Hi Michanel, thanks for great document! i'll put it into alsa-lib code for doxygen.
At Wed, 5 Dec 2001 01:23:15 -0500, Michael Ashton wrote: > > > Wrote a bit of "documentation" based on my reading of the sequencer > code. It's an attempt to answer the question "how does one send and > receive sys-ex using the sequencer?" Read and use at will, but don't > trust it until Takashi, Frank, and / or Jaroslav give their approval.. > > =================================================================== > > Sysex and the ALSA sequencer > ---------------------------- > > System exclusive messages are unique in the MIDI protocol; they are the > only MIDI messages which can vary in length, and they are the only > ones which have a start and a stop byte. Furthermore, MIDI places no > limit on the length of a sys-ex message. > > Because of this, sys-ex messages cannot be encoded directly into ALSA > sequencer events, which are fixed in length. Instead, the sequencer > uses the special _external_ event type, so named because its data > payload is kept outside the event; the event itself contains only a > pointer to the data and the length of the data. > > * Reception > > Sys-ex messages are not always packed into one sequencer > event. As of this writing, the sequencer's MIDI input system has a > buffer 256 bytes in length, which is, of course, not long enough to > hold all sys-ex messages. Large sys-ex messages are broken into chunks: > each time the buffer becomes full, the sequencer sends another sys-ex > event, which contains the most recently received bytes of the sys-ex > message. It is the client application's task to reassemble these into > a complete message, if necessary. > > This is actually a more favourable situation for the client, as it > allows the client to monitor the progress of sys-ex transfers, and to > process large messages without storing them, where possible. > > Sys-ex events are transmitted along with the beginning and > ending bytes 0xF0 and 0xF7. This is the only way for a client to tell > the beginning, middle, and ending parts of a series of sys-ex events > apart. > > Note that interleaved sys-ex events are not an issue, since it isn't > possible for sys-ex messages to be interleaved over a single MIDI > port. > > * Transmission > > Currently, ALSA does not require that sys-ex messages be sent in > pieces, but it does allow this. Sys-ex messages are not parsed for > output in any way; they are simply dumped to the destination port as > given. (It is therefore possible to send _any_ MIDI bytestream this > way!) Messages may be sent as a single event or multiple events, > regardless of length, which is limited only by available memory. > > To transmit events, the client calls snd_seq_event_output_buffer() or > snd_seq_event_output_direct(), both of which are library functions. > Both functions take as argument a sequencer handle and a pointer to > the event struct to be transmitted. > > snd_seq_event_output_buffer() places the event in an event buffer; the > event waits there, possibly with others, until snd_seq_drain_output() > is called or until the buffer becomes full, at which time > snd_seq_drain_output() is called automatically. If a variable-length > event is sent using this function, it is copied in its entirety onto > the buffer, along with the accompanying event record. If the event is > too large for the buffer, ALSA simply refuses to send it. > > The other method, snd_seq_event_output_direct(), writes the event > directly to the kernel driver; the event buffer is not used (or > affected). Variable-length events are treated slightly differently: a > block of memory is allocated to contain their payload, and this is > then written, preceded by the event record, to the kernel. > > Curiously, the kernel driver appears to have the ability to copy data > directly from user-space, but the library currently does not seem to > make use of this. Well, this is because SYSEX belongs to a VARIABLE-length event which is slightly different from VARUSR event (mainly for instrument layers). The former requires its data after the event header linearly while the latter passes only the pointer to the kernel. And, yes, alsa-lib could be more wise: When sysex is passed through snd_seq_output_event_direct(), this could behave as VARUSR to avoid overhead.. Takashi _______________________________________________ Alsa-devel mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/alsa-devel
