Raise on non-existing fields in Protobuf.Builder.new/2

[whatyouhide]

cc @keathley as mentioned in #76 and cc @tony612 for review 🙃

[whatyouhide]

Ping @tony612 :)

[tony612]

@whatyouhide Thanks for the PR. I'm thinking if it's better to have a new! function and keep the original one. The reason is sometimes I found it's useful to use a superset map to create a pb struct. And also, in this way we can avoid breaking current code. How do you like it? @keathley

[whatyouhide]

@tony612 we use this library pretty extensively at work and in my experience this is definitely an extremely unexpected behaviour, so much so that I've always considered this a bug. I think it's much more common to misspell a field name and only catch it when encoding than to desire this behaviour. Ultimately, up to you, I can create new! :)

[tony612]

@whatyouhide I guess that's why we have both struct and struct!? Sometimes we specify the keys explicitly

Foo.new(a: 1, b: "foo")

But sometimes we use a map to build it like

map = got_from_somewhere_else()
Foo.new(map)

It's hard to guarantee the map in the later example match very well with our struct, especially in a dynamic language like Elixir.

[whatyouhide]

@tony612 okay, we'll switch completely to new! and forget about new then 😄 I updated the PR, can you take a look?

[tony612]

@whatyouhide Looks like new! should be used in the test?

[whatyouhide]

@tony612 Yep, I actually forgot to push this code but we need to generate the new!/1 function in protobuf modules too 😄 I just pushed the new code and the updated test!

[keathley]

Sorry I didn't respond earlier. I'm glad y'all ended up using new! since it allows new to remain "open" which should allow more re-use of existing maps. 👍