Composite Type Support

[Hoverbear]

Addresses #499, #203, #326. Helps along #451.

Adds "Composite Type" support to PGX.

What's changed

• A new pgx::composite_type!("name") macro now exists. It can be used in most PGX related macros where one would use an argument. It expands to create a PgHeapTuple, but the PGX related macros are able to derive additional metadata from this macro for SQL generation.
• PgHeapTuple now implements FromDatum.
• PgHeapTuple acquired a new_composite_type constructor which allows users to create a heap tuple shaped for a given composite type string. Eg gHeapTuple::new_composite_type("dog").
• PgHeapTuple now has a into_trigger_datum function which will result in the correct Datum for returning from a #[pg_trigger] function. Since this was done as part of macro expanded code, no user changes should be required.
• The IntoDatum implementation of PgHeapTuple changed to better work with composite types. Specialized trigger related functions exist and should work transparently for users.
• PgTupleDesc now has a for_composite_type function which allows users to construct one for a given composite type by name.

Breaking Changes

• You can no longer do Option<default!(i32, 1)> in your #[pg_externs]. Change it to default!(Option<i32>, 1).
• The ordering of pgx related function argument macros is now fixed. Build errors should correctly guide users into the correct ordering with location context.
• Various types inside pgx::utils::sql_entity_graph were updated to include fields for composite types. MaybeNamedVariadicTypeList was removed, a new UsedType was added, and various internals around macro-expansion-time type resolution were tinkered with.

  As a reminder, while these structures are public, they are not recommended for users unless they are ok with them breaking nearly every 0.x version.

• If you are not using pgx::pg_module_magic!(); (or pgx::pg_sql_graph_magic!();) in your extension, they now expand to expose a pub extern "C" fn __pgx_sql_mappings() -> ::pgx::utils::sql_entity_graph::RustToSqlMapping instead of their prior. This was done to both reduce the number of exported symbols extensions you, as well as support composite types.

[Hoverbear]

@eeeebbbbrrrr I'd appreciate you giving this a lookie-loo if you can. :)

[Hoverbear]

Whoops...

[Hoverbear]

I should probably do an aggregate test...

[eeeebbbbrrrr]

comment no longer matches the code below. Why are we now using 'static?

[Hoverbear]

Ah, thanks!

We need anything using a lifetime to use the same lifetime so that the TypeIds match (&'a Foo and &'b Foo have different TypeIds)

[eeeebbbbrrrr]

Why did this go away? This is for supporting postgres ARRAY[]s that might contain NULL elements

[Hoverbear]

There is now a conflicting implementation which covers the same type and works the same:

error[E0119]: conflicting implementations of trait `datum::from::FromDatum` for type `std::vec::Vec<std::option::Option<_>>`
   --> pgx/src/datum/array.rs:334:1
    |
317 | impl<T: FromDatum> FromDatum for Vec<T> {
    | --------------------------------------- first implementation here
...
334 | impl<T: FromDatum> FromDatum for Vec<Option<T>> {
    | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ conflicting implementation for `std::vec::Vec<std::option::Option<_>>`

For more information about this error, try `rustc --explain E0119`.
error: could not compile `pgx` due to previous error
warning: build failed, waiting for other jobs to finish...
error: could not compile `pgx` due to previous error

[Hoverbear]

If I re-add this and remove that conflicting implementation I get a lot of type failures, eg

   Compiling pgx v0.5.0-beta.0 (/home/ana/git/tcdi/pgx/pgx)
error[E0277]: the trait bound `std::option::Option<heap_tuple::PgHeapTuple<'static, pgbox::AllocatedByRust>>: datum::PostgresType` is not satisfied
   --> pgx/src/lib.rs:263:26
    |
263 |             TypeId::of::<Array<Option<PgHeapTuple<'static, AllocatedByRust>>>>(),
    |                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ the trait `datum::PostgresType` is not implemented for `std::option::Option<heap_tuple::PgHeapTuple<'static, pgbox::AllocatedByRust>>`
    |
note: required because of the requirements on the impl of `from::FromDatum` for `std::option::Option<heap_tuple::PgHeapTuple<'static, pgbox::AllocatedByRust>>`
   --> pgx/src/datum/varlena.rs:344:14
    |
344 | impl<'de, T> FromDatum for T
    |              ^^^^^^^^^     ^
note: required by a bound in `datum::array::Array`
   --> pgx/src/datum/array.rs:16:25
    |
16  | pub struct Array<'a, T: FromDatum> {
    |                         ^^^^^^^^^ required by this bound in `datum::array::Array`

[Hoverbear]

@zombodb I added many changes around aggregate support, did you want to take another review pass or shall I merge once I'm comfortable with it? I need to do a self review still.