[Openchemistry-developers] Avogadro2 RPC?

[Marcus D. Hanwell]

On Sat, May 27, 2017 at 8:24 PM, Patrick Avery <psavery at buffalo.edu> wrote:
> I spent a good amount of time working through the Windows problem today, and
> I think it might actually be a bug with Qt on Windows. I've also figured out
> a work-around.
>
> Here is the problem. It has to do with operator<<(QDataStream &out, const
> QByteArray &ba).
>
> https://code.woboq.org/qt5/qtbase/src/corelib/tools/qbytearray.cpp.html#_ZlsR11QDataStreamRK10QByteArray
>
> https://github.com/OpenChemistry/molequeue/blob/master/molequeue/servercore/localsocketconnection.cpp#L153
>
> It is supposed to work with QLocalSocket by creating a packet where the
> first few bytes are a quint32 (which represents the size) and the rest of
> the bytes are the actual data, and then send this packet. On Linux and Mac
> this seems to work just fine. But on Windows, it creates two packets: the
> first packet is the quint32 size and the second packet is the data. I don't
> think it is supposed to create two packets, because the reading operator on
> the other end of the communication (operator>>(QDataStream &in, QByteArray
> &ba)) assumes it is one packet and fails to read it correctly.

That is weird, we tested this out on Windows, but that was years ago,
and I haven't done much testing recently. This would likely break
MoleQueue there too.
>
> My current work-around is to avoid using operator<<(QDataStream &out, const
> QByteArray &ba) to send that data via two packets by creating the single
> packet myself and sending it manually. It is done like so:
>
>   // the Json file we are wanting to send is the QByteArray, jsonData
>
>   QByteArray byteArray;
>   QDataStream tmpStream(&byteArray, QIODevice::WriteOnly);
>   tmpStream << jsonData;
>   socketDataStream.writeRawData(byteArray, byteArray.size());
>
> This still uses operator<<(QDataStream &out, const QByteArray &ba), but it
> only uses it to write to byteArray (instead of writing it directly to the
> socket). We then write to the socket manually by using writeRawData().
>
> This worked perfectly for me, and the messages are received just as they
> should be. I'll document this in my how-to markdown file I write.
>
> The return messages in the RPC server have the same problem. Perhaps I
> should implement my work-around in them so that Windows clients can receive
> correct messages back (instead of the two-packet problem).

Sound like a plan, is there an open bug report for Qt anywhere? This
seems like something that would hit more people, and it would be great
to resolve it upstream. The workaround mentioned seems reasonable, I
would like to test it, and it sounds like this same approach should
work on macOS and Linux too.