Fix handling of special implicit searches for ClassTag, quoted.Type, Eq #4436
Conversation
|
Based on #4435 |
They did not work if the searched-for type was an uninstantiated TypeVar that had the class as an upper bound. Getting them to work is surprisingly subtle.
odersky
force-pushed
the
fix-specialized-implicits
branch
from
ddbd04e to
4d95963
Compare
Roll the functionality of contextualBaseType into baseType itself.
|
test performance please |
|
performance test scheduled: 1 job(s) in queue, 1 running. |
|
Performance test finished successfully: Visit http://dotty-bench.epfl.ch/4436/ to see the changes. Benchmarks is based on merging with master (dfab5e5) |
|
test performance please |
|
performance test scheduled: 1 job(s) in queue, 1 running. |
The previous logic tended to test things several times, which made it harder to understand and possibly less efficient. We now combine the cachability decisions in the main `recur` method of `baseTypeOf`.
|
test performance please |
|
performance test scheduled: 1 job(s) in queue, 1 running. |
When testing on typer/*.scala, the differences are as follows:
Cache all Only cache successful
Total basetypes 386K 386K
Full computations in recur 85K 397K
Cache entries created 72K 46K
|
Performance test finished successfully: Visit http://dotty-bench.epfl.ch/4436/ to see the changes. Benchmarks is based on merging with master (d0f7846) |
|
test performance please |
|
performance test scheduled: 1 job(s) in queue, 1 running. |
|
Performance test finished successfully: Visit http://dotty-bench.epfl.ch/4436/ to see the changes. Benchmarks is based on merging with master (d0f7846) |
|
So this ended up rewriting |
|
@smarter, can we get this in? It would make the Tasty extractors somewhat nicer to write. |
|
Sure, I'll review this today. |
| val base = formalValue.baseType(cls) | ||
| if (base <:< formalValue) { | ||
| // With the subtype test we enforce that the searched type `formalValue` is of the right form | ||
| handler(base).orElse(ifNot) |
There was a problem hiding this comment.
In synthesizedClassTag there's a call to fullyDefinedType, I wonder if we should move this around here to do this consistently for all handlers.
|
@nicolasstucki Enjoy ClassTags! |
They did not work if the searched-for type was an uninstantiated TypeVar that
had the class as an upper bound. Getting them to work is surprisingly subtle.