Please make the initial allocation size in ReadVarColumn configurable

[defransen]

... as the comment suggests:

pyodbc/src/getdata.cpp

Line 95 in a4b0b75

// TODO: Make the initial allocation size configurable?

Ideally a configured value of 0 (or -1) would mean: use columnSize * cbElement.

This would help to work around a bug in the Oracle BI ODBC drivers SQLGetData implementation.
The driver expects to get cbAllocated instead of cbAvailable as the 5th parameter in the SQLGetData call. And even worse: it continues to return SQL_SUCCESS_WITH_INFO with a single NULL-byte when it has no more data until the 5th parameter is as big as cbData.
The result is that pyodbc seg-faults after all RAM has been allocated for the buffer.

The only sane way to work around this, without breaking compatibility, is to ensure to never get a SQL_SUCCESS_WITH_INFO return. Which means make the initial buffer as big as the maximum column size.

[mkleehammer]

I don't have an Oracle install to test against. Would a flag that caused pyodbc to pass cbAllocated actually fix the issue? That is, would it return SQL_SUCCESS if it thought it had the right amount of buffer?

Also, is there a bug report open somewhere on the driver we link to and follow?