This CL introduces new Factory/TypeInferenceListener methods to
communicate details about types that are literally named in the source
code.
Previously, when an expression like `x as List<int>` was analyzed, the
resolution information about `List<int>` was communicated to the
analyzer via a kernel DartType, forcing the analyzer to walk the type
representation, applying each of its constituent parts to the
corresponding part of the AST for `List<int>`. This was a lot of work
for the analyzer, it failed to resolve parts of invalid types (such as
`List<int, String>), and it didn't pave the way toward moving to the
Factory API.
The new approach is for information about each type appearing in the
syntax to be communicated to the analyzer at the time the BodyBuilder
converts it to a DartType. So for an invalid type like `List<int,
String>`, the resolution of `String` is communicated to the analyzer
at the time it is converted to a DartType; later, when `String` is
discarded (because `List` only accepts one type parameter), no crucial
information is lost, because the analyzer already knows the resolution
of `String`.
Work still to be done has been captured in issues:
- #33844: References to type parameter elements are not translated
- #33845: Function typed formal parameters are not handled
- #33846: Function types using `Function` syntax are not yet handled
Change-Id: I96dd9b6f3eb573b0b7a46335e8644c59fb78bae7
Reviewed-on: https://dart-review.googlesource.com/64262
Reviewed-by: Konstantin Shcheglov <scheglov@google.com>
Reviewed-by: Brian Wilkerson <brianwilkerson@google.com>
Commit-Queue: Paul Berry <paulberry@google.com>
This gets the tests running again. I'll follow this up with CLs that
reshape the ResolutionStorer API to match the Factory API we're
building, and then after that we can transition the code over to
calling the Factory API directly.
Fixes#33504Fixes#32258
Change-Id: I105841ed5b2ed1198ec5e3e7387f694c61d36ca7
Reviewed-on: https://dart-review.googlesource.com/61101
Commit-Queue: Paul Berry <paulberry@google.com>
Reviewed-by: Brian Wilkerson <brianwilkerson@google.com>
Reviewed-by: Konstantin Shcheglov <scheglov@google.com>
Reviewed-by: Kevin Millikin <kmillikin@google.com>
We have decided to move to a different API for Fasta/Analyzer
integration, one that doesn't involve compiling to Kernel or observing
type inference. The listener API will become broken, so let's just
remove it before that happens.
We lose the ability to trace type inference so let's restore that
ability next. The analyzer's resolution storer is not used, but that
code has been left. The resolution applier will no longer work
without the resolution storer, so those tests have been disabled.
Change-Id: If9ac5aba2d5de7788d8a65f7fa8410599e636d21
Reviewed-on: https://dart-review.googlesource.com/42400
Commit-Queue: Kevin Millikin <kmillikin@google.com>
Reviewed-by: Paul Berry <paulberry@google.com>
Reviewed-by: Peter von der Ahé <ahe@google.com>
This reverts commit 09eed74a8a.
Reason for revert: Too much SDK code is not yet compliant.
Original change's description:
> Make `void` a static warning to use almost everywhere.
>
> Changed the hint to a StaticWarningCode, since that's the new spec'd
> error type and the hint is no longer needed.
>
> Added a new set of methods to test the cases.
>
> Didn't try to solve the problem generally ("all usages except ... are
> errors" means it easier, in theory, to make a ReportVoidExpressions
> style visitor that catches absolutely all types) because most of the
> work is actually about suppressing errors that are no longer needed.
> Ie, from NO_SUCH_METHOD to USAGE_OF_VOID_RESULT which means we have to
> put the void handling logic into each AST method specially anyway.
>
> Some redundant tests removed.
>
> Don't flag: ternaries, void -> void assignments, void returns in
> dynamic.
>
> Change-Id: Ief8035dcfe582b36b6372180ddcf4e453d320d9c
> Reviewed-on: https://dart-review.googlesource.com/37441
> Commit-Queue: Mike Fairhurst <mfairhurst@google.com>
> Reviewed-by: Leaf Petersen <leafp@google.com>
TBR=leafp@google.com,scheglov@google.com,mfairhurst@google.com
Change-Id: I13ee4c6939468d35506779ade637a040833632f4
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://dart-review.googlesource.com/39848
Reviewed-by: Mike Fairhurst <mfairhurst@google.com>
Commit-Queue: Mike Fairhurst <mfairhurst@google.com>
Changed the hint to a StaticWarningCode, since that's the new spec'd
error type and the hint is no longer needed.
Added a new set of methods to test the cases.
Didn't try to solve the problem generally ("all usages except ... are
errors" means it easier, in theory, to make a ReportVoidExpressions
style visitor that catches absolutely all types) because most of the
work is actually about suppressing errors that are no longer needed.
Ie, from NO_SUCH_METHOD to USAGE_OF_VOID_RESULT which means we have to
put the void handling logic into each AST method specially anyway.
Some redundant tests removed.
Don't flag: ternaries, void -> void assignments, void returns in
dynamic.
Change-Id: Ief8035dcfe582b36b6372180ddcf4e453d320d9c
Reviewed-on: https://dart-review.googlesource.com/37441
Commit-Queue: Mike Fairhurst <mfairhurst@google.com>
Reviewed-by: Leaf Petersen <leafp@google.com>
This adds checks for allocations and type values, and provides a compile
time error for type declarations, is-checks, and as-casts.
Change-Id: I3bfed41b0dd45abeb02158f0537102c779511998
Reviewed-on: https://dart-review.googlesource.com/36383
Commit-Queue: Sigmund Cherem <sigmund@google.com>
Reviewed-by: Peter von der Ahé <ahe@google.com>
The analyzer used to emit a warning when a type parameter to a generic
method was used in an "is" check. This warning was there to help
users write code that worked correctly on both Dart 1.0 and 2.0. Now
that 2.0 generic methods are being more broadly supported and used,
this warning is blocking further library work, and is being removed.
Fixes https://github.com/dart-lang/sdk/issues/30530
Bug:
Change-Id: I70395305ad082aee3072b5beeb0b1b7f7883391b
Reviewed-on: https://dart-review.googlesource.com/29821
Reviewed-by: Brian Wilkerson <brianwilkerson@google.com>
This should fix the issue 30917 flakiness.
The reason why the tests were failing is that the code in the tests
was not quite correct Dart code, so Fasta generated a synthetic
`#errors` top-level variable with fileUri == null. We should skip
this field because it is not a part of any unit, because its fileUri
is not the fileUri of any unit.
But we lose fileUri for a Node if Program.uriToSource does not
include this URI. So, if we clear uriToSource in KernelDriver, we
lose all the file URIs. So, now we keep sources for the cycle files.
The reason for flakiness is because of the same race condition
between analyzing AnalysisDriver.addFile() and getResult(). If
we process getResult() before, we don't have to read the library
kernel file, we just create in first time, and fileUri(s) are
valid.
R=ahe@google.com, paulberry@google.com, sigmund@google.com
Bug: https://github.com/dart-lang/sdk/issues/30917
Change-Id: Ie559cf2a8a778ebf4a9e297f5cb13d30a98b5a3c
Reviewed-on: https://dart-review.googlesource.com/9744
Reviewed-by: Paul Berry <paulberry@google.com>
Commit-Queue: Konstantin Shcheglov <scheglov@google.com>
Mike Fairhurst