mirror of
https://github.com/nushell/nushell.git
synced 2026-11-23 01:48:49 +08:00
# Description Fixes #18807. ``` > echo $"('" "')" ``` should print `" "` but the string cannot even be closed: the lexer ends the interpolated string at the `"` inside the parenthesized subexpression, so the line lexes as `$"('"` followed by `"')"` and the REPL keeps asking for more input. `lex_item` treats a quoted region as opaque until the closing quote, which is right for plain strings and wrong for interpolated ones: there an unescaped `(` starts a subexpression where quotes and parens nest. The parser side already knows this. `parse_string_interpolation` scans the token with a delimiter stack, but it never sees the full text because the lexer hands it a truncated token. This change teaches the lexer the same delimiter matching the parser applies, so the token ends exactly where the parser will end the string: - a `"` or `'` opening directly after `$` marks the string as interpolated (backticks never interpolate) - an unescaped `(` in such a string opens a subexpression; an escaped `\(` does not, and `$'` has no escapes, both matching the parser's rules - while a subexpression is open, the innermost open delimiter decides what matches: a quote is only closed by the same quote, parens nest, and the string's own closing quote cannot end the string - at end of input an open subexpression is reported from its oldest open delimiter, so `$"(1` points at the `(` and asks for `)`. In the `$"foo (2 + 3"` typo the trailing quote now opens a nested string during lexing, and reporting the oldest delimiter keeps the error on the forgotten `(`, which is what the existing `unclosed_interpolation_subexpression` repl test expects Behavior checked against an unpatched build: - `--flag=$"(1 + 2)"` still interpolates (the `$` is mid token there, which is why the check is `previous char is $` rather than `token starts with $`) - the odd bareword form `a$"b(1)"c` still gives `a$"b1"c` - `^$"($nu.current-exe)"` still runs as an external head - `$"\('not an expr'\)"` still prints `('not an expr')` # User-Facing Changes Quotes can now be used inside `(...)` subexpressions of interpolated strings: `$"('" "')"` prints `" "`. Unclosed interpolated strings report the oldest unclosed delimiter of the subexpression instead of a misleading one. # Tests + Formatting Two new lexer test groups in `test_lex.rs`: five inputs that must lex to a single item token (the issue's expression, a double quote inside the subexpression, the `$'` form, nested parens, and escaped parens), and three unclosed inputs asserting which delimiter is reported and where it opened. Four new `eval_value_eq` cases in `tests/eval/mod.rs`, including the issue's expression evaluating to `" "`. The issue's lex case, the unclosed cases, and the issue's eval case all fail on current main and pass with this change; the remaining cases pass on both and guard against regressions. - `cargo fmt --all -- --check` clean - `cargo clippy --workspace --all-targets -- -D warnings` clean - suites: nu-parser 359 (includes the new cases), nu-command 3630, nu-lsp 785, nu-cli 171 (23 file completion tests fail identically with and without this change on my machine, environment related), repo tests 2026 including all repl string tests - `cargo check -p nu-parser --target wasm32-unknown-unknown --no-default-features` compiles # After Submitting ## Release notes summary - What our users need to know Quotes inside `(...)` subexpressions of interpolated strings now work: `$"('" "')"` prints `" "` instead of becoming an unclosable string.