Magic Lantern 2024 summary, 2025 preview

[Walter Schulz]

Or get yourself a 550D and help finding out where button triggering BGMT_PRESS_FLASH_MOVIE is located, what it does and why we have 3 related events. /s

[SpicyBoi]

Yup, discord is the best place for quick discussion.

6D2 is in the reliable early port stage.  This would let you have fair confidence that what's there (not much), works.  R is Digic 8.  None of these cams are reliable yet.  If you enjoy debugging, we'd welcome the help on Digic 8 (maybe I won't have to do it!).  If you prefer reverse engineering and normal dev work, 6D2 is a more obvious choice.

If you'd prefer to avoid reverse engineering, and stick to normal dev work - there's still lots to do on well supported cams.  The good parts of various forks want merging into official repo, known bugs want fixing, menus could do with a re-org and cleanup.  Etc!

I think the Canon R is a logical choice for me  as I like to shoot with vintage lenses and I have a lot of TV lenses which can be adapted to it.

Dose the canon R have at least hd raw recording at the moment?

[names_are_hard]

The R does not have working raw video of any kind.  We do know the principles required to get it working, and it will be possible to support on this cam.  Kitor has done proof of concept work on a few D8 cams so might have more detailed info.  No D8 cams are stable enough for public releases.

200D is the closest new cam to having raw video.  It works there with some significant glitches needing fixing (timing clocks work differently on modern cams and this breaks many assumptions the old code makes).

[kitor]

Technically R has all the hardware to do 1080p60 on a fast card, and with code I have on branch it can record full sensor width 1080p60 uncompressed 14 bit RAW with I believe line binning (that was checked by Walter).

But we don't understand how to control EDMAC on Digic 8 (it has different hardware than previous generations) so it was a sheer luck we found a working combination to do it at all. There was no attempts to do 1:1 zoom modes recording and with our current knowledge 12/10 bit and lossless can't be done. I mean in terms of "we don't know how to configure hardware to achieve this".

Digic 8 has more problems, it requires much rework in our codebase (Canon changed data structures for basic things like ISO or Aperture values in a way those are incompatible with our core data structures), and - as Names pointed out - for a good 3 years or more we are trying to catch major instability problem on those platforms.

I would like to explore possibility of such literally "unstable dev super-pre-alpha" Digic 8 builds in foreseeable future, but they will be fundamentally broken in many areas and targeted towards chasing the root cause of mentioned unstability. But that for sure won't happen before, or close to first "reboot" releases.

[Walter Schulz]

Technically R has all the hardware to do 1080p60 on a fast card, and with code I have on branch it can record full sensor width 1080p60 uncompressed 14 bit RAW with I believe line binning (that was checked by Walter).

At least strong indication. 5D4 and R may use the same sensor type. At least 1080p30 mode seems not to use line skipping (not sure I tested 60 fps).

BTW: 7D2 doesn't use line skipping either. The only Canon APS-C DSLR AFAIK.
BTW2: I kicked pre-ML from my R. Just beyond useless.

[SpicyBoi]

At least strong indication. 5D4 and R may use the same sensor type. At least 1080p30 mode seems not to use line skipping (not sure I tested 60 fps).

BTW: 7D2 doesn't use line skipping either. The only Canon APS-C DSLR AFAIK.
BTW2: I kicked pre-ML from my R. Just beyond useless.

Canon 7D2 sounds fascinating, what's the status on that can it record raw yet?

I want not even aware it existed. 😁

I looked more into the R and I think I will get a 6dmkii to help out the R stuff is beyond me a bit I need more experiences with hacking the system.

[names_are_hard]

I can extract raw sensor data from 7d2, there are enough problems remaining that I wouldn't call it "recording raw" yet.  The camera has an unusual design with EDMAC operations happening indirectly from the ICU.  We may need to build an abstraction layer to wrap EDMAC ops for this cam (which would be a benefit for all cams).  Combined card performance is nice and high on this cam.  Heder made useful progress last year but didn't come back to finish it off: https://www.magiclantern.fm/forum/index.php?topic=13746.625

I've re-implemented everything to a similar level as in that post but with cleaned up code.  Not yet pushed to my repo since it's unstable and missing some pieces.  Happy to share if you want to work on that cam.

Similar for 6d2.  It's 90% working but there's a hard problem remaining around EDMAC channel reservation / usage.

[SpicyBoi]

Heder made useful progress last year but didn't come back to finish it off: https://www.magiclantern.fm/forum/index.php?topic=13746.625

I've re-implemented everything to a similar level as in that post but with cleaned up code.  Not yet pushed to my repo since it's unstable and missing some pieces.  Happy to share if you want to work on that cam.

A 7d2 is on order will have it in a few days, please send me a link to the build and lets get things started.

[names_are_hard]

Cool - the repo is here: https://github.com/reticulatedpines/magiclantern_simplified/

Should build easily on modern linux.  Debian is preferred over Ubuntu (they have a bug around Qemu).  Is known to build on WSL or Mac.

I strongly recommend joining Discord for asking dev questions - you'll get answers a lot faster, and in more detail.