pinchartl: media build seems to be showing his age. Here (Fedora), it also doens't build as-is, as Fedora is now built with RC drivers builtin
I guess something needs to be changed there, in order for it to honor built-in core infrastructure, when possible
a change like that can be painful, though, as the kAPI may be different from the one inside media-build and the one at the main Kernel
at the case I used it (in order to build vidtv driver on one of my machines), I just disabled the CONFIG_* that I was not needing, so no issue. But doing something like that at the generic case can be very painful
hverkuil, syoung: fyi, I'm planning to apply today the vidtv patches I submitted, as they solve some troubles at linux-next build for media stuff
mchehab: you can always recompile the distro kernel to have all of the media subsystem as modules
the issue with dmabuf is that it can't be compiled as a module
pinchartl: yes, true...
but one of the original goals of media-build were to help non-developers to test changes at media drivers
for such usage, it would be good to be able to test things with distro-provided Kernels
I agree
btw, something like this can be used to test any driver against the current Kernel, without needing to build the entire media stack: https://pastebin.com/gJ4xkKws
one of the reasons why media-build is somewhat complex is that it used to allows using "make menuconfig" for configuring the media stack
it sounds that some recent change upstream broke support for it
so, maybe it is time to change how media-build works - if someone has the time for it
or instead add something like: ./build <driver> that would create a makefile like the above and build just a single driver using the existing media stack
not sure how easy/hard would be to maintain something like that, though
mchehab: your small makefile assumes that the APIs used by the drivers don't differ between the media tree and the running kernel
I was using media_build to test the UVC driver against my running kernel, which is v5.4.60
so that approach won't work
pinchartl: yes, it assumes that kAPI is identical
it should probably work for a driver like UVC that has been stable for a while
(probably with minimal efforts to address some minimal things(
it probably won't work for more complex changes, or when the running Kernel is too old
hverkuil: pinchartl: with regards to this patch:
https://patchwork.linuxtv.org/project/linux-media/patch/20200818133608.462514-7-gregkh@linuxfoundation.org/
can I merge it?
pinchartl argued against applying it
I suspect that Greg is doing such change treewide
mchehab: I'd like Greg to reply before making a decision
pinchartl: ok, I'll drop it from the patches hverkuil submitted me on his PR
you should ping him again about that, as you replied to him about one month ago
as he didn't answer yet, I assume he probably missed your reply
(or perhaps there were already some discussions about that at lkml)