Serial Design Issue: "Serial::isPending" Semantics

[Conrad T. Pino]

Hi David,
=======================================================================
The "Serial::isPending( pendingInput )" implementation looks clean.

A "true" return means data is available for reading on all platforms.
=======================================================================
The "Serial::isPending( pendingOutput )" implementation is problematic.

A "true" return on WIN32 platform means either:

        data was written but the write is NOT complete: 0 != cs.cbOutQue

        transmit buffer is empty

A "true" return on UNIX platform means either:

        device file handle is ready for writing

The second WIN32 case and UNIX case are sufficiently consistent to be
useful to thread that needs to know when a write can happen.

The first WIN32 case is inconsistent with other two cases and isn't
useful to a thread that wants to keep data moving whenever possible.
I recommend removing this case altogether.
=======================================================================
The "Serial::isPending( pendingError )" implementation is inconsistent.

A "true" return on UNIX platform means an exceptional conditions or a
disconnect has occurred.

The WIN32 platform *never* returns true.  Worse errors on WIN32 are
often cleared without being recorded which effectively ignores them.
Clearly WIN32 not supported well in this class family.  I believe the
documentation should state the limitations better.
=======================================================================
Best regards,

Conrad Pino