I (AGL) specifically participated in this effort to provide a
freely available Pascal source of Hydra, aiding further integration
of the protocol in programs.
The stuff works, and was used in the Communique 2.10 release which
supports Hydra internally.

So, here we are....


    Couple of things you have to know when including Hydra in your own
    program:

    The buffers are NEVER flushed or cleared, not even if errors occur.
    Only for an abort, just before the cancel sequence is sent.

    New data packets are only put in the transmit buffer if there are
    less bytes in that buffer than the maximum block length. This will
    keep the transmit buffer always full enough, but leave the Hydra
    code free to process incoming data.

    To enable maximum throughput without losing characters or slowing
    down transmission, FORCE your program to use async buffers with a
    minimum size of 4096 bytes each.
    Of course you can't directly set the size of those buffers if you
    use a FOSSIL driver. That's why The-Box Mailer and HydraCom call
    the FOSSIL information function, and print a warning message if
    either of the buffers is less than 4095 (yes 4095, the OS/2 VX00
    reports 4095 instead of 4096 ;-)
    Please don't argue, just do it. Alll this has been tried & tested.
    Larger buffers are  okidoki, but generally will not improve
    performance of either transfer direction....

    Depending on the system, you may need to lower RTS during disk
    access to prevent losing incoming data. DOS keeps interrupts
    disabled rather (too) long....
    X00 and VX00 have extended FOSSIL functions supporting this.
    If you use your own async stuff, make sure they can do it too.

    A general note, important not just for Hydra:
    Add timers to your lowlevel byte/block transmit functions, and
    the flush output buffer routine.
    Some (mail)sessions have been observed to just die, while the
    program stays online. Carrier is still present, and braindead
    timers in the highlevel code don't go off because the program
    is stuck in a lower level, waiting for some bytes to be passed
    to the FOSSIL or async code.
    This can happen because of hardware handshake problems, modem
    retraining, whatever. In any case, it happens....
    So it's in that lowlevel code that you need to add your timeout;
    something in the order of a minute will do fine. No need to pass
    back an error code, your high level timers should take care of
    the rest.

    Refer to HYDRA.DOC (HYDRA001.ZIP and FSC-0072.A01) for the original
    Hydra specs....

    Do read LICENSE.DOC before you grab the sources, and don't forget
    to mention what needs to be mentioned in your docs and (C) screens.

    Well, have fun, and integrate Hydra into your software soon!
    Please don't wait for someone else to do it first?
    This stuff was sent directly to developers of mailer, BBS and
    terminal software, those who were already aware of this source
    becoming available and in fact eagerly waiting for it.
    Won't tell you who they are to give them a fair chance of working
    on their programs without being harrassed about when they will
    release something with Hydra, but.... they are some biggies ;-)
    Better you hurry up too!

    Arjen Lentz
    FidoNet 2:283/512
    BBS/FAX +31-33-633916
    arjen_lentz@f512.n283.z2.fidonet.org