Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (qtdemux_parse_trak), (plugin_init):
po/POTFILES.in:
Throw an error when the file is encrypted. Move plugin_init stuff
to the end of the file, add stuff for i18n, make debug category
static.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (qtdemux_parse), (qtdemux_parse_trak),
(gst_qtdemux_handle_esds):
Figure out the real audio type in mp4a boxes by parsing the
optional descriptors in the optional esds box. Promote the
default AAC to mp3 when indicated. Fixes#330632.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (qtdemux_parse), (qtdemux_dump_unknown),
(qtdemux_parse_trak), (gst_qtdemux_handle_esds):
Parse version 2 sample descriptions.
Don't #define gst_util_dump_mem(), use something more
specific instead to avoid confusion.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_handle_src_query),
(qtdemux_dump_mvhd):
Don't cause side effects in a debugging function.
Also report duration in push mode since we can.
Original commit message from CVS:
Patch by: j^ <j at bootlab dot org>
* gst/qtdemux/qtdemux.c: (qtdemux_parse_trak),
(qtdemux_video_caps):
Never treat video streams as an audio stream.
Add qtdrw mime type.
Fixes#339041
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (qtdemux_parse_trak):
For VBR audio, don't try to calculate the samples_per_frame.
Fixes#338935.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_prepare_current_sample),
(gst_qtdemux_chain), (gst_qtdemux_add_stream), (qtdemux_dump_stsz),
(qtdemux_dump_stco), (qtdemux_parse_trak):
Don't make rounding errors in timestamp/duration calculations.
Fix timestamps for AMR and IMA4. Fixes (#337436).
Create a dummy segment even when there is no edit list.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_perform_seek),
(gst_qtdemux_do_seek), (gst_qtdemux_loop_state_movie),
(gst_qtdemux_loop):
Use duration as segment stop position if none is
explicitly configured.
Also perform EOS when we run past the segment stop.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_go_back),
(gst_qtdemux_perform_seek), (gst_qtdemux_do_seek),
(gst_qtdemux_loop_state_movie), (gst_qtdemux_loop),
(gst_qtdemux_chain), (qtdemux_parse_tree), (qtdemux_parse_trak):
More cleanups, added comments.
Mark discontinuities on outgoing buffers.
Post better errors when something goes wrong.
Handle EOS and segment end properly.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_init),
(gst_qtdemux_push_event), (gst_qtdemux_go_back),
(gst_qtdemux_perform_seek), (gst_qtdemux_do_seek),
(gst_qtdemux_handle_src_event), (plugin_init),
(gst_qtdemux_change_state), (gst_qtdemux_loop_state_movie),
(gst_qtdemux_loop), (gst_qtdemux_chain),
(qtdemux_sink_activate_pull), (gst_qtdemux_add_stream),
(qtdemux_parse), (qtdemux_parse_tree), (qtdemux_parse_trak),
(qtdemux_parse_udta), (qtdemux_tag_add_str), (qtdemux_tag_add_num),
(qtdemux_tag_add_gnre), (gst_qtdemux_handle_esds):
* gst/qtdemux/qtdemux.h:
Handle stss boxes so we can mark and find keyframes.
Implement correct accurate and keyframe seeking.
Use _DEBUG_OBJECT when possible.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_class_init),
(gst_qtdemux_init), (gst_qtdemux_dispose),
(gst_qtdemux_add_stream), (qtdemux_parse_trak):
Series of memleak fixes:
- Unref the GstAdapter in finalize.
- Use gst_pad_new_from_static_template(), shorter and safer.
- Free unused QtDemuxStream when not used.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (qtdemux_parse_udta):
Extract disc number and count from files that use
'disk' instead of 'disc' as node identifier for that
(fixes#332066).
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (qtdemux_parse), (qtdemux_parse_trak):
Use GST_WARNING instead of GST_ERROR for all the too short/long atoms
when parsing.
Also let's be a bit less vulgar in our warning messages :)
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (qtdemux_parse_trak):
Can't divide through zero (suppress warning in case of
stream with one single still picture) (see #327083)
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_add_stream),
(qtdemux_parse_trak), (qtdemux_video_caps):
Add support for palettised Apple SMC videos (#327075, based on
patch by: Fabrizio Gennari <fabrizio dot ge at tiscali dot it>).
Original commit message from CVS:
Reviewed by : Edward Hervey <edward@fluendo.com>
* gst/qtdemux/qtdemux.c: (qtdemux_video_caps):
Add 'dvsd' and 'dv25' to list of possible fourcc values for DV Video.
Add image/png for fourcc 'png '
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_loop_state_movie):
Don't GST_LOG timestamps from nonexistent index
entries (#331582).
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_loop_state_header):
Check that the size of the returned buffer is of the correct size
because the parser assumes that.
Fixes#331543.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_handle_src_event),
(gst_qtdemux_loop), (qtdemux_sink_activate_pull):
Don't stop the task if the pad isn't linked.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_post_buffering),
(gst_qtdemux_chain):
When buffering MDAT data, show the user something is
happening by posting 'buffering' messages on the bus.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_init),
(gst_qtdemux_handle_src_query), (gst_qtdemux_change_state),
(next_entry_size), (gst_qtdemux_chain):
* gst/qtdemux/qtdemux.h:
Make push-based work if mdat atom is before moov atom.
Don't answer duration query. This should be transformed into replying
FALSE to seek events.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (next_entry_size), (gst_qtdemux_chain):
Handle the case where data atoms are before moov atoms in push-based mode.
Errors out gracefully.
Original commit message from CVS:
* gst/qtdemux/Makefile.am:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_init),
(gst_qtdemux_handle_sink_event), (gst_qtdemux_change_state),
(extract_initial_length_and_fourcc),
(gst_qtdemux_loop_state_header), (gst_qtdemux_loop_state_movie),
(gst_qtdemux_loop_header), (next_entry_size), (gst_qtdemux_chain),
(qtdemux_sink_activate), (qtdemux_sink_activate_pull),
(qtdemux_sink_activate_push), (qtdemux_parse_trak):
* gst/qtdemux/qtdemux.h:
QtDemux can now work push-based.
It still needs some love for seeking.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_loop_header),
(qtdemux_parse_trak):
use the correct variable to check if we can calculate
the last chunk. Looks like an obvious bug, and makes
the dump of offsets comparable to other tools
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_loop_header),
(qtdemux_parse_trak):
clean up some debugging, using _OBJECT, moving recurring
messages to LOG level
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_handle_src_query),
(gst_qtdemux_handle_src_event), (gst_qtdemux_loop_header),
(qtdemux_inflate), (qtdemux_parse), (qtdemux_parse_trak),
(qtdemux_parse_udta), (qtdemux_tag_add_str), (qtdemux_tag_add_num),
(qtdemux_tag_add_gnre), (gst_qtdemux_handle_esds),
(qtdemux_video_caps), (qtdemux_audio_caps):
* gst/qtdemux/qtdemux.h:
Some QT demux loving.
Handle seeking in a less broken way.
Fix AMR caps to match the AMR decoder.
Set first timestamp on AMR samples to 0 for now.
Remove some \n in DEBUG strings.
Use _scale_int for maximum precision.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_loop_header),
(gst_qtdemux_add_stream), (qtdemux_parse_trak):
More coherent framerate setting on caps.
If sample_size is available, use that for the samples' duration in
the index. This enables single frame streams to work (and I imagine
fixes some other cases).
Tested on testsuite, no regression.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (qtdemux_parse_trak),
(qtdemux_audio_caps):
'twos' and 'sowt' fourcc can be 16bit or 8bit audio.
Fix 8bit case (#327133, based on patch by: Fabrizio
Gennari <fabrizio dot ge at tiscali dot it>).
Also, "G_LITTLE_ENDIAN" and "G_BIG_ENDIAN" are not
valid literals for endianness in caps strings,
only "LITTLE_ENDIAN" and "BIG_ENDIAN" are valid.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_init),
(gst_qtdemux_send_event), (gst_qtdemux_handle_src_event),
(gst_qtdemux_change_state), (gst_qtdemux_loop_header):
* gst/qtdemux/qtdemux.h:
Fix seeking for quicktime files. Could still use some more
love and sophistication.
Original commit message from CVS:
reviewed by: Edward Hervey <edward@fluendo.com>
* gst/qtdemux/qtdemux.c: (qtdemux_video_caps):
Add support for Indeo3 video in Quicktime files.
Closes#326524
Original commit message from CVS:
* ext/swfdec/gstswfdec.c: (gst_swfdec_class_init),
(gst_swfdec_chain), (gst_swfdec_render):
Add debugging category and return GstFlowReturn in the right places
* ext/wavpack/gstwavpackdec.c: (gst_wavpack_dec_link):
Get something from the peer pad once we've checked if there is a peer pad.
* gst/qtdemux/qtdemux.c: (gst_qtdemux_change_state),
(qtdemux_tree_get_child_by_type), (qtdemux_parse_trak),
(qtdemux_video_caps):
Couple of fixes
Original commit message from CVS:
* ext/faad/gstfaad.c: (gst_faad_chanpos_to_gst),
(gst_faad_update_caps):
Assume that an unknown channel mapping with 2 channels
is stereo and play it that way instead of erroring.
* gst/qtdemux/qtdemux.c: (gst_qtdemux_loop_header),
(gst_qtdemux_add_stream), (qtdemux_parse_trak):
Handle e.g. jpeg streams with 0 duration frames as having 0 framerate.
Debug fixes. Some 64 bit variable fixes
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_loop_header),
(gst_qtdemux_add_stream):
Memleak fixes.
Send out EOS for valid reasons (couldn't pull_range() from upstream
for example).
Original commit message from CVS:
* ext/faad/gstfaad.c: (gst_faad_srcgetcaps):
Handle gracefully the consequence of "Maximum number of scalefactor
bands exceeded", which results in 0 channels with samplerates of 0.
* gst/qtdemux/qtdemux.c: (gst_qtdemux_change_state):
Do upward transitions, then call parent state_change, then do
downward transitions.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_loop_header),
(gst_qtdemux_add_stream), (qtdemux_dump_mvhd),
(qtdemux_parse_trak):
Convert to fractional framerates
Original commit message from CVS:
Reviewed by: Tim-Philipp Müller <tim at centricular dot net>
* gst/qtdemux/qtdemux.c: (qtdemux_parse_udta):
Add support for custom genre tags.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_class_init),
(qtdemux_parse_tree):
Remove 'got-redirect' signal and post element message
on the bus instead.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_loop_header):
* gst/tta/gstttaparse.c: (gst_tta_parse_src_event),
(gst_tta_parse_parse_header):
newsegment API update.
Original commit message from CVS:
* gst/avi/gstavidemux.c: (gst_avi_demux_parse_stream),
(gst_avi_demux_stream_header):
* gst/qtdemux/qtdemux.c: (gst_qtdemux_change_state),
(gst_qtdemux_add_stream), (qtdemux_parse_tree):
Uncomment metadata and codec-name handling.
Original commit message from CVS:
* ext/faad/gstfaad.c: (gst_faad_class_init), (gst_faad_setcaps):
Add debug category, remove Close() call that made it crash
whenever reusing, renegotiating or anything; Close() actually
free()s the handle and should only be called on READY->NULL.
* gst/qtdemux/qtdemux.c: (gst_qtdemux_loop_header):
Actually set caps on buffer (in addition to pad), also.
Original commit message from CVS:
Reviewed by: Ronald S. Bultje <rbultje@ronald.bitfreak.net>
* gst/qtdemux/qtdemux.c: (gst_qtdemux_change_state),
(gst_qtdemux_loop_header):
Re-apply patch from #142272 that allows non-seekable sources,
re-proposed by Daniel Drake <dsd@gentoo.org>.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_change_state),
(gst_qtdemux_loop_header), (gst_qtdemux_handle_esds):
More memory leak fixes (#149162).
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (qtdemux_type_get), (qtdemux_audio_caps):
* gst/typefind/gsttypefindfunctions.c: (q3gp_type_find),
(plugin_init):
Add 3GP (variables name Q3GP because they can't start with a
number). Add samr audio fourcc (used in .3gp files), decoder
is work in progress. Also do a GST_WARNING instead of ERROR
in case of unknown nodes, to decrease output.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_loop_header),
(gst_qtdemux_add_stream), (qtdemux_parse), (qtdemux_parse_tree),
(qtdemux_parse_udta), (qtdemux_tag_add), (gst_qtdemux_handle_esds):
Change all g_print()s to debugging. Add a bunch of consistency
checks.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_loop_header),
(qtdemux_parse), (gst_qtdemux_handle_esds):
An esds box is not a container.
Fix parsing of mp4v boxes.
Do not try to renegotiate fps for each frame. Need to
find a better method. This should fix mp4 playback.
Original commit message from CVS:
* gst-libs/gst/riff/riff-media.c:
(gst_riff_create_video_caps_with_data),
(gst_riff_create_video_template_caps):
Add DIB fourcc (raw, palettized 8-bit RGB).
* gst-libs/gst/riff/riff-read.c:
(gst_riff_read_strf_vids_with_data):
Oops, fix strf_data reading bug.
* gst/avi/gstavidemux.c: (gst_avi_demux_add_stream):
Use a non-NULL tag.
* gst/qtdemux/qtdemux.c: (qtdemux_parse_trak):
Time for hacks. Sorry Dave. At least one quicktime movie (a
trailer) that I've encountered contains multiple video tracks.
One of those is the actual video track, the other are one-frame
tracks (images). Unfortunately, the number of frames according
to the trak header is 1 for each, so that doesn't help. So
instead, I look at the duration and discard tracks with a
duration shorter than 20% of the length of the stream. Better
than nothing.
Original commit message from CVS:
* ext/dirac/Makefile.am:
* ext/dirac/gstdirac.cc:
* ext/dirac/gstdiracdec.cc:
* ext/dirac/gstdiracdec.h:
Do something. Don't actually know if this works because I don't
have a demuxer yet.
* ext/gsm/gstgsmdec.c: (gst_gsmdec_getcaps):
Add channels=1 to caps returned from _getcaps().
* ext/ogg/gstogmparse.c: (gst_ogm_audio_parse_get_type),
(gst_ogm_video_parse_get_type), (gst_ogm_audio_parse_base_init),
(gst_ogm_video_parse_base_init), (gst_ogm_parse_init),
(gst_ogm_audio_parse_init), (gst_ogm_video_parse_init),
(gst_ogm_parse_sink_convert), (gst_ogm_parse_chain),
(gst_ogm_parse_change_state):
Separate between audio/video so ogmaudioparse actually uses the
audio pad templates. Both audio and video work now, including
autoplugging. Also use sometimes-srcpad hack.
* gst-libs/gst/riff/riff-read.c: (gst_riff_read_seek):
Handle events better. Don't hang on infinite loops.
* gst/avi/gstavidemux.c: (gst_avi_demux_class_init),
(gst_avi_demux_init), (gst_avi_demux_reset),
(gst_avi_demux_src_convert), (gst_avi_demux_handle_src_query),
(gst_avi_demux_stream_header), (gst_avi_demux_stream_data),
(gst_avi_demux_change_state):
* gst/avi/gstavidemux.h:
Improve A/V sync. Still not perfect.
* gst/matroska/ebml-read.c: (gst_ebml_read_seek),
(gst_ebml_read_skip):
Handle events better.
* gst/qtdemux/qtdemux.c: (gst_qtdemux_handle_sink_event),
(gst_qtdemux_loop_header), (qtdemux_parse_trak),
(qtdemux_audio_caps):
Add IMA4. Improve event handling. Save offset after a seek when
the headers are at the end of the file so that we don't end up in
an infinite loop.
* gst/typefind/gsttypefindfunctions.c: (qt_type_find):
Add low-priority typefind support for files with no length.
Original commit message from CVS:
* configure.ac: remove NASM check, since we don't use it. Update
dirac check to 0.4
* ext/dirac/gstdiracdec.cc: update to current 0.4 API
* gst/audioconvert/gstaudioconvert.c: (gst_audio_convert_link):
Initialized variables.
* gst/qtdemux/qtdemux.c: (gst_qtdemux_change_state),
(gst_qtdemux_loop_header), (qtdemux_parse), (qtdemux_parse_trak),
(gst_qtdemux_handle_esds), (qtdemux_audio_caps): Fix seeking, add
SVQ3 format
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_add_stream),
(qtdemux_parse_trak):
Don't crash by dividing by zero (see sample movie in #126922).
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_init),
(gst_qtdemux_handle_src_query), (gst_qtdemux_handle_src_event),
(gst_qtdemux_handle_sink_event), (gst_qtdemux_change_state),
(gst_qtdemux_loop_header), (qtdemux_dump_mvhd),
(qtdemux_parse_trak):
* gst/qtdemux/qtdemux.h:
Bitch. Also known as seeking, querying & co.
* sys/oss/gstosssink.c: (gst_osssink_init), (gst_osssink_chain),
(gst_osssink_change_state):
* sys/oss/gstosssink.h:
Resyncing is for weenies, this hack is no longer needed and was
broken anyway (since it - unintendedly - always leaves resync to
TRUE).
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_change_state),
(gst_qtdemux_loop_header): Patch from dcm@acm.org (David Moore)
to allow qtdemux to use non-seekable streams. (bug #142272)
Original commit message from CVS:
* ext/esd/esdsink.c: (gst_esdsink_chain): Fix crash when ESD
is killed while we're playing.
* gst/qtdemux/qtdemux.c: (qtdemux_parse): call
gst_element_no_more_pads().
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_loop_header),
(gst_qtdemux_add_stream), (qtdemux_parse), (qtdemux_type_get),
(qtdemux_dump_stsz), (qtdemux_dump_stco), (qtdemux_dump_co64),
(qtdemux_dump_unknown), (qtdemux_parse_tree), (qtdemux_parse_udta),
(qtdemux_tag_add), (get_size), (gst_qtdemux_handle_esds): More qtdemux
hackage -- parse a lot more atoms, extract a few tags. One might even
mistake this for tag support. Maybe it is.
* gst/qtdemux/qtdemux.h:
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_add_stream), (qtdemux_parse),
(qtdemux_parse_trak), (get_size), (gst_qtdemux_handle_esds): Hacked
up qtdemux to make it spit out codec_data. Do _not_ look at this
code; you will no longer respect me.
Original commit message from CVS:
* ext/jpeg/gstjpegdec.c: (gst_jpegdec_get_type):
* ext/jpeg/gstjpegenc.c: (gst_jpegenc_get_type),
(gst_jpegenc_getcaps):
move format setting to inner loop
* gst/ffmpegcolorspace/gstffmpegcolorspace.c:
(gst_ffmpegcolorspace_getcaps):
use GST_PAD_CAPS if available so that we use already negotiated
caps
* gst/qtdemux/qtdemux.c: (gst_qtdemux_loop_header),
(qtdemux_parse_moov), (qtdemux_parse):
extra debugging
* sys/qcam/qcam-Linux.c: (qc_lock_wait), (qc_unlock):
* sys/qcam/qcam-os.c: (qc_lock_wait), (qc_unlock):
move hardcoded path to DEFINE
Original commit message from CVS:
* ext/divx/gstdivxdec.c: (plugin_init):
Remove comment that makes no sense.
* ext/mad/gstid3tag.c: (gst_id3_tag_set_property):
Fix for obvious typo that resulted in warnings during gst-register.
* ext/xvid/gstxviddec.c: (gst_xviddec_src_link),
(gst_xviddec_sink_link):
Fix caps negotiation a bit better.
* gst/qtdemux/qtdemux.c: (qtdemux_parse_trak):
We call this 'codec_data', not 'esds'.
Original commit message from CVS:
* gst/qtdemux/qtdemux.c: (gst_qtdemux_handle_sink_event),
(qtdemux_parse), (qtdemux_type_get), (qtdemux_dump_mvhd),
(qtdemux_dump_tkhd), (qtdemux_dump_stsd), (qtdemux_dump_unknown),
(qtdemux_parse_trak), (qtdemux_video_caps), (qtdemux_audio_caps):
A number of new features and hacks to extract the esds atom and
put it into the caps. (bug #137724)
Original commit message from CVS:
reviewed by: <delete if not using a buddy>
* gst-libs/gst/riff/riff-media.c:
(gst_riff_create_video_caps_with_data),
(gst_riff_create_video_caps), (gst_riff_create_audio_caps),
(gst_riff_create_video_template_caps),
(gst_riff_create_audio_template_caps):
* gst-libs/gst/riff/riff-media.h:
* gst-libs/gst/riff/riff-read.c:
(gst_riff_read_strf_vids_with_data), (gst_riff_read_strf_vids):
* gst-libs/gst/riff/riff-read.h:
* gst/avi/gstavidemux.c: (gst_avi_demux_add_stream):
Add MS RLE support. I added some functions to read out strf chunks
into strf chunks and the data behind it. This is usually color
palettes (as in RLE, but also in 8-bit RGB). Also use those during
caps creation. Lastly, add ADPCM (similar to wavparse - which
should eventually be rifflib based).
* gst/matroska/matroska-demux.c: (gst_matroska_demux_class_init),
(gst_matroska_demux_init), (gst_matroska_demux_reset):
* gst/matroska/matroska-demux.h:
Remove placeholders for some prehistoric tagging system. Didn't add
support for any tag system really anyway.
* gst/qtdemux/qtdemux.c:
Add support for audio/x-m4a (MPEG-4) through spider.
* gst/wavparse/gstwavparse.c: (gst_wavparse_parse_fmt),
(gst_wavparse_loop):
ADPCM support (#135862). Increase max. buffer size because we
cannot split buffers for ADPCM (screws references) and I've seen
files with 2048 byte chunks. 4096 seems safe for now.
