fix fieldset.New bug when prefix slice has len < cap

[sivukhin]

Accidentally inspected gqlgen source code and found a bug in FieldSet.New implementation which can be reproduced if prefix slice is "not full" (length < capacity). In this case internal function parseUnnestedKeyFieldSet will overwrite same slice element multiple times - which will lead to incorrect FieldSet construction.

Bug fix is pretty simple: we should set max capacity equal to length in parseUnnestedKeyFieldSet (using extended slicing operation x[start:end:cap])

In order to cover this bug 1 test case were added: TestWithPrefix/prefix with len<capacity

From the brief source code search it looks like no code were exposed to this bug because prefix argument always passed as nil. But, as this is public library interface, someone can use this logic.

If it's possible - than removal of this argument can be considered (with potential breaking changes).

I have:

• Added tests covering the bug / feature (see testing)

[coveralls]

coverage: 79.357% (+3.5%) from 75.856%
when pulling 70d31c2 on sivukhin:fix-fieldset-bug
into cb3c1c8 on 99designs:master.

[StevenACoffman]

Thanks! There are a lot of things in the ecosystem that use gqlgen as a library: https://github.com/99designs/gqlgen-contrib so these kind of fixes are always welcome!

Yeah, this argument is mostly a leftover that we didn't want to break compatibility by removing. It's probably worth moving to an alternative and deprecating this for a few versions.