fix go vet error with go1.11

[dmitris]

currently builds of the package with go1.11 fail due to using %q qualifier with *string type in github/repos_contents_test.go:

$ go1.11beta2 tool vet .
github/repos_contents_test.go:59: T.Errorf format %q has arg tt.encoding of wrong type *string
github/repos_contents_test.go:62: T.Errorf format %q has arg tt.encoding of wrong type *string

In Travis, a number of PR builds fail with Go:master ("Allowed failures"), ex. https://travis-ci.org/google/go-github/builds/406237120

PR fixes the issue by treating the cases where tt.enconding and tt.content are nil or not nil separately and adding extra varaibles (encoding and content) for printing in error messages. It also adds a workaround for gofmt difference between go1.10 and go1.11 (as in golang/go#26228, Go commit: golang/go@542ea5a) by adding an empty line.

[gmlewis]

Since this affects only a test, why not just change the %qs to %vs?

[dmitris]

%v does not show the string, only the pointer address - I thought it would be useful to see the actual string value (if not nil) in the error message. Let me know if you prefer to change it to %v, I can change it.

[gmlewis]

Oh, right... doh. It seems like we could come up with a solution that has less repetition... like taking advantage of the same logic in the github-accessors.go file... for example:

...
got, err := r.GetContent()

stringOrNil := func(s *string) string {
  if s == nil {
    return "<nil>"
  }
  return *s
}
encoding := stringOrNil(tt.encoding)
content := stringOrNil(tt.content)
...

What do you think?

[dmitshur]

We'll need to think about a better long term way to deal with gofmt changes.

I think the best future-proof solution is to have the diff -u <(echo -n) <(gofmt -d -s .) check run only for one of the Go versions, not all of them. It might not be possible to make a single .go file compatible with gofmt of many Go versions (or possible only by sacrificing readability).

[dmitris]

@gmlewis - changed to use stringOrNil as you suggested

[dmitshur]

I don't think it's appropriate to add this much code in between got, err := ... and the if err != nil ... statements while keeping the variable name just err (see here). A future reader may think there's a missing error check after the got, err := ... line.

I would suggest factoring out the definition of stringOrNil to the very top of TestRepositoryContent_GetContent, or even outside to the package scope, below TestRepositoryContent_GetContent. And the stringOrNil(...) calls can just be inlined into the corresponding t.Errorf calls.

[dmitris]

agree, done with 2c7b329

[gmlewis]

Thank you, @dmitris and @dmitshur!
I like this a lot better.
One tiny nit, one random musing (feel free to ignore), and then LGTM.
After that, awaiting second LGTM before merging.

[gmlewis]

random musing - what about making this line return fmt.Sprintf("%q", *s) to preserve the quoting?

[dmitris]

if we want the function call to have quotes around the parameters as in a "real" one, I think we should add them to the t.Errorf's format, and change the surrounding quotes markers:

		if err == nil && tt.wantErr {
			t.Errorf(`RepositoryContent("%s", "%s") did not return unexpected error`,
				stringOrNil(tt.encoding), stringOrNil(tt.content))
		}

- this would allow to keep stringOrNil doing straightforward conversion to string or "<nil>" without adding additional formatting (surrounding quotes).

[gmlewis]

No, this is not correct because <nil> would then be quoted like "<nil>" which is not right.
It should either be <nil> or "string-value".
But no big deal... we can ignore it.

[dmitris]

you are right; changed stringOrNil to use fmt.Sprintf("%q", *s) for quoting

[gmlewis]

Please add period to end of sentence.

[dmitris]

done

[dmitris]

@gmlewis - could you merge this, or do you need the second approval? Thanks.

[gmlewis]

@dmitris - Sorry for the delay. I think this is good-to-go and doesn't need a second LGTM now.
Merging.