chore: add appveyor ci configuration

[Tapppi]

I added ci.appveyor.com to this project. Not much changed since #57, removed the core.autocrlf = true config. Is there any need for it really?

[coveralls]

Coverage remained the same at 95.402% when pulling 51444d0 on Tapppi:appveyor_config into 8369375 on conventional-changelog:master.

[coveralls]

Coverage remained the same at 95.402% when pulling ef74bdc on Tapppi:appveyor_config into 8369375 on conventional-changelog:master.

[bcoe]

@Tapppi landed this so that we'd kick off an AppVeyor build, looks like we might need to do some work:

https://ci.appveyor.com/project/bcoe/standard-version/build/job/dbrpfj3f570n8h1c

Perhaps git isn't available?

[Tapppi]

Yeah I just noticed that, I'm debugging on my fork, will report shortly.

[Tapppi]

Also, the config is missing the final newline 😢 Hurts my soul that I made that mistake.

[Tapppi]

So the build finished, with all the logging. Theres a bunch of git warnings about author info not being set, but I guess that doesn't matter much, except for the fact that it's annoying when debugging. The actual error seems to be Error in options.transform: Invalid date, happening after √ created CHANGELOG.md.

https://ci.appveyor.com/project/Tapppi/standard-version/build/10/job/ksaf6xchqvhavmwn

After debugging I have found that:
Doesn't seem to be related to node v6. Cannot reproduce locally.
Error is thrown in at https://github.com/conventional-changelog/conventional-changelog-core/blob/master/index.js#L45. Caused by the transform function parsing the gitRawCommit outputs committerDate and throwing an error there.

[Tapppi]

I have a nagging feeling this could be related to the git log --format command used by gitRawCommit including % signs in the format argument and this % sign thing in appveyor appveyor/ci#246 somehow.. 😞 Still not sure though. I'll look into it more when I can.

[bcoe]

@Tapppi tested on Node V6 on OSX, works BTW -- wish AppVeyor had better output. We should look at the npm/npm appveyor config, I think there are some tricks to get better output.

[stevemao]

@bcoe Can you add a badge to README? appveyor config is per user instead of per org... I can point the CI under my name to this repo... Which is weird ..

[bcoe]

@stevemao 👍 will add as soon as we hammer out this weird issue... yeah, appveyor doesn't do orgs too well.

[Tapppi]

So I put a debug build together. It basically has a node_modules.tar.gz that has all the deps with debug logs inserted that's unarchived instead of node install. As you can see, the git log output has a value of truen for committerDate (git log format specifier %ci) for some reason (the actual line is 344):
https://ci.appveyor.com/project/Tapppi/standard-version/build/19/job/v0a2h31kk5b97e73#L337

I have no idea how that can be 😞

[Tapppi]

Here is another test from earlier where I tested git log --format (executed in powershell) before running the tests. It's run with the same format string that's used in tests. The output is just fine..
https://ci.appveyor.com/project/Tapppi/standard-version/build/15/job/hc2a86i2va2mjsxa#L881

[bcoe]

@Tapppi I have quite a few cycles to put into OSS this weekend, so I can dive in and help debug too; if this is still stumping us.

[Tapppi]

I forgot to check the last test i put together and, lo and behold, it got the most interesting results so far. For the heck of it I put another debug together, where the git log --format string is always just logging all the dates (%ci %cI %ai %aI).

The committer dates (%ci %cI), which are used, are always truen even though the author dates are correct:

https://ci.appveyor.com/project/Tapppi/standard-version/build/20/job/piotnr784candb99#L592

As far as I can see, this could be either:

• git's intended behavior manifesting due to some strange way the tests are run
• a bug in the specific version of git in Appveyor

I can't find any documentation pointing out that truen could be a value for %ci though..