vilterp / datalog-ts Goto Github PK

Repro

• Put some rules in the fiddle that call infinite "virtual" relations like base.lt

Expected

• no errors in console

Actual
Errors in console:

Maybe "error" and "infinite" need to be distinct cases — maybe using a special InfiniteRelationError or something, which the explorer would then not log.

react-simple-code-editor seems to handle this fine, so I must be doing something wrong…

e.g.

foo{a: A} :-
  bar{b: b{b: A}}.

Would show an error as A being unbound in head.

Hi! I like the ideas in https://petevilter.me/post/datalog-typechecking/ a lot, have also been playing with a couple of projects around it.

One indexes Clojure codebases using a Datascript database https://github.com/jpmonettas/clindex

and then writing some exploratory features of an IDE on top of it here https://github.com/jpmonettas/clograms.

Graphviz was removed because it conflicted with react-flow, which was added for the drag-n-drop graph editor.

Not being rendered properly:

Probably an escaping issue somewhere in the parsing / highlighting pipeline

If there are two definitions in scope, both are highlighted as the definition.

Need some logic to choose the closer one, but not sure how to specify this in datalog without a reduce construct or something…

Hi. Let me tell you about some projects very similar to what you are doing.

Let's start with Makam ( https://github.com/astampoulis/makam ). This is prolog/datalog-based language for specifying syntax and semantics of programming languages, for playing around and arguing about programming languages semantics. Makam is implemented in Ocaml.

Makam is a very similar language to your Datalog dialect. For example, here is example from their manual:

typeof (plus A B) tint <- typeof A tint, typeof B tint.

Makam has a lot of advanced features: lambda expressions, ability to very succinct implicit context (i. e. scope) handling (but I personally consider this as a misfeature, I like explicit contexts), special features for list handling (i. e. for dealing with target language's lists), etc.

Makam is so powerful that you can write whole interpreter in it (see details in their blog).

Makam is able to parse target language using PEG ( http://astampoulis.github.io/blog/makam-tutorial-01/ ). Parsing is bidirectorial (!). Parsing is somehow implemented using JS (but I'm not sure exactly how Ocaml parts and JS parts are related).

(Now let me insert one self-advertisement note here: every time I see "bidirectorial parsing" somewhere, I'm very skeptical about it. Very often so-called bidirectorial parsing projects turn out to be not sound. But I wrote a CFG-based Haskell library, that allows you to write bidirectorial parser with actual guaranteed soundness, i. e. my library guarantees bidirectionality. Here is link: https://mail.haskell.org/pipermail/haskell-cafe/2021-July/134217.html . I can show you other my Haskell parsing libs.)

Makam code can be evaluated directly in browser. At link above you can simply click to "Evaluate code" button in the corner and see result of code evaluation. You can edit code, too.

Unfortunately, Makam is not designed for implementing IDE, it is for writing interpreters and arguing about language semantics. So this will be very cool to somehow combine your IDE with great Makam features and its ability to implement full-featured interpreter.

Also Makam is typed.

This article https://www.tweag.io/blog/2019-11-28-PCF-makam-spec/ can be seen as simple introduction to Makam.

Makam is instance of broad category of Lambda-Prolog. Here is list of various Lambda-Prolog-based projects: http://www.lix.polytechnique.fr/~dale/lProlog/ . Lambda-Prolog can be described as Datalog, but typed, and with lambda expressions.

In particular see this cool demo: https://trymlts.github.io/ . This is playground for toy language MLTS. Interpreter for MLTS is implemented in Lambda-Prolog. And Lambda-Prolog can be run in browser. So this is similar to your demo. In both cases we implement simple language (FP or MLTS) in prolog/datalog-like language (Datalog and Lambda-Prolog, which are very similar). But in your case we see IDE without runner, and they have runner with editor, but without full-featured IDE

Mixup in native implementation between scopeID and rule name

e.g.

foo{} :-
  bar{???}.
bar{x: X} :-
  xxx{}.

expected suggestion x; got nothing

expected:

• variables are only in scope after they're defined

actual:

• in scope where they're defined

example

y should not be an option here, since it's being defined.

Fix

Need to restructure the scope/defn logic.

Can fix by either porting RenameSpan to FP, or removing FP, but I don't want to remove FP until Lingo has multi-arg functions (see #156)

Should be able to rename things even if they don't have any usages. Should have some output here:

Lord knows how to do this in Monaco. lol

disabled in #130

https://github.com/vilterp/datalog-ts/pull/130/files#diff-89ebce8086202c48050e22f7a707349c38feb39217265501a159518da4dc124dR12

These errors shouldn't be there:

Not sure why this isn't finding them:

Via some kind of currying.

Blocked by not supporting a left-recursive grammar…

Last item for parity with hardcoded FP.

We should be seeing an error here, since bar has no attribute y:

foo{x: A} :-
  bar{y: A}.
bar{z: B} :-
  baz{boop: B}.

This seems not to be working… Defined here:

Variables that aren't declared in the head aren't being highlighted as definitions or usages.

It's an interesting question: are they definitions or usages? They're kind of both.

All language workbench examples show errors in the "Abstract Syntax Tree" and "Scope Graph" sections, and shows several errors in the console.

See log lingo-workbench.dev-1654054342512.log, snippets:

tree.tsx:65 Error: not found: astInternal.node
keyBindingsTable.tsx:11 error while checking availability of action "Jump to Definition": Error: not found: scope.Defn
keyBindingsTable.tsx:11 error while checking availability of action "Jump to First Error": Error: not found: tc.Problem
monacoIntegration.ts:149 in token provider: Error: not found: hl.mapping

probably just

-some_record{id: 22}

expected

putting the cursor over a usage causes all the other usages to be highlighted

actual

only the definition is highlighted

example

fix

fix scope.CurrentUser

Repro

• In the LWB, delete rules from datalog editor on the right pane

Expected

• Syntax and variable highlighting stops working on the right

Actual

• Variable highlighting (and some other features?) continue to work on the left, indicating that the DL rules are still active

Calling rule:

hl.Segment{type: T, id: I, span: S, highlight: H} :-
  hl.segmentInt{type: T, id: I, span: S, highlight: H} |
  hl.segmentString{type: T, id: I, span: S, highlight: H} |
  hl.segmentBool{type: T, id: I, span: S, highlight: H} |
  hl.keyword{type: T, id: I, span: S, highlight: H} |
  hl.segmentIdent{type: T, id: I, span: S, highlight: H} |
  hl.segmentSpecialVar{type: T, id: I, span: S, highlight: H} |
  hl.segmentIdentDefnHL{type: T, id: I, span: S, highlight: H} |
  hl.segmentIdentUsageHL{type: T, id: I, span: S, highlight: H} |
  hl.segmentComment{type: T, id: I, span: S, highlight: H}.

Works:

hl.segmentIdentDefnHL{type: "defn", span: S, highlight: true} :-
  ide.CurrentDefnOrDefnOfCurrentVar{span: S}.

Doesn't work:
inlining ide.CurrentDefnOrDefnOfCurrentVar:

hl.segmentIdentDefnHL{type: "defn", span: span{from: F, to: T}, highlight: true} :-
  scope.Defn{span: span{from: F, to: T}} &
  ide.Cursor{idx: CIdx} &
  F <= CIdx & CIdx <= T.

These should work the same. Something's going wrong with bindings in the simple interpreter.

After editing a bit:

Seems like it's desynced or something. Not sure what's going on. Never seen this happen with DL editors. Fixes itself on window reload.

See c95bd72

Had to rename I to RuleID because I had a binding from somewhere else. This shouldn't happen