From what I can tell from the introduction and initial HN thread Bel is an academic exercise in developing a programming language by focusing on axioms rather than implementation. Also:
I think the point of a high-level language is to make your programs shorter. All other things (e.g. libraries) being equal, language A is better than language B if programs are shorter in A. (As measured by the size of the parse tree, obviously, not lines or characters.) The goal of Bel is to be a good language. This can be measured in the length of programs written in it.
I struggled to install it on my own too, but GPT 4 fixed that issue splendidly. I highly recommend you try: create a free account on chat.openai.com if you haven't already (GPT 3.5 is free, GPT 4 costs $20/mo), and then give it a prompt similar to this:
"Help me understand how to install arc. When I try to install in racket it says:
racket: undefined; cannot reference an identifier before its definition >"
Just after that, Racket introduced an easier way to fix this, `unsafe-set-immutable-car!`, specifically as a stable way to mutate the so-called immutable cons cells going forward. :) We should probably move to that.
initializing arc.. (may take a minute)
serving from C:\Users\user\anarki\apps\news
starting app news
ready to serve port 8080
(or press ctrl and 'd' at once)
For help on say 'string':
arc> (help string)
For a list of differences with arc 3.2:
To run all automatic tests:
arc> (load "tests.arc")
If you have questions or get stuck, come to http://arclanguage.org/forum.
Arc 3.2 documentation: https://arclanguage.github.io/ref.
The `(:provide ...)` syntax is a special extension to the Arc language for the purposes of `#lang anarki`. Every `#lang anarki` file must start with `(:provide var1 var2 var3 ...)` fo r some number of variables (even zero). It describes what exports are visible to other Racket libraries.
I believe `#lang anarki` does something to make it so `(load "plain-anarki-deep-dependency.arc")` loads the file path relative to the directory the module is in.
There are a number of awkward things about using Arc to write libraries for Racket consumption:
- Racket languages typically have a notion of phase separation where some amount of the program (the compile-time part, i.e., phases 1 and up) happens during the process of compiling a module into a .zo file, and the rest of the behavior (the run-time part, i.e., phase 0) can be performed by loading the already-compiled .zo file. Arc programs are more like `#lang racket/load` in that they only begin macroexpanding the later expressions in the file after they've run the run-time parts of the previous expressions, so it's like the whole macroexpansion process has to wait until run time. The only things `#lang anarki` does at compile time are reading the file's s-expressions and processing the `(:provide ...)` directive.
- Racket and Arc both have macro systems. They both involve writing macro definitions in source-to-source styles that usually make it easy to write a macro that abstracts over another macro call. Sometimes, in more advanced cases, these styles require more attention to get the hygiene right. However, they use different "source" types (Racket's syntax objects vs Arc's plain s-expressions), and they have rather different approaches to hygiene (Racket's lexical information carrying module imports and sets of scopes, vs Arc's gensyms and global variable redefinition warnings). This means mixing Racket macros with Arc macros is one of the advanced cases; I expect that to get the interactions right, it may be necessary to do some explicit conversions between "source" types, as well as being proficient in both languages' approaches to hygiene.
- As far as a module system goes, Arc traditionally has no way to allow multiple pieces of code to see different bindings of the same top-level variable name. (Framewarc and I believe Arc/Nu have approaches to this, but they involve writing all macros differently than before.) Hence, two Arc libraries might fail to be usable together due to mutual clobbering of the same variable, which isn't a typical situation with Racket modules. If and when modular techniques catch on for Arc, those techniques may still be challenging to reconcile with Racket's techniques, so a `#lang anarki`-based library may still not present a very familiar interface to other Racket ecosystem programmers.
- Arc is a lisp-N with its various namespaces, like `setforms`, `defcall`, and `coerce`, stored as entries in global hash tables. Racket is a lisp-N with its various namespaces, like `set!` transformers and `match` expanders, stored on the same compile-time object by implementing multiple interfaces (structure type properties/generic interfaces). This is one particular place where reconciling the modularity approaches may be difficult.
- Racket's compiler tooling tries to determine statically what files a file depends on, both so that `raco make` and `raco setup` can recompile the file if its dependencies have changed and so that if a file is bundled up with `raco distribute` or `raco exe`, its dependencies are included in the bundle. This is something I didn't tackle at all with `#lang anarki`. Things like `load` and `dynamic-require` make it difficult to keep track of dependencies statically this way.
The design I chose for `#lang anarki` was basically to get something tolerable working with as little effort as possible. There's probably a lot of room for improvement, so I didn't consider it stable enough to show off in the `anarki` Racket package documentation.
I think a somewhat better approach would involve:
- Giving Anarki a read-time namespace system like Common Lisp's, to solve Arc's inadvertent name clobbering issues (while potentially still permitting intentional clobbering).
- Going ahead and macroexpanding a whole `#lang anarki` file at compile time, evaluating nothing but the `mac` definitions at first, even though that's not the usual Arc `load` semantics. User-defined macros can sometimes be slow, making it worthwhile to expand them before producing the .zo. Only so much Arc code actually cares about running certain expressions before others are expanded, and those cases can use a CL-style `eval-when` approach.
- Assembling static information about the module just by using `eval-when` compile-time mutable definitions. Then we could dispense with `(:provide ...)` and just have a global mutable table that collects all the exports of the current module.
I think Arc's load time is pretty typical of uncompiled Racket code. In Racket, people can speed up their development process by using `raco setup --pkgs my-package` to compile their libraries or `raco make filename.rkt`
to compile individual files. That might be harder to do for Arc, especially the Arc REPL, since Arc's macroexpansion is interleaved with run-time side effects.
I'm not sure how to answer that. If you're asking what the easiest way would be to speed up initialization, it would be to load fewer libraries. Since it's an anarchic fork, we've all added lots of ideas that are good in principle but not always needed.
It's a new project. However, Hubski has been using postgres for a few years, but it's currently an amalgam of Arc and Racket. I didn't write the db code, and am trying to start fresh. The Hubski racket connection looks like:
BTW, I am using the domain xyrth.com with a nginx reverse proxy. I just had the thought it may be my nginx configuration. It probably needs to listen explicity on 5432? Here's my current sites-enabled:
Is this on hubski.com? Have y'all been using postgres there for a long time? Or are you trying it for the first time?
Can you show what command you're running to check that you "can connect to the db from the Arc command line"? I wonder if the app is able to connect but not disconnecting for some reason. That would explain the "thread took too long" timeout message.
If you've written some custom code to connect to postgres, that'd be helpful to look at as well.
Yes, it's the fork that's been maintained by the people here at Arc Forum. :) I recommend it.
Whether you want the community additions or not, you might also be interested in the "official" and "stable" branches of that repo. The "official" branch tracks the official .tar releases that you'd find on the front page here, and the "stable" branch tracks the official releases plus a few small bug fixes and quality-of-life improvements.
Penknife really was just a pile of code before now. I originally wrote it on my commute using an Android phone. Now I've finally gotten around to giving it a readme, a .gitignore, and all that. Whew. :)
There was another language I called Penknife a couple of years later because I thought of it as having the same set of design goals, so technically this one's more like Penknife Mk. I. That's what I think I'll call it.
Something about Penknife Mk. I that I often think back to is its approach to macro hygiene.
Its syntax is primarily string-based, but with the ability for other values to be embedded inside the strings. It has a kind of quasiquotaton that surrounds the quoted section with an object wrapper that the macroexpander recognizes. When the macroexpander expands that expression, it switches over to using the scope at the macro's definition site. Interpolations in the quasiquotation are surrounded with another wrapper that causes the macroexpander to switch back to the caller's scope.
When the macro is defined, its lexical scope is captured and carried on the macro's binding. That way, it's the namespace that holds other namespaces inside it. The syntax trees don't have to hold namespaces; they can just hold paths to traverse the namespace hierarchy.
I still think of this as a nice sweet spot; the code has a context-independent enough identity to be compiled, while the namespaces are mutable enough to allow REPL interaction.
Racket's sets-of-scopes approach to hygiene is mature and supports advanced features like local macros and local definition blocks, but I think of it as kind of sloppy. It spray paints one piece of information over the whole syntax tree just to mark that a variable is in scope. This approach could be handy in cases where variable scopes overlap with each other in non-hierarchical ways -- where sometimes one variable is in scope, sometimes the other, and sometimes both -- but I feel like Penknife's wrapper objects are a much tidier model of the typical hierarchical structure. There've been many moments in my Racket programming when I would have liked Racket to have Penknife Mk. I's approach to hygiene instead.