Add new record completion operator #1375
Conversation
... as standardized in dhall-lang/dhall-lang#767
|
Note that this requires a few other changes to be merged so that this can be updated to use the corresponding |
| @@ -689,6 +689,8 @@ eval !env t0 = | |||
| vCombineTypes (eval env t) (eval env u) | |||
| Prefer t u -> | |||
| vPrefer env (eval env t) (eval env u) | |||
| RecordCompletion t u -> | |||
| eval env (Annot (Prefer (Field t "default") u) (Field t "Type")) | |||
There was a problem hiding this comment.
Given that we know that eval ignores the type annotation, can we just skip adding it here?
| eval env (Annot (Prefer (Field t "default") u) (Field t "Type")) | |
| eval env (Prefer (Field t "default") u) |
There was a problem hiding this comment.
In this case I prefer to pedantically follow the standard given that we don't have any benchmarks showing that this is a measurable optimization
| @@ -178,7 +178,10 @@ instance Arbitrary Directory where | |||
| instance (Arbitrary s, Arbitrary a) => Arbitrary (Expr s a) where | |||
| arbitrary = | |||
| Test.QuickCheck.suchThat | |||
| (Generic.Random.genericArbitraryRecG customGens weights) | |||
| (Generic.Random.withBaseCase | |||
There was a problem hiding this comment.
I believe this is good, but I don't actually quite understand why. Can you help me?
There was a problem hiding this comment.
This fixes an infinite loop. Adding one more constructor caused the weights to generate an infinitely growing expression tree. Using withBaseCase ensures that the random generator can still cap the recursion depth when this happens so that we don't have to be so diligent with weights.
There was a problem hiding this comment.
Oh! This sounds like a bug in generic-random. Should we report this?
There was a problem hiding this comment.
I'm not sure it is a bug. The documentation from withBaseCase seems to suggest that this behavior is expected if you don't use withBaseCase
Gabriella439
changed the title
... as standardized in dhall-lang/dhall-lang#767