[JENKINS-39968] Atomically create changelog#.xml file during checkout.

[tmcsantos]

see JENKINS-39968 for details.

[MarkEWaite]

@tmcsantos Thanks very much for the pull request!

Unfortunately, when I run it with my test job, I see the same behavior in the return value from the checkout step as I did before the change.

Can you confirm that the checkout step is returning the expected value when you use the plugin including the changes in this pull request? If you've confirmed that, could you compare your working case to my non-working case and help me understand my mistake?

[tmcsantos]

Hi @MarkEWaite
Thank you for your time reviewing this. Your test job seems to be testing a different issue, this one deals only with the changelog#.xml file creation, that was being shared across concurrent threads (failed to grab a unique changelog.xml file for creation). This file gets referenced in the build.xml as stated in the comments section for JENKINS-39968, meaning same changelog for distinct repositories.

In your example you seem to be fetching the same repo, but with different branch/tag. That still fails and is not a concurrent issue, because it fails even in a synchronous way.

So in a way, I think that JENKINS-53346 is not directly related to JENKINS-39968.

Hope it helped clear some things.

[MarkEWaite]

Thanks for the clarification @tmcsantos . You're absolutely correct!

I was surprised at that, since I assumed that the same problem was at the root of reporting changes from

• two checkouts of distinct branches from two distinct repositories
• two checkouts of distinct branches from the same repository

Before this change, they behave the same. After this change, the distinct branches case is much better behaved.

This change resolves JENKINS-39968 but does not resolve JENKINS-53346. A fix for JENKINS-53346 will require other changes.

[tmcsantos]

Thanks @MarkEWaite , as for JENKINS-53346 I may happen to have a fix, I'll submit one in time for the appropriate project.

[jglick]

As I noted in #32 (comment), this seems like it could introduce its own problems.

[tmcsantos]

I see, I didn't take in consideration failed checkouts, nor that empty changelog files posed a threat. I can go with #32, fixes the same issue, but the proposed change needs some tweaking. see #32 (comment)