On Nov 8 2017 16:05, Takashi Iwai wrote:
On Tue, 07 Nov 2017 23:36:28 +0100, Takashi Sakamoto wrote:
On Nov 7 2017 16:59, Takashi Iwai wrote:
On Tue, 07 Nov 2017 01:34:02 +0100, Takashi Sakamoto wrote:
This commit deprecates some APIs related to the dimension information. They are planned to be removed in a development period for Linux kernel v4.21.
IMO, the version to deprecate the feature may depend on LTS kernel version. 4.21 looks OK, but we may shift it.
I still get no merit of your suggestion. Do you mean that the removal of ABI should be done in several releases before target LTS will be actually released, to produce stable ABI? If so, we need to estimate the timing of next LTS. However, it's hard because for recent LTS we got no official announcement from stable maintainers till LTS version of Linux kernel is actually released. At least, I won't work with such uncertain estimations.
The feature removal isn't what we can guarantee at a certain kernel version. As long as we have a real (not theoretical) user, it shouldn't be dropped. It's still unwritten "no regression" rule.
That said, we may announce and prepare / plan the deprecation, but we can't define the exact kernel version in general. And considering LTS release is one of the decision factors.
If it comes from a general theory, I'm willing to follow it.
But in this case, we can be optimistic. As you know, the minor feature is just used in drivers/software related to models produced by Echo Digital Audio corp. The other probability is usage by user-defined control element set on any userspace application. As my recent work declared, this feature has been long-abandoned and no application exists (if such application exists, developers have already found such bugs which I've fixed recently before I worked...).
If the feature were used by majority, it would be more difficult to deprecate/remove it, and we need to be deliberate. Depending on cases, deprecation is impossible itself. But this feature is quite minor and got little popularity. Practically, we can safely remove the minor feature in any timing after modified version of echoaudio stuffs are disseminated. My proposed schedule takes 6 kernel releases for it. In the schedule, v4.21 release is the deadline of the feature, but if you'd like to keep the deadline undecided for such theory/rule, I have no objection to it. To me, deprecation of this feature for future removal is more important to avoid developers' misuses to which I addressed, than actual removal.
I have no objection of removal of API in alsa-lib, but my concern is that unnecessary announcement of the exact kernel version as if it's an ultimately fixed deadline. It's rather a "hopefully working" deadline, and we don't need to show it in alsa-lib side.
Now I understand your concern.
We can update the comment once after it's really deprecated, if any.
I'm OK for this suggestion. Later, I'll post updated version of this patchset.
Thanks
Takashi Sakamoto