[ANN] I will stop applying driver patches

[ANN] I will stop applying driver patches

[Johannes Berg]

Hi,

I'm not writing this happily, but seems I have no choice but to declare
"patch review bankruptcy".

When Kalle stepped down close to two years ago, I thought I made it
clear that I'd expect everyone here to help review patches, especially
driver ones, if I was going to start applying them.

Guess what didn't happen.

By my count, excluding iwlwifi and hwsim, I've applied ~360 driver
patches since, of which less than half have any Reviewed-by/Acked-by
tags [1]. In turn, this probably means I've (had to) review(ed) all the
others, and almost certainly did review more since I had to handle all
of them.

[1] I have an agreement with Arend that I apply patches to the
brcm[fs]mac driver, so those are included in the ones with tags.

That number doesn't even sound that big, but it never was something I
actually wanted to do, and never mind that I also applied 565 wifi stack
patches (plus ~300 of my own) with 22 (!) review tags from outside Intel
(where we have a process in place, so many of our own patches carry
reviewed-by tags) in the same time.

I also currently have/had (see below) another set of ~80 driver patches
in patchwork assigned to me right now, and 239 patches overall (though
about 1/3rd of that are probably my own [2]).

[2] which is a whole other problem, I've obviously also spent time
writing them, and almost nobody bothers (dares? [3]) reviewing those
either (at least in public)

[3] surely I must have a reputation for being acerbic ... right now I'm
just tired


And then ... at the same time, a lot of people really like to pile on
pressure:

 - regressions/bugs must be fixed 'within a week' (hi Thorsten)

 - patches to decades-old drivers are suddenly so important that people
   want them applied yesterday

 - roughly a million people who figured out how to push a button in an
   LLM chatbot think they're entitled to reviews that they immediately
   not read but just feed into said LLM again


So that's it, I can't do it any more. I'd say it was fun while it
lasted, but honestly, it really wasn't.

I don't know where we go from here.

I don't even know if this (alone) will help get my workload back to
reasonable levels, just the wifi stack alone feels more than enough
right now. Maybe I'll come back in a few weeks and say that I'll only
ever even read/apply patches from people/groups/companies who have
enough Reviewed-by lines themselves, or something like that.

I don't know if maybe someone will start maintaining some of those old
drivers and send pull requests, but right now I have no reason to think
there's anyone interested in it I'd even trust enough to pull code from,
almost nobody has shown that they even are able to review patches in any
way. I suspect the intersection of the sets "people who know anything
about the code" and "people who want to change it" is literally empty,
since everything is clearly just LLM driven.

I do know what I've done with the patches - I've delegated all driver
patches in patchwork to the 'wifibot' account (because nobody runs it),
and set up an auto-delegation rule.

And finally, lest this be misinterpreted otherwise:

I do also really like the actual technical work, figuring out the design
of how all the things fit together into the bigger picture, having those
discussions with people (!), thinking about problems and trying to fix
them, etc. So I'm not going away, I'm just saying I need to focus on the
wireless stack. There's more than enough going on to keep me busy for a
while yet: SMD transition AP & client side, NAN, (UHR) critical update
AP & client side, UHR mode management (client side), general
improvements like the RTNL work, latency reductions, TXQ work, etc.

johannes

Re: [ANN] I will stop applying driver patches

[Alessio Ferri]

Hi Johannes,

in defense of the people who "send patches to decades-old drivers":
Those drivers are incomplete/buggy/a pain to deal with, usually because
of the poor vendor support or totally missing documentation (no need to
name names). 
A lot of people (I am one of them) have had to deal with them for ages,
with great suffering and little hope of ever fixing the situation,
because of missing developers interested in fixing them / no
documentation to start from / no time to reverse the whole thing.
Finally today, with some assistance from an LLM, one could hope to have
working hardware with Linux, or at least try to; it seems then natural
to send patches upstream for review, instead of keeping them private.

Sometimes, even with an LLM helping, the work takes months.
Sometimes, the contributor has no idea of what they are doing, and
I get why this is a problem, after all maintainers are not review
machines.
But, there has to be a middle ground, maybe a small overall quota per
merge window would push people to self-review their own work more
carefully. That would avoid creating perverse incentives, like
forcing people to give reviews, with the risk of having people give away
random Reviewed-by tags just to reach the required number.

Best regards
Alessio