> 1. invention RAW_TEXT and RAW_BINARY > 2. for RAW_BINARY: PQbinaryTuples() returns 1 and PQfformat() returns 1 > 3.a for RAW_TEXT: PQbinaryTuples() returns 0 and PQfformat() returns 0, but > the client should to check PQcopyFormat() to not print "\n" on the end > 3.b for RAW_TEXT: PQbinaryTuples() returns 1 and PQfformat() returns 1, but > used output function, not necessary client modification > 4. PQcopyFormat() returns 0 for text, 1 for binary, 2 for RAW_TEXT, 3 for > RAW_BINARY > 5. create tests for ecpg
3.b certainly seems completely wrong. PQfformat==1 would imply binary data.
I suggest that PQcopyFormat should be understood as defining the format of the copy data encapsulation, not the individual fields. So it would go like 0 = traditional text format, 1 = traditional binary format, 2 = raw (no encapsulation). You'd need to also look at PQfformat to distinguish raw text from raw binary. But if we do it as you suggest above, we've locked ourselves into only ever having two field format codes, which is something the existing design is specifically intended to allow expansion in.
I wrote concept of raw_text, raw_binary modes.
I am trying to implement text data passing like text format - but for RAW_TEXT it is not practical. Text passing is designed for one line data, for multiline data enforces escaping, what we don't would for RAW mode. I have to skip escaping, and the code is not nice.
So I propose different schema - RAW_TEXT uses text values (uses input/output functions), enforce encoding from/to client codes and for passing to client mode is used binary mode - then I don't need to read the content with line by line. PQbinaryTuples() returns 1 for RAW_TEXT and RAW_BINARY - in these cases data are passed as one binary value. PQfformat returns 2 for RAW_TEXT and 3 for RAW_BINARY.