LINT-KEYWORD substring-matches inside strings and comments (`// Note:` warns "Keyword No"), and reports only the first hit per keyword

[logbie]

Summary

LINT-KEYWORD does a raw substring search over the whole source file, so it fires on text inside string literals, inside comments, and inside ordinary words. store s as "MNOP" is reported as Keyword 'NO' should be lowercase; so is a comment reading // Note: or // TODO:. Because source.find returns only the first hit per keyword, the rule is also unable to report genuine violations after the first.

Reproduction

store s as "MNOP"
display s

Command:

wfl --lint l2_lint_keyword_in_string.wfl

Expected

Rule LINT-KEYWORD found 0 issues, exit 0. There is no keyword in this program —
MNOP is the content of a string literal. The rule's own description is
"Keywords should be lowercase" (src/linter/mod.rs:236), and a sequence of
characters inside a string literal is not a keyword.

Actual

Rule LINT-KEYWORD found 1 issues
Linting complete. Found 1 issues
Lint warnings:
warning[LINT-KEYWORD]: Keyword 'NO' should be lowercase
 = Change to 'no'

Exit code: 1. Control: store s as "MXYP" → 0 issues, confirming the match is
the substring NO inside MNOP.

Scope is wider than string literals — it also hits comments and prose

Each row is a two-line program with the shown line added; all are valid, correct
WFL:

line	warnings
// Note: this explains the next step	1 — Keyword 'No' should be lowercase
// TODO: revisit	1 — Keyword 'TO' should be lowercase
store greeting as "Nothing to see"	2 — Keyword 'No', Keyword 'Nothing'
store label as "Ineligible"	1 — Keyword 'In' should be lowercase

So an ordinary English comment beginning "Note" or "TODO", and any capitalized
word containing a keyword substring, produce a lint warning and a nonzero exit.

Root cause

KeywordCasingRule::apply (src/linter/mod.rs:229-326) ignores the parsed AST
(_program: &Program, :242) and searches the raw source text:

// src/linter/mod.rs:292
if let Some(pos) = source.find(&uppercase_keyword) { ... }
// src/linter/mod.rs:308
if let Some(pos) = source.find(&mixed_case_keyword) { ... }

for each of 26 keywords, in both UPPERCASE and Mixedcase forms. There is no
tokenization, no word-boundary check, and no exclusion of string literals or
comments. The Mixedcase variant is what makes No, To, In, Nothing,
For, Each, End, Check, If, Count, From match ordinary English —
those are common English words and prefixes, so any file with prose comments is
likely to trip several.

It is also a false-negative

source.find returns only the first occurrence, so at most one diagnostic is
produced per keyword per casing — 52 possible warnings in a file of any size. A
file containing fifty genuine STORE keywords reports one. The rule can neither
avoid reporting non-keywords nor finish reporting real ones.

Suggested fix

Use the token stream (or the AST the rule already receives) instead of
source.find, and report a keyword only where the lexer actually produced a
keyword token. That fixes the false positives and the truncation together. If a
text-level implementation must be kept as an interim step, at minimum skip string
literals and comment text and require word boundaries — but the mixed-case English
collisions (No, To, In) argue for doing it properly on tokens.

Related

Same root-cause family as the companion LINT-INDENT issue in this batch: four of
the six rules receive the parsed Program and ignore it (_program at
src/linter/mod.rs:174, :242, :341, :386). For TrailingWhitespaceRule and
LineLengthRule that is legitimate. For IndentationRule and this rule it is the
direct cause of the defect. Independently fixable; noted so they are not triaged
as unrelated.

Environment

• wfl --version: WebFirst Language (WFL) version 26.8.4
• binary: system install C:\Program Files\wfl\bin\wfl.exe
• commit: c277d8f
• OS: Windows 11 Pro 10.0.26200
• build: release

Also reproduces identically on the repo build G:\repos\wfl\target\release\wfl.exe
(version 26.8.2, 1 issue) — not a regression.
No .wflcfg in scope for the repro.

Context

Found while porting G:/repos/JShrink/src/JShrink/Minifier.php (a 738-line PHP
JavaScript minifier) to WFL. The port needs an uppercase-alphabet constant to test
PHP's \w character class:

store word_chars as "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789_"

which contains MNOP and therefore reports Keyword 'NO' should be lowercase —
verified, 1 issue. The string cannot be changed; it is the definition of the
character class.

This did not block the port (the program runs and analyzes clean; nothing was
stubbed), but combined with the LINT-INDENT issue it means --lint cannot be
used as a gate on this codebase, and there is no in-language way to silence a
single warning.