git polling failure half-resolved by copying the job

[Nathaniel Irons]

Hello,

I've spent about two days trying to troubleshoot a git polling failure
on a new jenkins master, running 2.60.3 and git plugin 3.5.1. Polling
(kicked off manually or by timer or by webhook) will run, print the
names of all current branches, then declare "No changes". If I publish a
new branch, with the single branch specifier set to **, the polling log
will include the name of a branch it has never seen before, then "No
changes". It might as well say "Nevermore".

The jenkins master is in a docker container, which I eventually tried
running on my laptop. I planned to update it to 2.70.1 and poke it with
sticks, but polling the same job succeeded on the laptop on the first
try, before I'd modified anything. Then I tried duplicating the job on
the server, and the duplicate job instantly started building in response
to SCM webhooks. The original job still refuses.

I could rely on the duplicated job, but it's super frustrating that I
can't identify the cause of the problem. I'd also prefer to keep the
build history intact, and I don't want to rely on this workaround
indefinitely should the problem recur. Any suggestions for
troubleshooting the failure in the original job?

I did try comparing the config.xml files for the original and duplicate
jobs. Aside from the job title difference in <mappedURL>, and a swap of
child order under <properties>, they're identical.

Thanks,

Nat