praxis-syntax

Token and syntax-node definitions for the Praxis language.

The syntax tree is rowan-backed and lossless — trivia and all — so the language server can rewrite a span without disturbing the comments and whitespace around it. This crate contributes the vocabulary that tree is made of; praxis-parser is what builds one.

What it provides

• SyntaxKind — the single enum covering tokens, trivia and tree nodes.
• PraxisLanguage, SyntaxNode, SyntaxToken, SyntaxElement — the rowan Language impl and its node aliases.
• span_bridge — the only place praxis_source::Span and rowan::TextRange meet. The Praxis Span stays the source of truth for diagnostics.

It also holds the scanners that more than one consumer needs to agree about, in one copy each: the identifier character class, the text-literal decoder, the digit-separator rule, the rule for where an interpolated "…" literal ends, and the rule for where a backtick template ends. The lexer and the input parser both read the last of those, which is what stops one of them from accepting a template the other cannot.

Part of Praxis

Praxis is a small, statically typed, garbage-collected language for Advent of Code-style puzzles: the input parser is part of the language, types are inferred rather than written, and a program that falls over hands you its state instead of a stack trace.

To use the language, install praxis-cli — it provides the praxis binary. The repository has the book, the design document and the decision records.

This crate is one stage of that compiler, published so the pipeline is inspectable and so praxis-cli can be built from the registry. Its API tracks what the compiler needs and is not a stable platform for outside consumers.

Praxis was written with large language models against a human design. The repository's README says what that means for the license.

Licensed under either of Apache License 2.0 or the MIT license, at your option.