Support MATLAB-style leading-dot float literals

[nallana]

Motivation

• The lexer currently rejects MATLAB-style numeric literals that omit the leading zero (e.g., .1), causing parsing failures in normal scripts and range expressions like 0:.1:10.
• The fix must be done in the tokenization layer to be robust and avoid brittle downstream hacks while preserving existing dot-based syntax (obj.field, 1..3, .*, etc.).

Description

• Teach tokenize_detailed() in crates/runmat-lexer/src/lib.rs to contextually merge a standalone . into a Token::Float when it is followed by a valid fractional/exponent suffix, by detecting Token::Dot and scanning the remainder for a fractional/exponent sequence and underscores as digit separators.
• Add helpers can_start_leading_dot_float, leading_dot_float_suffix_len, scan_decimal_digits, and scan_float_exponent to safely identify valid leading-dot float suffixes and ensure the dot is not being used for adjacent member/transpose-style syntax.
• Preserve existing behavior for dot-based operators and property access (e.g., obj.field, .*, 1..3) and normalize underscore-separated digits in the emitted float lexeme consistently with other numeric tokens.
• Add regression tests: lexer tests in crates/runmat-lexer/tests/lexer.rs, parser tests in crates/runmat-parser/tests/range_parsing.rs, and runtime/ignition test in crates/runmat-ignition/tests/ranges.rs to cover standalone leading-dot literals, range-step forms like 0:.1:1, and ensure property-access dots are unchanged.

Testing

• Ran cargo fmt to ensure formatting completed successfully.
• Ran cargo test -p runmat-lexer --test lexer and all lexer tests (including new leading-dot tests) passed.
• Ran cargo test -p runmat-parser --test range_parsing and parser range tests (including leading-dot forms) passed.
• Ran cargo test -p runmat-ignition --test ranges and runtime/ignition range tests (including a leading-dot step execution test) passed.

---

Codex Task

---

Note

Medium Risk
Touches core lexing behavior by rewriting . into Float in specific contexts, which could subtly affect dot-based syntax (property access/operators) if the heuristics are wrong; changes are guarded and covered by new tests.

Overview
Adds MATLAB-style leading-dot float support (e.g., .1, .25e-2, .3_5) by teaching tokenize_detailed() to contextually merge a standalone Token::Dot plus a following digit/exponent sequence into a single Token::Float (with underscore normalization), while avoiding rewrites when the dot is adjacent to a prior value/transpose.

Extends test coverage across lexer, parser, and runtime execution to ensure leading-dot floats work in assignments and range steps (e.g., 0:.1:1) and that obj.field-style property access remains tokenized as Ident Dot Ident.

^{Written by Cursor Bugbot for commit ee12f68. This will update automatically on new commits. Configure here.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
runmat	Ready	Preview, Comment, Open in v0	Mar 23, 2026 10:33pm

[gneeri]

You need to fix the clippy things that are causing CI to fail, but otherwise, looks good to me.