fix: make use_testthat() work for brand-new package

[maelle]

Coming from r-lib/desc#117 (comment)

I understand why this doesn't happen in usethis test where testthat edition field is already set by create_local_package(); after digging a bit I see that prior to 747931b there was as.character(). Now I am at peace, having the whole story 🧩 😌

@krlmlr

[jennybc]

@ateucher will you have a look? I think you were most recently in this headspace. But seems good to me.

[ateucher]

Yes, happy to.

[ateucher]

Thanks @maelle this is great. Along with this change I think it's also worth making proj_desc_field_update() stricter here so this would be handled there rather than in desc (probably using check_string(value)).

Happy to make that change myself if you'd prefer.

[jennybc]

Should proj_desc_field_update() be in charge of coercing to character?

[ateucher]

I did certainly wonder about that but thought it may end up receiving something unexpected that goes awry when trying to coerce, whereas the calling functions would be more suited to formatting the value properly since they know about them. Right now use_testthat_impl() is the only function that calls it with a non-character value.

[jennybc]

OK I'm happy for you decide. So, yes, I think it would be good to check for character input, as you suggest above.

[ateucher]

I'm going to merge this - thanks @maelle!

[jennybc]

I'm not sure this would be worth the fiddliness. But arguably the set logic should be applied before we message. Maybe just worth keeping in mind if we touch this again.