This is tricky due to driver.Valuer returning any. For example, we can
plan for fmt.Stringer because it always returns a string.
Because of this driver.Valuer was always handled as the last option. But
with pgx v5 now having the ability to find underlying types like a
string and supporting fmt.Stringer it meant that driver.Valuer was
often not getting called because something else was found first.
This change tries driver.Valuer immediately after the initial PlanScan
for the Codec. So a type that directly implements a pgx interface should
be used, but driver.Valuer will be prefered before all the attempts to
handle renamed types, pointer deferencing, etc.
fixes https://github.com/jackc/pgx/issues/1319
fixes https://github.com/jackc/pgx/issues/1311
e.g. 2000-01-01 instead of 2000-1-1. PostgreSQL accepted it without
zeroes but our text decoder didn't. This caused a problem when we needed
to take a value and encode to text so something else could parse it as
if it had come from the PostgreSQL server in text format. e.g.
database/sql compatibility.
The preferred format may not be possible for certain arguments. For
example, the preferred format for numeric is binary. But if
shopspring/decimal is being used without jackc/pgx-shopspring-decimal
then it will use the database/sql/driver.Valuer interface. This will
return a string. That string should be sent in the text format.
A similar case occurs when encoding a []string into a non-text
PostgreSQL array such as uuid[].
Reverted almost all of 976b1e0.
Still may consider a way to get DecodeValue to be strongly typed but
that feature isn't worth the complications of generics. Especially in
that applying this style to ArrayCodec would make Conn.LoadType
impossible for arrays.
Added ValueRoundTripTest to pgxtest
Removed pgtype/testutil
pgtype tests now run with all (applicable) query modes. This gives
better coverage than before and revealed several bugs which are also
fixed in this commit.
