The sqlite3_value_*() API is almost identical to the sqlite3_column_*() API. sqlite3_value_bytes() should be called after we've converted the value using sqlite3_value_blob().

See also bpo-43249.

Related:
If sqlite3_value_blob() returns NULL, we should check if sqlite3_errcode() equals SQLITE_NOMEM and raise MemoryError if it does.

If not, we should initialise cur_py_value to None, because as the PyBytes_FromStringAndSize docs says: "If v is NULL, the contents of the bytes object are uninitialized."

> If sqlite3_value_blob() returns NULL, we should check if sqlite3_errcode() equals SQLITE_NOMEM and raise MemoryError if it does.

This also applies to sqlite3_value_text().

It also applies to sqlite3_value_bytes() if a conversion takes place. We don't need to care about that as long as we call sqlite3_value_bytes() after conversion.

Also, PyTuple_SetItem() errors are not checked.

I would also suggest using PyUnicode_FromStringAndSize() iso. PyUnicode_FromString() in case SQLITE_TEXT, to avoid the unneeded strlen(). After sqlite3_value_text() is called, we can just use sqlite3_value_bytes(), since the length is precomputed as a result of the conversion.

Berker, would you allow a PR to improve all these issues in _pysqlite_build_py_params(), or would you prefer them split up in multiple PR's?