not sure if regression and what specific annex command caused it yet.
initially detected with 8.20210621 released and bisected to 8.20210621-12-g4b1b9d7a8 AKA 8.20210630~39: datalad tests started to fail to git push because freshly created remote was already carrying some master. Some more details on original datalad issue #5759
In the linked issue, it is apparently bisected to 4b1b9d7a83f0da71665658e6234e699e182cbddc which added annex.freezecontent-command.
But unless you're setting that config, I don't see it changing behavior, and if you are setting that config, it kind of follows that the behavior change is due to whatever you set it to?
annex.freezecontent-command. May be bisection was not entirely correct and it is some prior commit where it failed to build or smth like that - I didn't double check manually the status of prior commit etc.https://git-annex.branchable.com/bugs/regression__58___index.lck_in_shared-group_is_not_g+w/ fixed a problem where git-annex init entered an adjusted branch unexpectedly.
Entering an adjusted branch involves creating a master branch at init time, because it has to have a branch existing in order to adjust it.
So, these bugs appear to have the same cause. I'll close this dup.
But, note that the same thing could happen if git-annex init were run on a crippled filesystem, so it may be your test needs to deal with such a new-born master branch.