.transform() preserves check constraints, refs #762

This commit is contained in:
Simon Willison 2026-08-11 21:59:57 -07:00
commit 2303b80aef
6 changed files with 447 additions and 2 deletions

View file

@ -10,6 +10,7 @@ Unreleased
----------
- New ``table.checks``, ``table.column_checks`` and ``table.table_checks`` introspection properties expose column-level and table-level ``CHECK`` constraints. (:issue:`834`)
- ``table.transform()`` now preserves ``CHECK`` constraints, including comments within their expressions. Renaming a column rewrites identifier references in them without changing string literals or function names. Dropping a column drops a check owned by that column, and raises ``TransformError`` if a remaining check depends on it. (:issue:`762`)
- ``table.transform()`` now works for tables that are referenced by views. Previously the ``ALTER TABLE ... RENAME TO`` step raised ``no such table`` if a view referenced the table being transformed. View definitions are left unchanged - see :ref:`python_api_transform_views`. This also fixes a bug where ``transform(keep_table=...)`` silently rewrote dependent views to point at the frozen backup table instead of the live one. (:issue:`831`)
.. _v3_39_1:

View file

@ -1986,6 +1986,15 @@ A bare column name drops any foreign key that column participates in, including
Renaming a column with ``rename=`` updates any foreign keys that use it, and dropping a column with ``drop=`` also drops any foreign keys it participates in - for a compound foreign key this removes the whole constraint.
.. _python_api_transform_check_constraints:
CHECK constraints
-----------------
``.transform()`` preserves both column-level and table-level ``CHECK`` constraints. If a column is renamed, references to that column in the check expression are renamed too.
A column-level check is removed if its owning column is dropped. Dropping a column referenced by any remaining check raises ``TransformError`` instead of creating an invalid or unexpectedly weakened schema.
.. _python_api_transform_views:
Tables referenced by views

View file

@ -549,3 +549,86 @@ def parse_checks(create_sql: str) -> list[Check]:
_column_checks(item, item_tokens, column, body_start + item_start)
)
return checks
def _is_identifier_token(tokens: list[_Token], index: int) -> bool:
token = tokens[index]
if index + 1 < len(tokens) and tokens[index + 1].text in ("(", "."):
return False
if index and (
tokens[index - 1].is_keyword("COLLATE") or tokens[index - 1].is_keyword("AS")
):
return False
if token.kind == "identifier":
return True
if token.kind != "word" or token.text.upper() in _SQLITE_KEYWORDS:
return False
return True
def check_references_identifier(expression: str, identifier: str) -> bool:
tokens = _meaningful(_lex(expression))
folded = _ascii_fold(identifier)
return any(
_is_identifier_token(tokens, index)
and _ascii_fold(_unquote(token.text)) == folded
for index, token in enumerate(tokens)
)
def check_expression_ends_in_line_comment(expression: str) -> bool:
"""Return True if appended SQL would be swallowed by a ``--`` comment."""
tokens = _lex(expression)
if not tokens:
return False
final = tokens[-1]
return (
final.kind == "comment"
and final.text.startswith("--")
and not final.text.endswith(("\n", "\r"))
)
def _valid_bare_identifier(identifier: str) -> bool:
if not identifier or identifier.upper() in _SQLITE_KEYWORDS:
return False
first = identifier[0]
if not (first == "_" or first.isalpha() or ord(first) >= 0x80):
return False
return all(
char == "_" or char == "$" or char.isalnum() or ord(char) >= 0x80
for char in identifier[1:]
)
def _quote_replacement(original: str, replacement: str) -> str:
if original.startswith('"'):
return '"{}"'.format(replacement.replace('"', '""'))
if original.startswith("`"):
return "`{}`".format(replacement.replace("`", "``"))
if original.startswith("[") and "]" not in replacement:
return f"[{replacement}]"
if _valid_bare_identifier(replacement):
return replacement
return '"{}"'.format(replacement.replace('"', '""'))
def rewrite_check_expression(expression: str, rename: dict[str, str]) -> str:
"""Rewrite column identifiers in a CHECK expression, preserving trivia."""
if not rename:
return expression
tokens = _lex(expression)
meaningful = _meaningful(tokens)
replacements = {_ascii_fold(key): value for key, value in rename.items()}
edits: list[tuple[int, int, str]] = []
for index, token in enumerate(meaningful):
if not _is_identifier_token(meaningful, index):
continue
replacement = replacements.get(_ascii_fold(_unquote(token.text)))
if replacement is not None:
edits.append(
(token.start, token.end, _quote_replacement(token.text, replacement))
)
for start, end, replacement in reversed(edits):
expression = expression[:start] + replacement + expression[end:]
return expression

View file

@ -27,7 +27,14 @@ from typing_extensions import Self
from sqlite_utils.plugins import ensure_plugins_loaded, pm
from .create_table_parser import Check, parse_checks
from .create_table_parser import (
Check,
ParseError,
check_expression_ends_in_line_comment,
check_references_identifier,
parse_checks,
rewrite_check_expression,
)
from .utils import (
OperationalError,
chunks,
@ -86,6 +93,12 @@ def quote_identifier(identifier: str) -> str:
return '"{}"'.format(identifier.replace('"', '""'))
def _check_constraint_sql(check: Check) -> str:
prefix = f"CONSTRAINT {quote_identifier(check.name)} " if check.name else ""
newline = "\n" if check_expression_ends_in_line_comment(check.check) else ""
return f"{prefix}CHECK ({check.check}{newline})"
_IDENTIFIER_CASEFOLD = str.maketrans(
"ABCDEFGHIJKLMNOPQRSTUVWXYZ", "abcdefghijklmnopqrstuvwxyz"
)
@ -1380,6 +1393,7 @@ class Database:
extracts: dict[str, str] | list[str] | None = None,
if_not_exists: bool = False,
strict: bool = False,
_checks: Iterable[Check] | None = None,
) -> str:
"""
Returns the SQL ``CREATE TABLE`` statement for creating the specified table.
@ -1425,6 +1439,19 @@ class Database:
defaults = {resolve_casing(n, columns): v for n, v in (defaults or {}).items()}
if column_order is not None:
column_order = [resolve_casing(c, columns) for c in column_order]
checks = list(_checks or ())
checks_by_column: dict[str, list[Check]] = {}
table_checks: list[Check] = []
for check in checks:
if check.column:
column = resolve_casing(check.column, columns)
if column not in columns:
raise AlterError(
f"No such column for CHECK constraint: {check.column}"
)
checks_by_column.setdefault(column, []).append(check)
else:
table_checks.append(check)
if not columns:
raise ValueError("Tables must have at least one column")
if not all(n in columns for n in not_null):
@ -1481,6 +1508,10 @@ class Database:
column_extras.append(
f"REFERENCES {quote_identifier(fk.other_table)}({quote_identifier(cast(str, fk.other_column))}){_fk_actions_sql(fk)}"
)
column_extras.extend(
_check_constraint_sql(check)
for check in checks_by_column.get(column_name, ())
)
column_type_str = COLUMN_TYPE_MAPPING[column_type]
# Special case for strict tables to map FLOAT to REAL
# Refs https://github.com/simonw/sqlite-utils/issues/644
@ -1520,6 +1551,9 @@ class Database:
actions=_fk_actions_sql(fk),
)
)
column_defs.extend(
f" {_check_constraint_sql(check)}" for check in table_checks
)
columns_sql = ",\n".join(column_defs)
sql = """CREATE TABLE {if_not_exists}{table} (
{columns_sql}{extra_pk}
@ -2677,6 +2711,34 @@ class Table(Queryable):
if column_order is not None:
column_order = [resolve_casing(c, existing_columns) for c in column_order]
try:
existing_checks = self.checks
except ParseError as ex:
raise TransformError(
f"Could not parse CHECK constraints for table {self.name!r}: {ex}"
) from ex
create_table_checks: list[Check] = []
for check in existing_checks:
owner = (
resolve_casing(check.column, existing_columns) if check.column else ""
)
# A column-level constraint disappears with the column that owns it.
if owner and owner in drop:
continue
for dropped_column in drop:
if check_references_identifier(check.check, dropped_column):
raise TransformError(
f"Cannot drop column {dropped_column!r}: it is used by "
f"CHECK constraint {check.name or check.check!r}"
)
create_table_checks.append(
Check(
rewrite_check_expression(check.check, rename),
name=check.name,
column=rename.get(owner) or owner,
)
)
create_table_foreign_keys: list[ForeignKeyIndicator] = []
if foreign_keys is not None:
@ -2826,6 +2888,7 @@ class Table(Queryable):
foreign_keys=create_table_foreign_keys,
column_order=column_order,
strict=self.strict if strict is None else strict,
_checks=create_table_checks,
).strip()
)

View file

@ -0,0 +1,154 @@
import pytest
from sqlite_utils import Database
from sqlite_utils.utils import sqlite3
BASELINE_ROWS = [(1, "one"), (2, "two")]
def insert(table):
table.insert({"id": 3, "value": "three"}, pk="id")
def insert_all(table):
table.insert_all(
[
{"id": 3, "value": "three"},
{"id": 4, "value": "four"},
],
pk="id",
batch_size=1,
)
def upsert(table):
table.upsert({"id": 2, "value": "TWO"}, pk="id")
def upsert_all(table):
table.upsert_all(
[
{"id": 2, "value": "TWO"},
{"id": 3, "value": "three"},
],
pk="id",
batch_size=1,
)
def update(table):
table.update(2, {"value": "TWO"})
def delete(table):
table.delete(2)
def delete_where(table):
table.delete_where("id > ?", [1])
MUTATOR_CASES = (
pytest.param(
insert,
[(1, "one"), (2, "two"), (3, "three")],
id="insert",
),
pytest.param(
insert_all,
[(1, "one"), (2, "two"), (3, "three"), (4, "four")],
id="insert_all",
),
pytest.param(
upsert,
[(1, "one"), (2, "TWO")],
id="upsert",
),
pytest.param(
upsert_all,
[(1, "one"), (2, "TWO"), (3, "three")],
id="upsert_all",
),
pytest.param(
update,
[(1, "one"), (2, "TWO")],
id="update",
),
pytest.param(delete, [(1, "one")], id="delete"),
pytest.param(delete_where, [(1, "one")], id="delete_where"),
)
class RollbackTest(Exception):
pass
def seed_database(path):
conn = sqlite3.connect(str(path))
try:
conn.execute("create table items (id integer primary key, value text)")
conn.executemany("insert into items values (?, ?)", BASELINE_ROWS)
conn.commit()
finally:
conn.close()
return Database(path)
def current_rows(db):
return db.conn.execute("select id, value from items order by id").fetchall()
def persisted_rows(path):
conn = sqlite3.connect(str(path))
try:
return conn.execute("select id, value from items order by id").fetchall()
finally:
conn.close()
@pytest.mark.parametrize("mutate,expected_rows", MUTATOR_CASES)
def test_mutator_commits_by_default(tmp_path, mutate, expected_rows):
path = tmp_path / "default.db"
db = seed_database(path)
assert not db.conn.in_transaction
mutate(db["items"])
assert current_rows(db) == expected_rows
assert not db.conn.in_transaction
db.close()
assert persisted_rows(path) == expected_rows
@pytest.mark.parametrize("mutate,expected_rows", MUTATOR_CASES)
def test_mutator_commits_with_outer_atomic(tmp_path, mutate, expected_rows):
path = tmp_path / "atomic.db"
db = seed_database(path)
with db.atomic():
assert db.conn.in_transaction
mutate(db["items"])
assert current_rows(db) == expected_rows
assert db.conn.in_transaction
assert current_rows(db) == expected_rows
assert not db.conn.in_transaction
db.close()
assert persisted_rows(path) == expected_rows
@pytest.mark.parametrize("mutate,expected_rows", MUTATOR_CASES)
def test_mutator_rolls_back_outer_atomic(tmp_path, mutate, expected_rows):
path = tmp_path / "rollback.db"
db = seed_database(path)
with pytest.raises(RollbackTest), db.atomic():
mutate(db["items"])
assert current_rows(db) == expected_rows
assert db.conn.in_transaction
raise RollbackTest
assert current_rows(db) == BASELINE_ROWS
assert not db.conn.in_transaction
db.close()
assert persisted_rows(path) == BASELINE_ROWS

View file

@ -2,7 +2,7 @@ import sqlite3
import pytest
from sqlite_utils.db import ForeignKey, TransactionError, TransformError
from sqlite_utils.db import Check, ForeignKey, TransactionError, TransformError
from sqlite_utils.utils import OperationalError
@ -1065,3 +1065,138 @@ def test_transform_restores_legacy_alter_table_setting(fresh_db):
assert sqls[-1] == "PRAGMA legacy_alter_table=ON;"
dogs.transform(types={"name": str})
assert fresh_db.execute("PRAGMA legacy_alter_table").fetchone()[0] == 1
def test_transform_preserves_check_constraints(fresh_db):
fresh_db.execute("""
CREATE TABLE scores (
id INTEGER PRIMARY KEY,
score INTEGER CONSTRAINT valid_score CHECK(score BETWEEN 0 AND 100),
CONSTRAINT nonzero_id CHECK(id != 0)
)
""")
scores = fresh_db["scores"]
scores.insert({"id": 1, "score": 50})
scores.transform()
assert scores.checks == [
Check("score BETWEEN 0 AND 100", name="valid_score", column="score"),
Check("id != 0", name="nonzero_id"),
]
with pytest.raises(sqlite3.IntegrityError, match="CHECK constraint failed"):
scores.insert({"id": 2, "score": 101})
def test_transform_preserves_check_ending_in_line_comment(fresh_db):
fresh_db.execute("""
CREATE TABLE inventory (
quantity INTEGER,
CHECK (
quantity >= 0 -- Quantity cannot be negative
)
)
""")
inventory = fresh_db["inventory"]
inventory.transform(types={"quantity": float})
assert inventory.checks == [Check("quantity >= 0 -- Quantity cannot be negative")]
with pytest.raises(sqlite3.IntegrityError, match="CHECK constraint failed"):
inventory.insert({"quantity": -1})
def test_transform_renames_columns_inside_check_constraints(fresh_db):
fresh_db.execute("""
CREATE TABLE inventory (
quantity INTEGER CONSTRAINT positive
CHECK(quantity > 0 AND 'quantity' != ''),
maximum INTEGER,
CONSTRAINT within_maximum CHECK(quantity <= maximum)
)
""")
inventory = fresh_db["inventory"]
inventory.insert({"quantity": 2, "maximum": 3})
inventory.transform(rename={"quantity": "amount"})
assert inventory.checks == [
Check(
"amount > 0 AND 'quantity' != ''",
name="positive",
column="amount",
),
Check("amount <= maximum", name="within_maximum"),
]
with pytest.raises(sqlite3.IntegrityError, match="CHECK constraint failed"):
inventory.insert({"amount": 4, "maximum": 3})
def test_transform_check_rewrite_preserves_functions_and_quotes(fresh_db):
fresh_db.execute("""
CREATE TABLE items (
length TEXT,
"old name" TEXT,
CHECK(length("old name") > 0 AND length != '')
)
""")
items = fresh_db["items"]
items.insert({"length": "label", "old name": "hello"})
items.transform(rename={"length": "description", "old name": "new name"})
assert items.checks == [Check("length(\"new name\") > 0 AND description != ''")]
def test_transform_check_rewrite_quotes_keyword_column(fresh_db):
fresh_db.execute("CREATE TABLE t(old_name TEXT CHECK(old_name != ''))")
fresh_db["t"].insert({"old_name": "value"})
fresh_db["t"].transform(rename={"old_name": "select"})
assert fresh_db["t"].checks == [Check("\"select\" != ''", column="select")]
def test_transform_check_rewrite_does_not_rename_collations_or_cast_types(fresh_db):
fresh_db.execute("""
CREATE TABLE t (
nocase TEXT,
kind TEXT,
other TEXT,
CHECK(
other COLLATE nocase != ''
AND CAST(other AS kind) != ''
AND nocase != ''
AND kind != ''
)
)
""")
fresh_db["t"].insert({"nocase": "n", "kind": "k", "other": "o"})
fresh_db["t"].transform(rename={"nocase": "label", "kind": "category"})
check = fresh_db["t"].checks[0].check
assert "COLLATE nocase" in check
assert "AS kind" in check
assert "AND label != ''" in check
assert "AND category != ''" in check
def test_transform_drops_check_owned_by_dropped_column(fresh_db):
fresh_db.execute("""
CREATE TABLE t (
id INTEGER,
obsolete INTEGER CHECK(obsolete > 0),
CHECK(id > 0)
)
""")
fresh_db["t"].insert({"id": 1, "obsolete": 2})
fresh_db["t"].transform(drop={"obsolete"})
assert fresh_db["t"].checks == [Check("id > 0")]
def test_transform_refuses_to_drop_column_used_by_remaining_check(fresh_db):
fresh_db.execute("""
CREATE TABLE ranges (
minimum INTEGER,
maximum INTEGER,
CHECK(minimum <= maximum)
)
""")
ranges = fresh_db["ranges"]
ranges.insert({"minimum": 1, "maximum": 2})
schema_before = ranges.schema
with pytest.raises(
TransformError,
match="Cannot drop column 'maximum'.*CHECK constraint",
):
ranges.transform(drop={"maximum"})
assert ranges.schema == schema_before