Multiple Changesets Without Ancestors

[Pierre-Yves David]

On 05/10/2015 03:18 AM, Lasse Kliemann wrote:
> After some stress-testing (automated change of files, commit, pull,
> push, merge, ...) my test repos shows a couple of changesets that have
> no ancestors (parent is -1).
>
> 'hg verify' runs fine.

Having multiple roots is not an error per se.

> Questions:
>
> 1.) This should not happen, or did I miss something?

This should not happen. You mention stress-testing and I this ring some 
bell. I think I've already heard someone saying this happened during 
stress testing. This could be a subtle race condition in the commit code 
when multiple process are doing a lot of them?

Adding Durham and Martin to the thread as they may be the one who 
reported that.

> 2.) How can the situation be diagnosed further?

Can you reliably reproduce this behavior? What version of everything are 
you using? What is you script doing exactly?

> Of course, the repo is not important to me (just dummy data), but I'm
> apprehensive that something similar might happen in a production repos,
> and then I need to know what to do.

in 10 years of Mercurial. I've never heard of this outside of tress 
testing (which I did once, recently). So you are most probably fine for 
your real life usage

> Ok... just now I see, the hg I use on some of the systems is
> ancient. It's 2.8.2, still shipped with Ubuntu in 2015!
> Maybe that's the problem?

Please try again with 3.4


-- 
Pierre-Yves David