Last Lilith commit I promise

This commit is contained in:
Reid 'arrdem' McKenzie 2021-08-23 10:10:03 -06:00
parent db9c0f7bf8
commit 7bff95fe32
2 changed files with 57 additions and 19 deletions

View file

@ -51,20 +51,28 @@ def foo(a, b):
(+ a b)) (+ a b))
``` ```
But these are both deeply syntactically limited. As we're thinking about "comments" and "documentation", docstrings are a lot better than reader discard comments because we can at least do something with them.
DSLs inside of strings .... are inside of strings. But they're still very much second class in that they presuppose "documentation" for an artifact or a program is limited to API docs.
Special separate parsers are required to work with or on them.
Users have to write those DSLs within the context of embedded strings.
It just doesn't work well.
What if we wanted to put other media in a "source" file? To the extent docstrings are overloaded for other DSLs like doctests, those DSLs are extremely second if not third class.
What if the "doctest" grammar was something you could just type at the top level? There's no syntax support for them in the language, a separate parser must be used, editors don't understand when you embed code or other languages in strings ... the entire experience is just limited.
Or you could embed YAML or SQL into a file without special escaping airs?
### Literate Programming (tangle/weave)
Old-school Knuth [literate programming](https://en.wikipedia.org/wiki/Literate_programming) is AN answer for how to center documentation over the program.
In the LP world, your program isn't so much a program as it is a book containing code blocks that can be "woven" together into a program a compiler can eat.
From a literature perspective, this is a great idea.
Programs are absolutely literature, and being able to produce a fully fledged and formatted book out of a "program" rather than just an API listing out of a "program" is great.
Unfortunately tangle/weave systems don't align well with language boundaries.
They deliberately leap over them, and in doing so disable language tooling.
So what if we could build a language that natively has something like tangle/weave?
Make comments or perhaps more accurately document DSLs "first-class" and values that can reference terms from the language, and in turn be referenced by the language.
## Enter Lilith ## Enter Lilith
Lilith is a sketch at what if you took the ideas from literate programming (having fragments of text from which programs are composed) but deliberately DID NOT privilege the "source" for a "document" over the "source" for the "program". Lilith is a sketch at what if you took the ideas from literate programming (having fragments of text from which programs are composed) but deliberately DID NOT privilege the "source" for a "document" over the "source" for the "program".
Documents and programs could be co-equal and co-resident artifacts. Documents, DSLs and programs could be co-equal and co-resident artifacts.
To achieve this vision, Lilith uses a context sensitive block-prefixed syntax which SHOULD be uncommon enough not to involve collisions with other languages. To achieve this vision, Lilith uses a context sensitive block-prefixed syntax which SHOULD be uncommon enough not to involve collisions with other languages.
@ -136,19 +144,25 @@ $ lil src/lark/designdoc.lil
.... ....
``` ```
## Hackweek disclaimers ### Presentation videos
Python packaging is the devil. - [short walkthrough](https://twitter.com/arrdem/status/1429486908833292295)
This code was originally developed in [my monorepo](https://github.com/arrdem/source/tree/trunk/projects/lilith) where it is build and tested via a reproducible Bazel setup. - [long walkthrough](https://twitter.com/arrdem/status/1429486908833292295)
I don't want to try and make the judges install Bazel, so `python3 setup.py develop` is the lowest common denominator.
Note that `setup.py install` doesn't work for resource packaging reasons I haven't sorted out.
While this README may be stale of language features, at this time Lilith is mostly concerned with parsing, building a runtime/namespace system of and evaluating fragments. ## Limitations and broken windows
While the machinery is there in the form of a relatively classical lisp `eval[]` operation to implement `if`, `lambda` and other such traditional special forms, that really wasn't the intent of the language syntax.
Most of Lilith's standard library (`src/python/lilith/prelude.lil`) consists of establishing a FFI prelude to Python, which makes the language seem more fully featured than it is.
**Currently missing:** While this Lilith prototype succeeds in making document terms usable from the "host" language (Lilith), it doesn't currently support going the other way.
- `eval[]` A more complete Lilith would probably feature a way to embed Lilith evaluation within non-Lilith languages (ab)using read-time eval and treating document bodies like Python f-strings.
An even more context sensitive parser would be needed to implement this, but that's eminently doable.
The Lilith language itself is a PITA to write.
While `!def[]` is a nice hack, `!def[] lambda` is ... kinda silly.
Especially because there's no way to attach a docstring to a def.
Perhaps a convention `!def[foo, ...]`, `!def[foo.__doc__, ...]` could hack around that, allowing docstrings to be defined separately but it's not thought out.
The Lilith language itself is wildly incomplete.
Being able to piggy-back off of the host Python interpreter has been good for a lot, but ultimately the language is missing a lot of basic lisp-isms:
- `eval[]` (all the machinery is here to do this)
- `apply[]` (although this is trivial to implement) - `apply[]` (although this is trivial to implement)
- `or[]` - `or[]`
- `not[]` - `not[]`
@ -156,6 +170,8 @@ Most of Lilith's standard library (`src/python/lilith/prelude.lil`) consists of
- `let[]` - `let[]`
- `if[]` - `if[]`
The module/namespace/def system is clearly Clojure derived and worked out pretty well, but `!import` can't trigger code loading as presently factored.
## License ## License
This code is copyright Reid D. 'arrdem' McKenzie 2021, published under the terms of the MIT license. This code is copyright Reid D. 'arrdem' McKenzie 2021, published under the terms of the MIT license.

View file

@ -134,3 +134,25 @@ return hasattr
return max return max
!def[round, py] !def[round, py]
return round return round
!def[add, py]
return lambda *args: sum([0] + (args or []))
!def[sub, py]
return lambda x, *ys: if not ys 0 - x else x - sum(ys)
!def[mul, py]
from functools import reduce
return lambda *args: reduce(lambda x, y: x * y, 1, args)
!def[mod, py]
return lambda x, y: x % y
!def[xor, py]
return lambda x, y: x ^ y
!def[shr, py]
return lambda x, y: x >> y
!def[shl, py]
return lambda x, y: x << y