Original issue https://github.com/jackc/pgtype/issues/68
This crash occurred in the recursive assignment system used to support
multidimensional arrays.
This was fixed in 9639a69d45. However,
that fix incorrectly used nil instead of an empty slice.
In hindsight, it appears the fundamental error is that an assignment to
a slice of a type that is not specified is handled with the recursive /
reflection path. Or another way of looking at it is as an unexpected
feature where []T can now be scanned if individual elements are
assignable to T even if []T is not specifically handled.
But this new reflection / recursive path did not handle empty arrays.
This fix handles the reflection path for an empty slice by allocating an
empty slice.
Adds array support - previously only slices were supported.
Adds new test cases for multidimensional arrays and slices.
All previous test cases are unmodified and passed (fully backwards compatible).
Removes hard-coded type conversions for arrays, instead now relies on the type support of the array element's type conversion support.
Less maintenance for arrays, new type conversions are automatically supported when array's element gains new type support.
Simplifies typed_array_gen.sh generator script by removing the hard-coded single-dimensional types for arrays.
Only typed_array.go.erb and typed_array_gen.sh have been changed + 1 new auxiliary function in array.go file + additional tests in test files for each array. Other changes are from generated code.
It was a mistake to use it in other contexts. This made interop
difficult between pacakges that depended on pgtype such as pgx and
packages that did not like pgconn and pgproto3. In particular this was
awkward for prepared statements.
Because pgx depends on pgtype and the tests for pgtype depend on pgx
this change will require a couple back and forth commits to get the
go.mod dependecies correct.
Also remove PrepareEx. It's primary usage was for context. Supplying
parameter OIDs is unnecessary when you can type cast in the query SQL.
If it does become necessary or desirable to add options back it can be
added in a backwards compatible way by adding a varargs as last
argument.
Instead of needing to instrospect the database on connection preload the
standard OID / type map. Types from extensions (like hstore) and custom
types can be registered by the application developer. Otherwise, they
will be treated as strings.
Jack Christensen
Simo Haasanen
ferhat elmas