Fix StringLiteral source spans covering only the closing quote

[revarbat]

Every StringLiteral's position spanned a single " instead of the whole literal — and so did any node wrapping one:

a = "txt";          StringLiteral      span='"'      (should be '"txt"')
f("x,y", b);        PositionalArgument span='"'      (should be '"x,y"')

Cause

A string is matched by several lexer rules — opening quote, content runs, escape sequences, closing quote — and YY_USER_ACTION calls updateLocation() on every one of them, so tokenStart always describes the most recent match. Building the token's location from currentTokenLoc() in the closing-quote rule therefore described just that one character.

STRING is the only multi-rule token, which is why nothing else was affected.

Fix

ParserDriver::stringStart records the opening quote's position when that rule fires; stringTokenLoc() spans from there to the current end.

a = "txt";                    -> "txt"
a = "";                       -> ""
a = "with \"quote\" in it";   -> "with \"quote\" in it"
x = ["a,b","c"];              -> "a,b"  /  "c"

Why it mattered

A consumer slicing source by these offsets could land inside string content. BelfrySCAD's reformatter rewrote a comma inside "h,l,height,length" in BOSL2's isosurface.scad, silently changing the string's value.

That was found only because the reformat was checked for AST-shape preservation across the whole corpus rather than eyeballed — the output looked entirely plausible.

Tests

Four new (630 total): the plain case, empty strings, embedded escaped quotes, escape sequences and commas inside strings; the spans of enclosing nodes, which is how the bug actually did damage; strings inside vectors; and a multi-line string still reporting its opening line rather than its closing one.

🤖 Generated with Claude Code