PG 1.5.0 may have broken COPY exports

[askreet]

The following example fails with:

.../lib/ruby/gems/3.2.0/gems/pg-1.5.1/lib/pg/connection.rb:406:in `sync_get_copy_data': no COPY in progress (PG::Error)
	from .../lib/ruby/gems/3.2.0/gems/pg-1.5.1/lib/pg/connection.rb:406:in `get_copy_data'
	from .../lib/ruby/gems/3.2.0/gems/pg-1.5.1/lib/pg/connection.rb:249:in `copy_data'
	from test.rb:8:in `<main>'

require 'pg'

conn = PG::Connection.new(host: 'localhost', user: 'postgres', password: 'password', port: 5432, dbname: 'dbname')

conn.copy_data('COPY (SELECT * FROM users) TO STDOUT WITH BINARY') do |res|
  loop do
    data = conn.get_copy_data
    break if data.nil?
  end
end

This works in 1.4.6. It seems that we call get_copy_data an additional time as of v1.5.0 to handle some expected padding from PG, but that does not happen on my system.

 PostgreSQL 14.6 (Debian 14.6-1.pgdg110+1) on aarch64-unknown-linux-gnu, compiled by gcc (Debian 10.2.1-6) 10.2.1 20210110, 64-bit

[larskanis]

I added automatic handling of header and trailer data to binary COPY format, which is necessary for PG::BinaryEncoder::CopyRow and PG::BinaryDecoder::CopyRow. I think the best fix is to apply this only when a coder is given to copy_data. Then binary encoding and decoding can be done equally simple as in text format through the new coders and compatibility is preserved to use unmodified binary format with no en/decoder. This is implemented in #527 .

[askreet]

Thank you for the rapid fix!