tonyg

Wellll not two years, no, a couple-few weeks of initial work and then, embarrassingly, it sat idle for two years. As to why: at least in part because one of the best ways to learn is to do.

This looks great!

Thanks so much!

In many benchmarks, Racket is significantly faster than many other Scheme implementations, including Chicken: https://ecraven.github.io/r7rs-benchmarks/ This is because it is now based on Chez.

It is certainly not compact, though.

New? It stretches back industrially to the WS-* stuff in the early 00s (this is where the term originated), and on the academic side has roots in process calculi from the 90s and session types from the late 00s and early 10s.

Absolute slop. Interesting that the end result, the place where Claude ended up, is similar to a (human-authored, windmill-tilting) project of mine, https://preserves.dev/ . Though with a much prettier website and far, far, far more words.

That's one of the reasons RFC 9485, "I-Regexp: An Interoperable Regular Expression Format", is important.

https://datatracker.ietf.org/doc/html/rfc9485

*Mitchell Wand

I will not. I refuse

Is .map specialcased or do user functions accepting callbacks work the same way? Because you could do the Scott-Mogensen thing of #ifTrue:ifFalse: if so, dualizing the control-flow decision making, offering a menu of choices/continuations.

(1998). Java existed, but neither Scala nor Java-with-generics did.

From the conclusion:

"We have presented a programming protocol, Extensible Visitor, that can be used to construct systems with extensible recursive data domains and toolkits. It is a novel combination of the functional and object-oriented programming styles that draws on the strengths of each. The object-oriented style is essential to achieve extensibility along the data dimension, yet tools are organized in a functional fashion, enabling extensibility in the functional dimension. Systems based on the Extensible Visitor can be extended without modification to existing code or recompilation (which is an increasingly important concern)."

I don't think that's right - IIRC it used to be possible to write out a file, if loaded from a file:// URL, directly from JavaScript. Then that ability got nobbled because security (justifiable) without properly thinking through a good alternative (not justifiable). I mourn the loss of the ability, TiddlyWiki was in a class of its own and there should have been many more systems inspired by its design. Alas.

ETA: Wikipedia has reminded me the feature was called UniversalXPConnect, and it was a Firefox thing and wasn't cross-browser. It still sucks that it was removed without sensible replacement.

> Undermining the credibility of computer science research is the best possible outcome for the field, since the institution in its current form does not deserve the credibility that it has.

Horseshit. This might be true for AI research (and even there that's an awfully broad brush you're using, mate), but it's certainly not true for other areas of computer science.

> Static analysis can tell what forms are invoking an fexpr and which are function calls. It's not got different from knowing which are macros. That problem can be solved.

I don't think this is the case. Consider Kernel's

  ($lambda (f) (f (+ 3 4)))

Is `f` a fexpr or a closure? We cannot know until runtime.

There are no string keys in the Person example above. You could add some, though, or use numbers instead with the same host-language API:

  Person = <person @name String @address Address>

as above, or

  Person = <person {
    @name "name": String
    @address "address": Address
  }>

or

  Person = {
    @name 1: String
    @address 2: Address
  }

etc. all produce the same host-language record, e.g. in TypeScript

  export type Person = {
    name: String,
    address: Address,
  };

Thank you! I wonder if something a bit contrived such as small-caps could help. I'll experiment.

Date and datetime are left to convention: https://preserves.dev/conventions.html#dates-and-times

Decimals I'm on the fence about. Some discussion here: https://gitlab.com/preserves/preserves/-/issues/10

Yeah I struggle with "Preserve" vs "Preserves" sometimes. Was there something in particular that struck you as unfortunate, though?

XML would make a fine choice. It lacks atomic data types other than text, and compound data types other than sequences, unless you count element attributes, which are in a kind of awkward position because of the historical development of the language. Preserves has a richer suite of primitive data types and decomposes XML's elements into separate notions of map, sequence, and tagged value.

Thanks. Yes there are: see sections 15, 16 and 17 of https://synit.org/book/, where Preserves, Preserves Schemas, and the Syndicated Actor Model make a reactive replacement system layer for linux (essentially an alternative to systemd)

The version number is the schema language version, not the version of the collection of types described in the file.

The schema language is extensible/evolvable in that pattern matching ignores extra entries in a sequence and extra key/value pairs in a dictionary. So you could have a "version 1" of a schema with

  Person = <person @name String> .

and a "version 2" with

  Person = @v2 <person @name String @address Address>
         / @v1 <person @name String> .

Then, Person.v2 from "version 2" would be parseable by Person from "version 1", and Person from "version 1" would parse using "version 2" as a Person.v1.

The schema language is in production but the design is still a work in progress and I expect more changes before a 1.0 release of the schema language.

(The schema language is completely separate from the preserves data model, by the way -- one could imagine other schema languages being used instead/as well)

Yeah EDN is quite similar. Preserves has no nil, allows any value as a tag, gets into the weeds more on when strings are equal or not, doesn't distinguish lists and vectors, and doesn't require each kind of tagged element to define an equivalence. And it has annotations (vs EDN's comments) and embedded values.

https://gitlab.com/preserves/preserves/-/blob/main/preserves... -- crude but effective! (I use it all the time)

No, a record is a tagged (sequence of) value(s).

  <tag v1 v2 v3>

If you put a single dictionary-valued "field" in a record, you get a variation with named fields

  <tag {
    field1: value1
    field2: value2
    field3: value3
  }>

Records have positional "fields" because of the Scheme heritage of the design.

--

Re bytestring -- yes there are some concessions to real machines/languages in there that aren't absolutely required. Other examples include booleans and strings, which could have been <true> and <false> and <string [65 66 67]> etc respectively.

There's a little more on this topic in footnote 2 on the "conventions" page: https://preserves.dev/conventions.html#fn:why-dictionaries

Thank you!

Do you have a link for CSTML, please? Googling is showing a bunch of possibilities none of which look quite relevant enough to be right...

Python remembers order, and exposes it in its iterations, but doesn't use it in its equivalence over dictionaries (== semantics).

(ETA: What are you quoting there? I don't think that text appears on the Preserves site) (ETA2: Ah, it's the tutorial. Cool)

Useful if you have a JSON-keyed table, for example: JSON lacks a useful (standardised) equivalence relation, meaning you get weak and/or implementation-specific guarantees about how key lookup works. Equivalences were the motivation for developing Preserves: I was (and am) working on a generalized approach to messaging middleware, you might say, meaning that things like "patterns over values" and "filters" and "value-keyed tables" are all things I need to talk about. (This all comes out of RabbitMQ/AMQP thinking back in the day and my PhD-and-after work subsequently.)

Or, in "quick reference card" form: https://preserves.dev/cheatsheet.html

The syntax isn't the most interesting part though; the thing that distinguishes it from most other data languages out there is that it has semantics (= a rigorous definition of when values are equal and when they aren't). So you can use Preserves semantics with JSON syntax (a subset of Preserves' text syntax) as one way of getting actually-meaningful JSON.

Plus, comments (and other annotations) ;-)

> It confuses folks into thinking that there is some garbage collection that isn’t tracing. There’s no such thing.

It is standard to consider reference counting as garbage collection.

Bacon, D.F., Cheng, P. and Rajan, V.T. 2004. A unified theory of garbage collection. ACM SIGPLAN Notices. 39, 10 (Oct. 2004), 50–68. DOI: https://doi.org/10.1145/1035292.1028982

Abstract:

Tracing and reference counting are uniformly viewed as being fundamentally different approaches to garbage collection that possess very distinct performance properties. [...] Using this framework, we show that all high-performance collectors (for example, deferred reference counting and generational collection) are in fact hybrids of tracing and reference counting.