fix(sqlite): support INSERT IGNORE and offset-only limits [CODEX]

[russellromney]

Summary

Fix SQLite-targeted transpilation for three cases that currently produce SQL SQLite rejects:

• MySQL INSERT IGNORE ... -> SQLite INSERT OR IGNORE ...
• Postgres LIMIT ALL -> no SQLite LIMIT clause
• Postgres offset-only OFFSET n -> SQLite LIMIT -1 OFFSET n

These are represented directly in SQLite syntax rather than unsupported semantics.

Examples

sqlglot.transpile(
    "INSERT IGNORE INTO t (id, a) VALUES (1, 2)",
    read="mysql",
    write="sqlite",
)[0]
# before: INSERT IGNORE INTO t (id, a) VALUES (1, 2)
# after:  INSERT OR IGNORE INTO t (id, a) VALUES (1, 2)

sqlglot.transpile("SELECT x FROM t LIMIT ALL", read="postgres", write="sqlite")[0]
# before: SELECT x FROM t LIMIT ALL
# after:  SELECT x FROM t

sqlglot.transpile("SELECT x FROM t OFFSET 1", read="postgres", write="sqlite")[0]
# before: SELECT x FROM t OFFSET 1
# after:  SELECT x FROM t LIMIT -1 OFFSET 1

I also checked the combined case:

sqlglot.transpile("SELECT x FROM t LIMIT ALL OFFSET 1", read="postgres", write="sqlite")[0]
# after: SELECT x FROM t LIMIT -1 OFFSET 1

Why

SQLite uses INSERT OR IGNORE for this conflict behavior, not MySQL's INSERT IGNORE. SQLite's INSERT docs describe the INSERT OR action form for choosing a constraint conflict algorithm: https://www.sqlite.org/lang_insert.html

SQLite also requires OFFSET to appear as part of a LIMIT clause; LIMIT -1 OFFSET n is the common SQLite spelling for offset without an upper row bound. Related prior art/background on emulating MySQL conflict behavior in other dialects: https://stackoverflow.com/questions/1009584/how-to-emulate-insert-ignore-and-on-duplicate-key-update-sql-merge-with-po

I searched for existing SQLGlot issues/PRs mentioning these exact SQLite-target gaps and did not find an existing report.

Tests

• python3 -m unittest tests.dialects.test_sqlite.TestSQLite.test_sqlite
• python3 -m unittest tests.dialects.test_dialect.TestDialect.test_limit tests.dialects.test_mysql.TestMySQL.test_mysql
• python3 -m unittest tests.dialects.test_sqlite
• python3 -m ruff check sqlglot/generators/sqlite.py tests/dialects/test_sqlite.py
• git diff --check

[georgesittas]

Thank you for the PR!

[georgesittas]

Hmm... this is fine for now, but looks like a code smell to me. Having two ways to represent the IGNORE semantics doesn't look right. Perhaps something to clean up in the future...

[russellromney]

good catch here. let me think about a better way to do this

[georgesittas]

This is fine for now, let's not deal with it in this PR. Just wanted to leave a note for future reference.

[russellromney]

All feedback addressed, thanks. Should have checked tests a little more thoroughly