Trial: Go back to witness syntax

[odersky]

This is an attempt to validate the current given design against my very first proposal
from more than a year ago: witnesses. I have duplicated the docs, replacing syntax
and terminology.

To get started with browsing: https://github.com/dotty-staging/dotty/blob/try-witness/docs/docs/reference/contextual-witnesses/motivation.md

[jducoeur]

I still think it doesn't read as idiomatic English -- I don't know anybody outside of hardcore computer science who ever uses the word "witness" like that. Can't speak for the rest of the world, but I personally don't think it would teach well to native English speakers...

[odersky]

@jducoeur

https://english.stackexchange.com/questions/167683/witness-to-vs-witness-of/168408
https://forum.wordreference.com/threads/witness-of-to.2600118/
https://www.youtube.com/watch?v=nAvH8I7o8Ek

The noun form is "witness of". So this looks like idiomatic english to me. It might be rarely used in common language, but that's OK as far as I am concerned. We are talking about a concept that will be novel to new users. So it's OK to use an uncommon term and fill it with meaning.

[odersky]

Also: the very first example in Merriam Webster says:

the ruins are a compelling witness of the ancient civilization's cultural and aesthetic achievements

https://www.merriam-webster.com/thesaurus/witness

[jducoeur]

Sure, but that's a fine example of what I mean: that's essentially an archaic form of speech. It would have sounded entirely normal a century ago, and a bit quaint when I was a child, but the language has evolved rapidly since then. I can't remember seeing a sentence like that outside of academia or religion in recent decades, and rarely even there.

I'm using "idiomatic" quite precisely here. It is legal English -- but so is all of Shakespeare. It isn't how people actually use the language in regular practice any more, though. In practice, the non-verb form "witness" is rarely used except when describing a human being, and even that not especially often outside of a courtroom...

[tellurion]

While witness isn't commonly used out of legal settings, it works well with the term evidence, which the compiler currently uses. And it is grammatically correct. I still don't like them because they are used to suggest or talk about something abstractly and not used in a concrete sense.
I do think a term like provide would work better with given.
e.g. The clause provide Ord[Int] { ... } would read as "provide this Ord of Int to compiler" and the clause (given Ord[Int]) would read as "given that an Ord of Int has been provided." I think provide (or a similar, more solid term) and given communicate both intent and what is actually happening better.

[SethTisue]

(at this point I don't have a clear overview in my head of all the different possibilities and their pros and cons, so I can't confidently say what I think is best, big-picture, but regardless:)

I would love it if we can find a noun that isn't "given". I have tried, but I have been unable to get used to talking/thinking about "givens" as a thing. I'm delighted to hear that choosing a different noun is still a possibility (I'd sort of "given" up on it). perhaps "witness" isn't the perfect noun; perhaps there isn't any perfect noun; but IMO it's a better noun than "given"

@jducoeur I don't know anybody outside of hardcore computer science who ever uses the word "witness" like that

the important thing is not to give a clear impression that is also a wrong impression. I think "witness" doesn't give a clear but wrong impression; rather, it gives a "hmm, what's that, tell me more" impression and that's okay:

@odersky it's OK to use an uncommon term and fill it with meaning

agree

[godenji]

/s/witness/instance would work. I agree with @jducoeur that the usage of witness here doesn't read particularly well, but it could work.

The important piece is finding the right terminology to transform the world of Scala 2 implicits such that whatever new keyword(s) are added somehow convey the semantics of use. For example, with typeclasses instance is a natural fit, and given also works quite well as a replacement for implicit parameters, but these two terms are perhaps not enough to capture the range of functionality provided by Scala implicits.

Side note: the multi-soft keyword approach term ... of ... with {} is a bit unpleasant; what's the resistance to adding one or two hard keywords to the language, particularly for such an important area as implicits?

[odersky]

My evaluation so far:

• witness works pretty well for one half of the implicit universe: typeclasses and other typelevel computatoons.
• It's a bit of a stretch for context parameters. Might sound a tad too archaic to call an implicit ExecutionContext a "witness of ExecutionContext". But one could probably get used to it over time, when the novelty wears off.

[bmeesters]

Outside the Scala community, Scala is often regarded as an 'academic' language. This is not a view I agree with, but it is also not something we should ignore. Using keywords like witness is bound to strengthen this view, rather than diminish it.

Personally, I don't care that much about which keyword is chosen and I could get used to witness. However, I am already hooked to Scala, and I think things like witness might be off-putting to potential new Scala developers.

[kavedaa]

I'm not sure if it's been suggested before, but evidence of could be an alternative to witness of.

Pros:

• evidence is less archaic
• metaphorically it's more concrete and "solid"
• if witness were to be combined with the with form, their likeness (wit-) could be confusing

Cons:

• it's a bit awkward to talk about "evidences", in contrast to "witnesses" (better: "evidence instances")

I'd like to add that I think the of form works better than the as form. witness/evidence of Foo makes sense grammatically, but given as Foo doesn't quite.

[odersky]

I decided to pursue #8017 instead.