cadence, with branches and merge reviews. HyperCard scale is someone making something really cool, and dozens, maybe hundreds, of personalizations. Some shared, some not.

The distinction between, say, adding photos to a gallery, and adding a photo editor to the stack, is not sharp in HyperCard.

I'm very pleased to see this project, although I think the retro aesthetic might be self-limiting at some point. The loss of HyperCard was a real one, we're suffering from it to this day.

This makes me think of the many programmers who started with BASIC on the early 6502/Z80 systems. BASIC of that era was a truly atrocious language (it got better later) but many of those people went on to learn better tools and make the computing world we know today. The original primitive tool served its purpose, which was largely to get people excited about the possibility of shaping their own experience instead of just consuming prepackaged applications. AFAICT Lil is a far better language than early BASIC, and that's enough.

Turning mistakes into possibly unwanted run-time behaviour is dodgy. The user may end up with a problem that's harder to debug than just a crash.

Perhaps have a switch that can enables stronger type checking.

Reply on news.ycombinator.com