search

mkleehammer / pyodbc

Python ODBC bridge
https://github.com/mkleehammer/pyodbc/wiki
MIT No Attribution
2.89k stars 562 forks source link

Please make the initial allocation size in ReadVarColumn configurable #1071

Open defransen opened 2 years ago

defransen commented 2 years ago

... as the comment suggests: https://github.com/mkleehammer/pyodbc/blob/a4b0b75dc88c910ed69561038bc21fe0ce7da00e/src/getdata.cpp#L95

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 commented 1 year ago

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?