Re: [PATCH net-next] docs: netdev: clarify the need to sending reverts as patches

Re: [PATCH net-next] docs: netdev: clarify the need to sending reverts as patches

[Thorsten Leemhuis]

On 27.03.23 19:26, Jakub Kicinski wrote:
> We don't state explicitly that reverts need to be submitted
> as a patch. It occasionally comes up.
>  [...]
> +In cases where full revert is needed the revert has to be submitted
> +as a patch to the list with a commit message explaining the technical
> +problems with the reverted commit. Reverts should be used as a last resort,
> +when original change is completely wrong; incremental fixes are preferred.
> +

FWIW, I see how this is well meant, but I'm not really happy with the
last sentence, as one of the problems I notice when handling regression
is: it sometimes takes weeks to get regressions fixed that could have
been solved quickly by reverting the culprit (and reapplying an improved
version of the change or the change together and a fix later). That's
why Documentation/process/handling-regressions.rst strongly suggest to
revert changes that cause regressions if the problem can't be fixed
quickly -- especially if the change made it into a proper release. The
two texts thus now not slightly contradict each other.

I noticed that this change was already applied, but how would you feel
about changing the second sentence into something like this maybe?

```
Use reverts to quickly fix regressions that otherwise would take too
long to resolve. Apart from this reverts should be used as a last
resort, when the original change is completely wrong; incremental fixes
are preferred.
```

Or maybe this?

```
Incremental fixes in general are preferred over reverts, but the latter
are useful to quickly fix regressions that otherwise would take too long
to resolve. Apart from this reverts should be used as a last resort,
when the original change is completely wrong.
```

Ciao, Thorsten

Re: [PATCH net-next] docs: netdev: clarify the need to sending reverts as patches

[Jakub Kicinski]

On Wed, 29 Mar 2023 11:04:01 +0200 Thorsten Leemhuis wrote:
> FWIW, I see how this is well meant, but I'm not really happy with the
> last sentence, as one of the problems I notice when handling regression
> is: it sometimes takes weeks to get regressions fixed that could have
> been solved quickly by reverting the culprit (and reapplying an improved
> version of the change or the change together and a fix later). That's
> why Documentation/process/handling-regressions.rst strongly suggest to
> revert changes that cause regressions if the problem can't be fixed
> quickly -- especially if the change made it into a proper release. The
> two texts thus now not slightly contradict each other.
> 
> I noticed that this change was already applied, but how would you feel
> about changing the second sentence into something like this maybe?

Please escalate the cases which can be fixed by easy reverts because 
I can't think of any in networking :(

The entire doc is based on our painful experience telling people the
same thing over and over again, we don't want to include things which
don't actually happen on netdev. Longer the doc is the less likely
people will actually read it :(