[Python-buildbots] Idea: have a different buildbot server for unstable workers?

[Victor Stinner]

Hello,

Le dim. 24 mars 2019 à 16:32, Michael <aixtools at felt.demon.nl> a écrit :
> Understood - and have, after I better understood the reluctance for any
> AIX related patch - took, I think it was yours (Victor) advice - worry
> about master, and when that is done (all test pass) - re-evaluate.

I told you that AIX is not my priority. I have limited time to work on
Python upstream, and so I make choices to prioritize and organize my
time. I told you how I work so you can adjust your expectations in
term in availability of core developers. I contacted you, since I
guess that other core devs will not, and you was asking questions
about (the lack of) reviews.

There is no such "reluctance" from me.

It seems like your underestimate how much time I spent on your AIX
changes. I had to not review other changes while reviewing your
changes. I had to not fix other bugs while looking at your bugs. Time
is precious :-) You likely also underestimate the cost of just basic
maintenance on Windows, macOS and Linux: just ensure that the CI is
"not broken": it's between 20% and 40% of my whole week. Supporting a
new platform would only increase this ratio for me, whereas I would
like to reduce this ratio. Hopefully, slowly, I'm getting help on
maintainaing these platforms :-) Sometimes, I don't fix regressions, I
just report them. But producing an useful bug report requires a look a
little bit at the bug, it takes me a lot of time (more than what I
would like to spend).


> I have also already started maintaining my own forks for 2.7, and for a
> while 3.4 through 3.6.
>
> I have no issue doing the backports myself. I might need to learn some
> tricks with git to make it nearly automated, but I do not want to have
> the expectation that they will be considered, when in fact, there seems
> to be more aversion than willingness to accept.

All Linux distributions maintain downstream patches for whatever
reasons. There is a lot of tooling to help you in this task. Git is a
good start :-)



> I gave my word, and shall keep it. But only if I feel there is some
> level of acceptance. If you really do not want AIX supported - I am
> finally ready to accept the hint. I'll leave you all alone (and wish you
> well).

I told you how I work, it's a matter of priorities. For example, Red
Hat customers issues are more important than any Python upstream
issues for me ;-)


> I too am a volunteer. I have no commercial benefit from anything I have
> done. Nor do I expect any from any future action. No manager of mine is
> encouraging me to do this - so no "bonus" at all. I hoped I was
> contributing.

I'm not sure why you are writing that.

Are you expecting that your changes deserve to be reviewed because you
are working on your free time on Python? Do you think that other
contributors are paid to write their changes?

If you expected to get rewarded, maybe you chose the wrong project :-)
Python doesn't work like that. The motto is "We are all volunteer" :-)


> So, to get back to core question: If I backport the patches to the
> failing tests will they be considered for merge?

Just to elaborate what I wrote previously: while I had limited time to
review AIX changes on the master branch, fixing issues specific to AIX
on 3.7 is a low priority for me. I only see 3.7 changes as a threat to
3.7 stability. I'm talking about the general case, not the specific
case, so I'm always very careful with changes in stable branches. For
example, we are rejecting more and more bugfixes in 2.7 because of the
high risk of regressions introduced by subtle behavior changes, even
when the reported bug is severe (sadly, more and more severe bugs are
known).

For AIX, my priority remains the master branch, and not all AIX
buildbots are green yet. The following bug seems severe, it's a crash:
https://bugs.python.org/issue36273

Victor
-- 
Night gathers, and now my watch begins. It shall not end until my death.