Reference gotcha with deployment featuring changes to .gitignore

[dougjq]

We recently had a bad deploy and this comment is to remind us and others about the complexity of WordPress plugins in roots and interactions with .gitignore. If the plugin is composer managed, it's best to specify that in .gitignore because basically composer in roots wants to install that in web/app/plugins, which is source controlled. We don't want to see that plugin in our local working copies. But then we decided to move the advanced custom fields plugin from composer, where because of recent events it was not working like before, into directly managed source. And what we found recently was the first trellis deploy didn't come with the advanced custom fields source. We don't know why; maybe it's a trellis bug. Even if so, I don't think it would need to be fixed. Just tell people to deploy twice, problem solved. It repeatably failed for us the first time and worked for us the second deploy.

[dalepgrant]

@dougjq we've found in the past that splitting out the commits seems to help. For example, we've more often gone the other way: added a plugin via composer that was previously committed. Removing the files in one commit, editing gitignore in another and finally re-adding in composer in a third commit seems to have stopped any funny business for us. That may help avoid having to deploy twice.