- User Since
- Apr 8 2016, 6:16 AM (222 w, 3 d)
Mon, Jun 29
Sun, Jun 28
Remove Issue? "Holding, Then Releasing Key Should Always Dismiss Pie" - This mentions the menu sticking open if the mouse isn't moved... I cannot replicate this behaviour. The pie menu will activate/cancel if menu key released.
May 14 2020
May 11 2020
Commit message: Use refcounted_frames so ffmpeg can properly manage internal buffers. Use AVDictionary to set option.
Apr 15 2020
Mar 29 2020
I don't have any further additions to this patch except any requested by review.
Added more UI notifications of ffmpeg frame seek issues.
Mar 17 2020
"Clean up clip user notification warnings/errors"
Mar 16 2020
For anyone interested:
Changes and tuning thanks to ffmpeg fate-suite videos. - handles a greater range of edge cases. - warns user if a frame seek mismatch occurs. - more efficienct seeking with fallback to earlier seek. - change deprecated av_free_packet to unrefs. - use stream specific pts timing for seeks.
Mar 2 2020
Hi Sybren, Thanks for asking. I was able to get back to this after an unforeseen break.
I've tested quite a few edge cases thanks to the ffmpeg "fate" test videos - quite a learning experience.
Feb 20 2020
Updated to more recent Master as a result of user discussion at Blender Community: Right-Click Select
Jan 24 2020
Quick note: Downloaded FFMPEG test videos ( accessible via ffmpeg source: make fate-rsync ).
I'll hold off on updating code here a few more days to test these edge cases.
Jan 17 2020
Sergey and Sybren,
Jan 8 2020
Caught some bugs ( from latest changes but also one found during simulating re-use of packets )
Modified some AV_LOG_DEBUG messages to be more descriptive.
No longer uses last found pts in repeated search... impedes debugging and otherwise not necessary.
Jan 7 2020
Notes to self/FYI:
Jan 6 2020
Manage double decode buffers in own function.
Still some complex intermingling.. hoping to clean it up further and/or document.
Dec 22 2019
Thanks for your vote of confidence.
My goal is to get this cleaned up and documented such that it requires the least amount of time to review.
Dec 21 2019
Troy: Yep, the flushing of the codec is being done. ( I also double buffer the frames which adds more complexity here ). I've upgraded the ffmpeg calls away from the deprecated ones as well. If you can test with other codecs please do. The error message (with settings in replicating bug above ) I'm most interested in hearing about is "PTS for frame differs from previous decode!" and "PTS search for frame differs from previous scan!"
Dec 20 2019
This, as you might guess, makes it very challenging to seek as there's no hard and fast rules on where those I frames will reside in a bitstream.
It turns out the video proxy time indexing code in Blender ( indexer.c ) is unfinished(?). All the index files produced are identical despite the differing types.
Dec 19 2019
- fix length calculation and tune pts drift allotment.
Dec 18 2019
I've downloaded the Blender ffmpeg test videos ( and all the other test via make test ).
- improved variable names, plus changed _all_ camelcase names to underscored versions.
wow-- did I write that!!? ;) - thanks for catching that Sergey.
Dec 17 2019
- reinstate commented line from testing.
- replace deprecated ffmpeg calls
- handle EAGAIN result from avcodec_send_packet
- upgrade deprecated call to write proxy videos
Dec 16 2019
Dec 13 2019
Just a few thoughts to this point:
There are certainly users, that would appreciate VFR support, but I didn't consider this to be priority. Sequencer already have time-altering effect, so I want to create method of mapping frame you see in preview to frame number in domain of the source media (because we currently use frame numbers).
As for implementation - I would check if there is any convention of positioning elements in time with sub-frame accuracy and work from there. I only know, that curves can use frame and decimal fraction. We could also use time directly. It may be good idea to generalize and document approach in cases like this, but I am not sure at this point.
Then on FFMPEG side there should be layer, that translates Blender time (or what the convention may dictate) to pts and vice-versa. Perhaps we could use that to retrieve stream length, because now it seeems, that length is calculated only from amount of frames.
In any case, to me this sounds like effort for it's own patch
Dec 11 2019
I see that arc squashed the commits :(
I'm attaching the first commit (39e37c9) - I had referenced it above.
Nov 8 2019
"I don't think use of such strong imperative statement is something what belongs to a manual."
Thanks for being so diplomatic. In my defense I was recovering from a visit to the dentist.
Oct 21 2019
My best effort at writing some documentation.
Feb 2 2019
Please note that this small diff has been cleaned up from its originally intention (testing) to something I hope is more efficiently reviewable.
Jan 28 2019
I cleaned up my original diff and added an Interface->Pie Menus preferences entry "Pie Key Again Click" - "Pressing pie shortcut again will act as click ( to cancel or select )" which defaults to enabled.
Added a preferences item to turn on or off this action (defaults to on ). This should help avoid any unforseen issues.
Jan 17 2019
Complex Pie Menus for testing?
Are there any test pie menus hidden away... something involving pie menus with inner block menus/pie menus etc?
Default Pie Menu Action.
Also... I'm experimenting on ways to allow the short 'tap' of a pie-menu key to perform an action ( instead of sticking the pie menu ).
For example, z held down brings up the Shading Pie menu however tapping it would simply toggle wireframe mode.
Jan 8 2019
I should've removed the
/* distance from initial point */
line just above my changes in source/blender/editors/interface/interface_handlers.c
That comment looks like it was left over from earlier code?
Fix for: Holding, Then Releasing Key Should Always Dismiss Pie
I believe this fix complies with the intended behaviour of the pie menu as per the discussion at devtalk
Dec 2 2018
I believe this is fixed now.
See my comments at related bug T57157
This bug is now fixed!
I was able to reproduce it before but not after this commit:
Nov 13 2018
Yes, Bug is still there:
I updated and used my test file to get crash.. bug is still there. I also tried the original UVsphere steps from scratch and it didn't immediately crash.. but didn't paint either. I then rotated the sphere a bit and tried to draw and then it crashed.
Oct 14 2018
My observation: ps.mvert_eval.co is getting corrupted in a memcpy during python gui call.
Sep 10 2018
Issue solved via latest upgrade to all library versions.
Aug 30 2018
The NDOF ( i.e. space mouse ) menu has been adversely affected by this change ( T56547) and the earlier (T56511 ).
The menu can be brought up without a space mouse: edit->User Preferences->Input
In the Window section look for the "Call Menu" item with "NDOF Menu" key and change it to your preferred key.
Here's the affected menu:
Jul 22 2018
Running a program from Windows console ( aka command line) will spit out information in the console.
Window's Blender has a command to see the console ( Window->Toggle System Console ) but this will disappear when Blender crashes.
Jul 21 2018
I tried this out ( on Linux however ) and it segfaults after saying "Exception: Add-on 'io_curve_svg' has not been upgraded to 2.8, ignoring"