Permalink
Commits on Nov 8, 2018
Commits on Aug 31, 2018
  1. mpeg parse update.

    karlheyes committed Aug 31, 2018
    handle very large frames (typicall ID3 tags) without dropping to failed parsing.
    drop the restricted block size adjustment, use the incoming bitrate as a scaling
    factor. This allows for huge bitrates while keeping smaller blocks for typical
    streams.
Commits on May 27, 2018
  1. push kh10

    karlheyes committed May 27, 2018
Commits on Apr 26, 2018
  1. drop flag setting for indicating the presence of in-memory content.

    karlheyes committed Apr 26, 2018
    Needs to be renamed really, but this should only be used for content that has been
    prepared and treated like intro data but is not actually from a file.
Commits on Apr 25, 2018
Commits on Apr 24, 2018
Commits on Apr 2, 2018
  1. filter out ID3 when streaming for now.

    karlheyes committed Apr 2, 2018
    seems like browsers have issues with ID3 embedded in content when not at the beginning. This
    poses a problem when a fallback file is used with ID3.
    
    Some slightly processing change in mpeg handling. To prevent a bunch of resync messages
    appearing, a failed frame process like for ID3 frames can report the frame size which can
    then be skipped or assumed to be 1 byte as before.
Commits on Apr 1, 2018
  1. the rate avg shorten was incorrectly increasing the range by 1.

    karlheyes committed Apr 1, 2018
    The increase generally meant a slightly inaccurate rate used for fallback files
Commits on Mar 31, 2018
  1. fix memory leak case, the queue was not being purged properly from re…

    karlheyes committed Mar 31, 2018
    …cent commits.
    
    A recent commit to lessen the frequency of purges tripped over the check for the end time
    check. This caused the purge to fail and the queue to expand
Commits on Mar 26, 2018
  1. make incoming rate to update far more frequently, but report the stat…

    karlheyes committed Mar 26, 2018
    … at the same rate
    
    A number of internal checks are based on the current assessment of the incoming bitrate, which may
    not be good enough if it is updated only every few seconds.  So update the internal value frequently,
    eg every few read attempts, but report the stat for it at the same rate.
Commits on Mar 18, 2018
  1. minor typos

    karlheyes committed Mar 18, 2018
  2. windows build script update

    karlheyes committed Mar 18, 2018
  3. cut a kh9 release

    karlheyes committed Mar 18, 2018
Commits on Mar 16, 2018
  1. fix stupid bug, causing FD closure at source exit, eg listener socket…

    karlheyes committed Mar 16, 2018
    …s or sigfd etc
    
    Wrong api call with log handle, looks like this was causing a problem for socket closures
    on source shutdowns, workers, signalfd and listener sockets.  Would only occur if the preroll
    setting was in use though.
Commits on Mar 13, 2018
  1. fix some errno related possibities. not ususally a problem.

    karlheyes committed Mar 13, 2018
    While glibc is ok with this, we cannot assume certain behaviour across the board, eg strtol can set errno but
    return a valid figure and errno may never be zerod.
Commits on Mar 11, 2018
Commits on Mar 10, 2018
  1. Merge commit 'c7e2c01'

    karlheyes committed Mar 10, 2018
  2. trigger a preroll log cycle reopen on the apply routine.

    karlheyes committed Mar 10, 2018
    The id is attached to the source_t not the mount in these cases
  3. average rate handler update.

    karlheyes committed Mar 10, 2018
    make source push a 0 value on each run through to allow for a regular drop of the bitrate if
    nothing keeps it up (stalled listeners were not handled previoudly). Ideally we should make it
    more time dependant eg every 200ms instead of every read attempt.
    
    change in the way the next block is reused in the circular list. We still have to allow for
    inserts as the purge can delete blocks normally.
    
    On a reduce trigger, set a cycle off which initially reduces the sampling to the last N samples
    and then keep that number of samples at least for 1 go.  This means that after the reduce, any
    residual high samples can be dropped out of the average quickly.
Commits on Mar 8, 2018
  1. make sure the error flag is treated sanely in the 404 path

    karlheyes committed Mar 8, 2018
    we don't want to send a response on a dead socket.
Commits on Mar 7, 2018
  1. intro handling update. reset timer for a lagging listener

    karlheyes committed Mar 7, 2018
    if a listener is kicked back to intro with a partial block then reset timer so the drop case
    is handled better. Increase the low threshold for such cases to half bitrate. Also do not
    trigger a intro log entry for that case.
Commits on Mar 6, 2018
  1. fix possible double close FD on older setups (ipv4 api only).

    karlheyes committed Mar 6, 2018
    This should on be on pre IPv6 setups and only with relays involved.
Commits on Mar 4, 2018
  1. add ssl-capath option is required.

    karlheyes committed Mar 4, 2018
    usually the trusted CA pem is located by the openssl library in a system wide place but this
    allows for a localized version if required.
Commits on Mar 3, 2018
  1. cleanup some log messaging

    karlheyes committed Mar 3, 2018
  2. worker balancing adjustments

    karlheyes committed Mar 3, 2018
    mostly to prevent source clients bouncing around too much. allow listeners to move around a
    bit more.