Unable to close milestone when running on AppVeyor but it works locally

[Jericho]

The last step of my Cake script when I publish a new release of my library is to close the GitHub milestone. Unfortunately, it's failing when my Cake script is running on AppVeyor (see AppVeyor log here and also screenshot provided below).

I'm running GitReleaseManager in Verbose mode but there is very little info about what went wrong which makes it tough to investigate the problem.

I executed the exact same command from my laptop and it completed successfully:

I have checked and double-checked the GitHub token configured in my AppVeyor account and I am confident that it is correct therefore I have ruled out that this could be caused by an authentication issue.

One thing I have noticed is that I am getting a warning about No valid version was found on Paging. In fact this warning is highlighted in red on AppVeyor which leads me to suspect that it could be more serious that a mere warning. I can see this same warning when I run GRM on my laptop but it's not highlighted, and GRM is able to continue to completion seemingly without any problems.

• Is it possible that warnings are treated differently when running GRM on a CI versus locally?
• Also, what does "no version found on paging" mean? Is this something I should investigate further or is it just an innocuous warning that I can safely ignore?

[Jericho]

Ok, I figured out a few things:

1. the warning about "no valid version" is reported by the MilestoneExtensions.Version extension method
2. As I understand it, this extension method is invoked when GRM fetches all the milestones from my GitHub repo and it tries to infer a Version(major, minor, build) for each milestone with the assumption that the title contains a string that looks something like 'x.y.z'
3. Paging which is mentioned in the warning refers to a milestone I created in my GitHub repo to conveniently group a bunch of issues that will be addressed in a future release. The name of this particular milestone does not follow the naming convention and therefore it's perfectly understandable for the extension method to warn that it was unable to derive a Version number from the title of this milestone.

So, either I am on a wild goose chase and this warning has nothing to do with the problem causing my build to fail or, as I mentioned in my original comment, this warning somehow gets treated as an exception but only on AppVeyor.

[gep13]

@AdmiringWorm are you aware of any issues in this area? Specifically with regard to how things work on AppVeyor compared to locally?

@Jericho are you in a position to run a test without the Paging milestone in place, to see if the AppVeyor build succeeds?

[Jericho]

are you in a position to run a test without the Paging milestone in place, to see if the AppVeyor build succeeds?

I renamed the "Paging" milestone to "99.99.99 - Paging" to satisfy the naming convention and launched a new build before going to lunch. Just came back and I see that the build completed successfully.

I think this proves that the warning is interpreted as a failure causing the build script to fail but only on AppVeyor.

[AdmiringWorm]

@gep13 I am not aware of anything. There is however one thing that could come into play, it is possible that both Warnings and Errors are outputted to stderr which maybe cause AppVeyor to abort the execution if something is written there.

[gep13]

@AdmiringWorm is there a way to confirm what output is being used by GRM for warnings and errors? i.e. are we doing this wrong, and we should alter how these messages are being written out of the application?

[Jericho]

While @AdmiringWorm is researching, does it make sense to change GRM to avoid invoking the Warning method and replace it with calls to either Verbose or Information?

There are 7 calls to Warning that would need to be modified and also 5 unit tests because they assert whether Warning was invoked or not.

I'm volunteering to submit a PR if you think it makes sense.

[AdmiringWorm]

@gep13 we are specifically outputting warnings and higher to stderr. Anything lower will output to stdout.

There are divided opinions on the correct approach, and some believe only errors should be shown on stderr. In contrast, others believe both warnings and errors should be outputted there, as both are considered important information that should be outputted to the terminal even if the normal output is redirected.

I am hesitant to change it, though, but we may want to detect if we are running on appveyor and then output warnings to stdout instead (or if we are running on a CI in general).