…-rs#4337)

* fix(postgres): check if statement caching is enabled before assigning an id

* test(postgres): add regression test for transact-rs#4328

…-checked (transact-rs#4345)

Restructures the "Nullability: Bind Parameters" section of the query!
macro documentation to state upfront that the nullability of bind
parameters is not verified at compile time, why (SQLx does not parse
SQL), and give the concrete example from transact-rs#2642.

The previous phrasing understated the limitation and hid it behind
guidance about WHERE clauses, which is now demoted to a sub-section.

Closes transact-rs#2642

…t-rs#4334)

… terminals (transact-rs#4263)

* sqlx-cli: use cyan instead of white for help text literals

White is invisible on light terminal backgrounds. Cyan is readable on both
light and dark backgrounds, so it's a better default for the literal style.

Fixes transact-rs#4112

* sqlx-cli: drop the literal style override

Default terminal color is readable on both light and dark backgrounds,
which is the original bug. No need to pick a specific replacement color.

Signed-off-by: Charlie Tonneslan <cst0520@gmail.com>

---------

Signed-off-by: Charlie Tonneslan <cst0520@gmail.com>

transact-rs#4205)

* fix: make MySqlQueryResult -> AnyQueryResult conversion more consistent & robust

- replaced custom map_result function with From<T> implementation

* fix: unify PgQueryResult -> AnyQueryResult conversion code path

- replaced custom map_result function with From<T> implementation

* fix: make SqliteQueryResult -> AnyQueryResult conversion respect last_insert_rowid

- replaced custom map_result function with From<T> implementation
- fixed bug causing AnyQueryResult.last_insert_id to always be None for Sqlite backend

* test(sqlite): add any_sets_last_insert_id

- added test
- updated required-features

* test(mysql): add any_sets_last_insert_id

- added test
- updated required-features

* test(postgres): add any_sets_last_insert_id

- added test
- updated required-features

* fix(tests): use different syntax to address compiler errors

* fix(test/postgres): slightly change bind syntax as required for this backend

…o docs (transact-rs#4359)

…ransact-rs#4340)

Datetimes stored as a REAL Julian day number (e.g. via SQLite's
julianday()) were decoded by splitting the UNIX timestamp with
trunc()/fract().abs(). timestamp_opt() always adds the nanoseconds
forward in time, so for values before 1970 this pushed the result up to
~2 seconds off and could move it onto the wrong side of the epoch
(e.g. 1969-12-31 23:59:59.5 decoded as 1970-01-01 00:00:00.5).

Round seconds toward negative infinity with floor() and take the
sub-second remainder relative to that, so nanos is always a valid
forward offset. Post-epoch values are unaffected. Adds unit tests for
the pre- and post-epoch paths.

Some simple clarification for the use of the examples.

* feat(sqlx-core): split connector creation from handshake

* feat(sqlx-mysql, sqlx-postgres): cache the TLS connector

`num_idle` was documented as returning connections "active and idle",
which reads as active plus idle. It only counts idle ones.

`size` was documented as "currently active. This includes idle connections" which is incomplete: `try_increment_size()` bumps the counter before the connection is established, so connections still
being opened are counted too.

…act-rs#4344)

Adds a "Note: TimeZone" section to PgConnectOptions documentation
explaining that SQLx sets the session TimeZone to UTC on connection,
why this is done, and how it differs from psql's default behavior
(which can surprise users of functions like date_trunc, now, etc.).

Closes transact-rs#3226