* Use a non-UTC timezone when testing the PostgreSQL timezone support.
* Fix serialization of time of day values for PostgreSQL
* Fix serialization of PostgreSQL values for "timestamp with time zone" and "time with time zone" fields.
Add support for connection pooling
* Add support for connection pooling to the core code.
* Add support for PostgreSQL connection pooling with tests.
* Add support for SQLite3 connection pooling with tests.
* Add support for MySQL connection pooling with tests.
- The postgres connection does not change the connection_config thus passing it as const.
- Implies 'thread safety' when using the same config for multiple connections
Co-authored-by: Carel Combrink <carel.combrink@vastech.co.za>
* When inserting values into "timestamp with time zone" fields treat the value as being in the UTC time zone.
* Simplify parsing of PostgreSQL date/time responses by using regular expressions. Always convert response times with time zone to UTC.
* Add tests which check if timestamp and date fields are treated as having a UTC time zone.
* Clarify the test comments.
* value: add missing includes
* mysql, sqlite3: use explicitly *out* stringstreams for serialization
* postgresql: fix: use max_digits10 instead of digits10
* detail: add float_safe_ostringstream, ensuring floats are serialized with enough precision
* mysql, postgresql, sqlite3, tests: fix: set float-precision for all streams
* run clang-format on changed files
* inline remove_cvref.h
* replace unspecific "wrapper" with concrete type name
* tests: split connector specific tests out of serialize/Float
Leak: The sqlite3 connection handle destructor was
accidentally turned into a default constructor during the
migration of repositories.
Replaced the sqlite3 raw pointer with a unique_ptr to prevent
this kind of accident.
Removed timezone handling in the process (needs to be documented).
Note: on_conflict does not check for constraints (needs to be documented).
Note: some of the constraints tests seem to be aiming for a different static_assert.
Changed sqlite3 usage tests to be linked into one executable.
This provokes multiple definition errors for non-inlined free
functions.
Also ran clang-format on the changed headers.
The compiler actually does a much better job at telling you if two
result columns have the same name as soon as you are trying to use
them.
In contrast to the static_assert, the compiler will even say which
name is ambiguous.
