SqlDataReader.WillHaveEnoughData is pessimistic, can it read ahead?

[Wraith2]

This is a nice-to-have idea bourne out of some performance investigation.

When doing calling into SqlDataReader.ReadAsync the internal function WillHaveEnoughData is called to see if it's possible to read another row from the current packet without needing to do IO, it calculates the maximum size of each field that could be fetched and uses that to work out if there is enough data present. In the case of strings the maximum size may often exceed the real size by a large amount causing preeimistic behaviour and allocation of a TaskCompletionSource where one Is not needed.

If possible it would be nice if WillHaveEnoughData could have a way to discover if enough column header information is present in the buffer already to determine the size of the next row and avoid the IO if possible. This may require the ability to peek forwards into the buffer by multiple tokens without consuming or doing io.

In practice this behaviour causes a ReadAsync with any variable length type to assume large amounts of data are needed causing task machinery overheads on all invocations where it is only needed in a fraction of the cases.

[cheenamalhotra]

Thanks @Wraith2

Thanks for bringing suggestion for improvement, we will consider this sometime in future when we have bandwidth from high priority issues.