Fix Terraform trying to set default_branch to null

[abitrolly]

Fixes #299 where Terraform tries to reset the default_branch to null for non-empty repositories if not set in .tf

The default_branch defaults to master on GitLab side, but it is only set if the repository is not empty. Otherwise it is null on remote side.

---

The debug in 24d43d8 showed no trace of null value from remote.

2020/06/19 20:55:55 [DEBUG] gitlab.Project default_branch (string) ""
2020/06/19 20:55:55 [DEBUG] d.Get default_branch (string) ""
2020/06/19 20:55:55 [DEBUG] d.GetOk default_branch (string) "" - false
2020/06/19 20:55:55 [DEBUG] d.HasChange default_branch false
2020/06/19 20:55:55 [DEBUG] d.GetChange default_branch (string) "", (string) ""

But the actual response shows that null was returned in JSON.

2020/06/19 20:55:55 [DEBUG] GitLab API Response Details:
...
 "default_branch": null,

---

The fix is to skip the change if TF is trying to set null value for remote side. This PR also adds a warning if default_branch is specified in .tf file, but the remote repo is empty. And also removes more specific check for initilize_with_readme towards the generic case above. default_branch should not make assumptions on how the repo was created.

[abitrolly]

After understanding what SchemaDiffSuppressFunc really does hashicorp/terraform-plugin-sdk#479 it looks like it is possible to suppress the plan if new planned value is empty. What would ignore all changes to default_branch until it is set in Terraform config.

[abitrolly]

Setting default branch on an empty repo results in an error.

Error: PUT https://gitlab.com/api/v4/projects/19526446: 400 {message: {base: [Could not change HEAD: branch 'sdfdf' does not exist]}}

[armsnyder]

It would be nice to see a regression test for this, since it's difficult to tell just by looking at the change how it will behave.

[horjulf]

This looks simple enough, can we get this merged please ?
Currently the provider breaks in all cases where you don't want to track or enforce a default branch in Terraform code.

@nicholasklick

[abitrolly]

With the ongoing governmental terror in Belarus, I am emotionally drained, and can not think clearly to write the tests, sorry.

[roidelapluie]

@armsnyder WDYT about this?

@abitrolly no worries. take care!!!!

[armsnyder]

@roidelapluie The PR looks like an improvement, so I'm fine with merging. I think it could be improved further at a later time by using the Default or Computed fields of the attribute, and I was also hoping for a test. But on visual inspection it looks ok.

[nicholasklick]

Merging due to customer request and Maintainer approval

[roidelapluie]

Merging due to customer request and Maintainer approval

Thanks to* I hope :)