Several fixes around fresh and Unscoped.#26193
Conversation
odersky
force-pushed
the
change-unscoped-fresh
branch
2 times, most recently
from
36a4c5d to
2412b6a
Compare
|
@bracevac Can you take a look at the scaladoc test failure? It's hard to see what goes on here. |
odersky
changed the title
|
@odersky I fixed the test in question. But tbh, this one was no different than a cc compilation test, so it's not unreasonable to expect someone working on cc to fix them. Certain other tests for scaladoc and REPL I will absolutely insist on keeping them, as they affect user experience. |
odersky
left a comment
odersky
left a comment
There was a problem hiding this comment.
Yes, that's much better! I completely agree we need snippet tests for specific cc code. But we should be 100% dumb for generic functionality such as whether more than one expectation per line is checked.
|
We now see extra errors on already |
|
I ended up with a simpler change that lets |
|
|
||
| val _: () -> One^{fresh} = () => One() // ok | ||
| val _: () => One^{fresh} = () => One() // ok | ||
| val _: () => One^{fresh} = f2 // ok |
There was a problem hiding this comment.
I wonder: does the notion of fresh here ever make sense without an ExclusiveCapability?
Which is to say, if One is already known to be exclusive, why can't One^ just mean fresh?
odersky
force-pushed
the
change-unscoped-fresh
branch
from
5ff528f to
4b737ca
Compare
Previously, anonymous functions were excluded. lambda-fresh-2.scala shows why this is unsound. We now drop the special treatment of anonymous functions, and at the same time exclude LocalCaps that come from outside the function from checking.
Previously, a closure could return an unscoped type with `any` in it (e.g., `Ref^`) but its type would be a pure function type. We now propagate the `any` into the function type. This has to be done with care though, since the `^` of a closure might come from the expected type, in which case we should ignore it. The tricky bit is distinguishing one from the other.
Drop two unused parameters.
Don't elide even if the capset belongs to a Capability class.
This replaces the special treatment of UnScoped in the ToResult map and makes Unscoped capabilities more versatile.
A LocalCap never accounts for a ResultCap, even if the classifier is Unscoped.
We need to do it after we widened an Unscoped fresh to any.
odersky
force-pushed
the
change-unscoped-fresh
branch
from
8a0dd7c to
920375d
Compare
odersky
changed the title
odersky
changed the title
fresh and Unscoped.
|
@bracevac I accidentally deleted your commit to fix the scaladoc when I did a |
Both `// error` and `// warn` will now match one _or more_ error/warning. In user-facing docs, we usually do not want to mark every single error/warning, but just indicate that there should be some.
|
@odersky done |
|
@bracvac Thanks! Ready for review them. |
This is no longer needed since we now ensure monotonocity of closure types.
odersky
force-pushed
the
change-unscoped-fresh
branch
from
ee87a8a to
21e6dca
Compare
odersky
force-pushed
the
change-unscoped-fresh
branch
from
0817851 to
d6f69db
Compare
Co-authored-by: Oliver Bračevac <bracevac@users.noreply.github.com>
3 participants
1. Ensure monotonicity in the types of closures
Previously, a closure could return an unscoped type with
anyin it(e.g.,
Ref^) but its type would be a pure function type. We now propagatethe
anyinto the function type. This has to be done with care though, sincethe
^of a closure might come from the expected type, in which case weshould ignore it. The tricky bit is distinguishing one from the other.
2. Never elide ResultCaps when printing Capability classes.
They always print as
fresh.3. Tighten
Unscopedexception inacceptsLevelOf.Previously, an
UnscopedLocalCap could subsume a ResultCap. This is no longer possible.4. Do existential widening in
testAdapated.If a closure has a
Ref^{fresh}return type of anUnscopedclassRef, and this leads to atype mismatch in
testAdapted, convert toRef^{any}, and try again after ensuring monotonicity ofthe mapped type. This means that for
Unscopedclasses created in closures we have a choicewhether we want to treat tha instance as
freshor as a locally createdany.