fix(parser): accept string literal aliases after AS

[gauravs]

Problem

SQLite, MySQL, and Snowflake accept a string literal as a column or table
alias, but the parser rejects it:

$ echo "select \"deals\".\"record_id\" as 'Record Id', sum(amount) from \"deals\" group by 1" \
    | sqlglot format --read sqlite

# Before (v0.10.9, unpatched):

error: Parser error: Expected identifier, got String ('Record Id') at line 1 col 31

# After:

SELECT
  "deals"."record_id" AS "Record Id",
  SUM(amount)
FROM
  "deals"
GROUP BY
  1

This form is common in machine-generated SQL (e.g. ORMs that quote
display-name aliases as string literals).

Fix

parse_optional_alias now accepts a String token in alias position after
an explicit AS, for both column and table aliases. The alias is an
identifier despite the quoting, so it is normalized to
QuoteStyle::DoubleQuote and the generator re-quotes it in each target
dialect's canonical style:

-- SQLite/Postgres:  SELECT record_id AS "Record Id" FROM deals
-- MySQL:            SELECT record_id AS `Record Id` FROM deals
-- T-SQL:            SELECT record_id AS [Record Id] FROM deals

The implicit form (no AS, e.g. SELECT 'a' 'b') is intentionally not
accepted — adjacent string literals are concatenation in MySQL, so treating
a trailing string as an alias would silently change semantics.

Escaped quotes in the literal are handled by the tokenizer as usual
(AS 'It''s' → AS "It's").

Tests

Six new tests in tests/test_quoted_aliases.rs: column alias, table alias,
alias alongside quoted column refs, escaped inner quote, SQLite→MySQL
transpile, and a regression guard that a trailing string without AS
remains a string expression.

[yigalrozenberg]

Approved — small, focused, and correct. ✅

The key detail that makes this robust: the string-literal alias is returned as a quoted identifier (QuoteStyle::DoubleQuote) rather than an unquoted name, so it flows through write_quoted → QuoteStyle::for_dialect and emits the right quote style for every dialect family, with proper embedded-quote escaping:

• double-quote → PostgreSQL / Oracle / Snowflake / ANSI
• backtick → MySQL / BigQuery / Hive / Spark
• bracket → T-SQL / Fabric

Choosing quoted (not None) is exactly right — it preserves case under PG/Oracle/Snowflake folding. Gating strictly behind an explicit AS is also correct (a trailing string with no AS stays an expression / MySQL concatenation).

Verified locally: all 20 tests in test_quoted_aliases pass, no regressions across the test_transpile / test_dialects / test_expressions suites.

Merging with a few small maintainer follow-ups: broaden coverage with a T-SQL (bracket) test, a Snowflake case, and PG/Oracle transpile targets, and clarify the code comment (T-SQL also uses AS 'col', and the acceptance is intentionally dialect-agnostic / lenient on input). Thanks for the clean, well-tested fix!

[yigalrozenberg]

🎉 Thanks so much for this contribution, @gauravs — a clean, focused, and well-tested fix. Really appreciate the thorough test cases (the escaped-quote and "no AS ⇒ not an alias" ones especially).

This has been merged to master and released in v0.10.11, now live on crates.io:

[dependencies]
sqlglot-rust = "0.10.11"

I added a couple of small maintainer follow-ups on top of your change (commit cab52c9):

• Broadened the tests so all three quote families are exercised end-to-end — T-SQL (AS [Mixed]), Snowflake and PostgreSQL/Oracle transpile targets — in addition to your MySQL/backtick case.
• Clarified the code comment to note that T-SQL also accepts AS 'col', and that the acceptance is intentionally dialect-agnostic (lenient on input, consistent with the existing TRUE/FALSE and DuckDB-keyword branches).

Thanks again for making the parser better! 🙌