tds_generic_put: Correct BCP 7.x new blob type handling

[ucko]

... i.e., of VARCHAR(MAX) and the like. Split from #555.

[freddy77]

That's... interesting! Tests passes with both this commit and without, so I suppose this path it's not really tested. Worth to write a test, and check TDS documentation!

[ucko]

An analogous patch to 1.4 has been helpful for some of my usage. However, when I locally extended ctlib's long_binary test to incorporate BCP, I found that it succeeded under TDS 7.2+ only without this change. I'll take a closer look and get back to you.

[ucko]

AFAICT, whether this patch is helpful or harmful depends on runtime context TBD; I'll try to identify the relevant condition and adjust it accordingly.

[freddy77]

TDS protocol documentation seems to indicate PLP length is needed. Yes, it would be useful to extend a test to catch this. I vaguely remember this specific code (I wrote it) and it was kind of a bug in a specific server version, or a combination of server and protocol version. Wondering why I didn't write a test at that time.

[freddy77]

Added a test, see 3006113. Looks like it's working with current code, not with this patch applied.

[freddy77]

Could it be that this commit was here to fix an old bug handling empty strings?

[ucko]

Aha, the supporting logic for blk_textxfer in #558 doesn't account for possibly needing to send a final PLP length of zero. I'll fix that instead.

[ucko]

Withdrawn in favor of #558.