Repository navigation
Note
Releases before v1.0.0 are beta releases: options and generated output may still change between minor versions. Everything that ships is tested against real databases and already used in production.
v0.9.0 - 2026-08-16
Added
- #248 MySQL support: the new
sql_drivervaluesasyncmyandpymysqlgenerate code for asyncmy (asyncio) and PyMySQL (synchronous), usingengine: "mysql"and the same?placeholders as the SQLite drivers - the plugin rewrites them to the drivers' pyformat%sat generation time. InlineENUM(andSET) columns generateenums.pyclasses,tinyint(1)maps tobool,timetodatetime.timedelta, and returnedjsoncolumns staystrlike on every other driver.:execlastidreturnscursor.lastrowid(Nonewhen nothing was inserted);:copyfromstays PostgreSQL-only. Reused parameters (sqlc.argorsqlc.sliceused at several sites) merge into one function argument even though MySQL binds every occurrence separately. (Rayakame)
Fixed
- #248 With
omit_unused_models: true, an enum referenced only through an overridden column no longer breaks generation in either direction: an overridden enum parameter kept itsenums.X(...)conversion while the class was filtered away (aNameErrorat import), and an enum used only by overridden return columns was retained as a deadenums.pynothing imports. (Rayakame) - #248 Query modules whose only queries return whole model rows (no
:many) no longer import the row's column-type modules (datetime,decimal, ...) into theirTYPE_CHECKINGblock: nothing in such a module spells those types, and ruff flagged the imports as unused on generated output. (Rayakame) - #257 Arguments of a reused
sqlc.sliceare no longer misordered when the query also contains a backtick- or bracket-quoted identifier holding a?(SQLite family), and a slice whose name contains%now expands correctly on the MySQL drivers - its marker was doubled in the SQL constant but not in the replacement the generated code performs, so the expansion silently matched nothing. (Rayakame) - #260 SQLite-family queries that combine a
sqlc.slicewith a named argument (sqlc.arg,@name,:name) now bind their arguments to the right columns. sqlc numbers every placeholder of such a query (?1,?2, ...) and assumes the slice marker occupies one slot, but the marker expands to one slot per element, so every index after it named the wrong slot: the query returned wrong rows, or the driver rejected it. The generated SQL constant now carries bare?placeholders and the arguments are passed in text order, which is correct for any slice length including an empty one. (Rayakame) - #262 A backslash in a query now survives into the generated SQL constant. The SQL was written into a plain Python string literal, so Python read every escape it recognised:
LIKE 'a\tb'reached the server with a tab,'C:\name'with a newline, and an escape Python does not know (REGEXP '\d+') raised a SyntaxWarning that is slated to become a SyntaxError. Query text holding a backslash - and the docstring that repeats it - is now emitted as a raw literal. Affects every driver. (Rayakame) - #263 A query containing
"""no longer produces a module that does not parse. The SQL went into a"""literal, so three consecutive double quotes in the query - from a string literal, or from a quoted identifier ending in a quote - closed the constant early and left the rest of the file as stray syntax. The constant and the docstring now pick a delimiter the query does not contain. A query holding both"""and'''has no block spelling left: the constant falls back to an escaped one-liner, and the docstring, which cannot be escaped without tripping ruff D301, leaves the SQL out. (Rayakame)
plugins:
- name: python
wasm:
url: https://github.com/rayakame/sqlc-gen-better-python/releases/download/v0.9.0/sqlc-gen-better-python.wasm
sha256: d1787aa32e61f2e73c81a4f93b3e5a9beeec918952cff3183fb96313057f58a8