FR: Support for VARLEN fields with different size

[GitMensch]

The size of sub-items to store the length of variable-length-fields can be currently chosen at compile-time like this:

gixsql/libgixpp/TPESQLProcessing.cpp

Lines 107 to 120 in f28e208

	// These must be in sync with the ones in SqlVar.h
	#ifdef USE_VARLEN_32
	#define VARLEN_LENGTH_PIC "9(8) COMP-5"
	#define VARLEN_PIC_SZ 9
	#define VARLEN_LENGTH_SZ 4
	#define VARLEN_LENGTH_T uint32_t
	#define VARLEN_BSWAP COB_BSWAP_32
	#else
	#define VARLEN_LENGTH_PIC "9(4) COMP-5"
	#define VARLEN_PIC_SZ 4
	#define VARLEN_LENGTH_SZ 2
	#define VARLEN_LENGTH_T uint16_t
	#define VARLEN_BSWAP COB_BSWAP_16
	#endif

The problem is that short/long field length indicators cannot be currently mixed, so in practice there is no difference between VARCHAR and LONG VARCHAR, provided that you define USE_VARLEN_32 (except at declaration time, where I think LONG VARCHAR is still not supported, so you have to use standard VARCHAR).

Originally posted by @mridoni in #38 (comment)

So things to do:

• support both 2 and 4 byte sizes in the runtime together, likely adding an internal type
• adjust the preprocessor to insert the matching type when explicit specified via EXEX SQL VAR and to use the 4byte type for VARYING groups which extend the 2byte size

[GitMensch]

Rechecked: OCESQL uses the following:

#define OCDB_VARCHAR_HEADER_BYTE 4 // size of PIC S9(8) COMP-5`

So I do wonder: Shouldn't USE_VARLEN_32 be the default as long as there's a fixed varlen only?

[mridoni]

So I do wonder: Shouldn't USE_VARLEN_32 be the default as long as there's a fixed varlen only?

Yes, that's probably for the better, I'll add this to the next batch of fixes.

[GitMensch]

Should the "variable" support instead of GixSQL compile time be in the backlog or planned for a version?
For now I have no need for it, just want to get some overview where GixSQL stands and what issues are "in progress".

[mridoni]

This will be partially solved in v1.0.21dev (to be released in about a week). It will be possible to specify the size of the length indicator both during preprocessing and at runtime (with an environment variable). It is a stopgap but at least it allows for different indicator field sizes without having to recompile (or use a separate version of) GixSQL. This will be definitely fixed by adding a flag to the ones currently passed to define SQL parameters (along with type, size, etc.) but I don't know if it will be possible for v1.0.21 "final".