Add table literal sugar for lists of records

[haydenflinner]

Sorry about the original contents of this post, Cursor jumped a little ahead -- I ran out of Claude Code earlier today and have been trying the new old tool again :-) The below is human edited.

I had Claude add a table syntax to Roc's syntax, because I like tables, and I wondered if it was possible without knowing Roc to modify it.

Why tables matter

I think tables are critical to the future of computing. Some references:

• A Data-Centric Introduction to Computing with Shriram Krishnamurthi (jump to 11:24)

or in text form:

• A Data-Centric Introduction to Computing, especially Our Perspective on Data — tabular data as the ideal way to think about collections of records, and one of the oldest:

The homepage todos example (couldn't find the source) is the kind of thing that currently repeats every field name on every row. A table literal makes that more convenient, IMO:

Suggested syntax (taken from Pyret)

Homepage writes:

    todos = [
        { name: "Learn Roc",       status: Done },
        { name: "Buy groceries",   status: Done },
        { name: "Write blog post", status: InProgress },
        { name: "Call mom",        status: NotStarted },
    ]

The suggested change in idiom is:

todos = table name, status {
    "Learn Roc",       Done,
    "Buy groceries",   Done,
    "Write blog post", InProgress,
    "Call mom",        NotStarted,
}

The most obvious difference compared to Pyret's choice is the lack of row. Pyret's row: would enable a single logical row span multiple source lines, I guess. Without it, my intuition is that each table row can only be one row in the source text.
Claude says that in the impl vibecoded, newlines at table-body depth end the row; newlines inside (), [], {}, or a lambda do not.
I don't have an opinion on that tradeoff, it's just what fell out.

table is a contextual keyword (table(x) stays valid, i.e. table is not reserved). Columns may take types (table name : Str, age : U8 { ... }) so numerals pick up the column type instead of defaulting to Dec.

Implementation

This PR is an allegedly working implementation of the suggested syntax that Claude wrote. I have not personally used it yet, I wanted to discuss the idea more before sinking in more time. I had claude write an impl mostly to see if it was possible.

It claims that this is tested:

parse + format + canonicalize desugar, with snapshots. Invalid tables (duplicate columns, width mismatch, zero columns) fail at parse so later stages never see a bad table.

[greptile-apps]

Greptile Summary

The PR adds table-literal syntax and lowers it into existing lists of records, with parser, formatter, canonicalizer, LSP, documentation, and snapshot support.

• Adds contextual recognition and row-oriented parsing for table literals.
• Adds AST and node-store representations plus list-of-records canonicalization.
• Adds formatting, selection-range traversal, diagnostics, documentation, and snapshots.

Confidence Score: 4/5

The PR is not yet safe to merge because a valid complete-width table row followed by a newline and comma is still parsed as malformed.

The newline branch finishes the row without consuming the comma, and table_row_next immediately starts the next cell at that same comma token, leaving the previously reported parser failure outstanding.

Files Needing Attention: src/parse/Parser.zig

Important Files Changed

Filename	Overview
src/parse/Parser.zig	Adds contextual table recognition and row parsing, but the previously reported complete-row newline-before-comma separator failure remains.
src/canonicalize/Can.zig	Desugars validated table literals into lists of records and applies optional column annotations.
src/fmt/fmt.zig	Formats table headers and rows while integrating table expressions into multiline layout decisions.
src/parse/AST.zig	Adds table AST nodes, diagnostics, and S-expression serialization.
src/parse/NodeStore.zig	Adds storage, scratch spans, and retrieval APIs for table columns and rows.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
    Source[Table literal source] --> Parser[Parser row and column handling]
    Parser --> AST[Table AST]
    AST --> Canonicalizer[Canonicalization]
    Canonicalizer --> Records[List of records]
    AST --> Formatter[Formatter]
    AST --> LSP[LSP selection ranges]

Loading

_{Reviews (3): Last reviewed commit: "Treat table as a table literal only when..." | Re-trigger Greptile}