| commit | a61ae333768ef892edecf6a95cf050de5caac7c4 | |
| author | Andrew Kennedy <akenn@fb.com> | |
| Thu, 4 Jul 2019 12:06:43 +0000 (4 05:06 -0700) | ||
| committer | Hhvm Bot <hhvm-bot@users.noreply.github.com> | |
| Thu, 4 Jul 2019 12:10:23 +0000 (4 05:10 -0700) | ||
| tree | 81b54e7c05eb457fe4d53ac731c5edb2a41a3c23 | treesnapshot (tar.gz zip) |
| parent | b44d22033f69c6c770d746c6480669cb823ab254 | commitdiff |
Fix unsoundness bug with arithmetic and unresolved types
Summary:
Arithmetic is broken wrt unresolved-type arguments. For example, Hack thinks that
```
$x ==> $x + 1
```
has type `function(num):int` if the lambda is passed to a generic function.
To fix this, we revamp the type checker code for arithmetic, using `expand_type_and_narrow` to use the lower bound if possible, defaulting to `int`.
The original bug was due to `is_sub_type_LEGACY_DEPRECATED` returning `true` for type variables (e.g. a test of `#1 <: int` would return `true`).
Reviewed By: CatherineGasnier
Differential Revision: D15780423
fbshipit-source-id: d1e68e7cc633c6f3a7f15b3e6f195f04d828c4e7
Summary:
Arithmetic is broken wrt unresolved-type arguments. For example, Hack thinks that
```
$x ==> $x + 1
```
has type `function(num):int` if the lambda is passed to a generic function.
To fix this, we revamp the type checker code for arithmetic, using `expand_type_and_narrow` to use the lower bound if possible, defaulting to `int`.
The original bug was due to `is_sub_type_LEGACY_DEPRECATED` returning `true` for type variables (e.g. a test of `#1 <: int` would return `true`).
Reviewed By: CatherineGasnier
Differential Revision: D15780423
fbshipit-source-id: d1e68e7cc633c6f3a7f15b3e6f195f04d828c4e7
31 files changed: