Do not create phis for length since it confuses bounds checker and
results in ugly code for for-in Array expansion.
Change-Id: I75160d38a8ad924b5a3f72fa4dd3d9f0d04e88b1
Reviewed-on: https://dart-review.googlesource.com/22303
Reviewed-by: Stephen Adams <sra@google.com>
Commit-Queue: Stephen Adams <sra@google.com>
This involves a few main pieces:
- Add code to the GN scripts to generate DDC's patched SDK and then
compile it to summaries and JS in the build output directory.
- Add support to the underlying DDC build scripts to support controlling
which files are built where.
- Update test.dart to use the DDC SDK from the build directory.
- Update create_sdk to use the built SDK instead of the checked in one.
- Fix various internal DDC tools to build their own copy of the SDK
(since they can't easily find the one in the build directory because
it's path if config-specific) and use those.
- Delete the checked DDC SDK JS and summaries.
I think I got everything working. The built Dart SDK looks fine -- it's
identical to one built using the old build scripts.
The various tools and DDC's little test runner I *think* work, but there
may be a bug or two in there. I tried the various things I could and it
seems like they work but it's hard to tell since they may be kind of
broken right now anyway.
Bug:
Change-Id: Iea77915a5c1cc8450f60ebfbdf8c725c7ea2f32c
Reviewed-on: https://dart-review.googlesource.com/18144
Commit-Queue: Bob Nystrom <rnystrom@google.com>
Reviewed-by: Zach Anderson <zra@google.com>
Reviewed-by: Vijay Menon <vsm@google.com>
We never need to add assignability checks when doing top level
inference; putting the check inside checkAssignability avoids the need
to remember to put it at every call site.
In order to keep type inference consistent between top level and local
type inference, I've modified the handling of assignments so that the
computations performed by checkAssignability do not influence the
inferred types. For example, when performing type inference on `a =
b`, we used to consider the resulting expression to have the type of
`a` if a downcast was required, otherwise the type of `b`. Now we
always consider the type of the assignment to be the type of `b`.
This is consistent with the informal spec of type inference. This
change introduces some mal-typed kernel nodes in error recovery
situations, which I will fix in a future CL--see issue #31417.
Change-Id: I02e8dff54e2184e6f059d1339e51edf42c4cd058
Reviewed-on: https://dart-review.googlesource.com/22260
Reviewed-by: Konstantin Shcheglov <scheglov@google.com>
Commit-Queue: Paul Berry <paulberry@google.com>
At this point KernelDriver is used only by Analyzer, so I don't include
anyone from the FrontEnd team. We will merge to get the same performance
benefits of the latest IKG approach, but later.
Only just enought integration is done - we can store and apply
types in the local variable declaration.
Everything is in disarray, I will follow up with improvements.
R=brianwilkerson@google.com, paulberry@google.com
Bug:
Change-Id: Ie3434e26233b84fb019c50a7cc834a08d0dadbed
Reviewed-on: https://dart-review.googlesource.com/22202
Commit-Queue: Konstantin Shcheglov <scheglov@google.com>
Reviewed-by: Paul Berry <paulberry@google.com>
Reviewed-by: Brian Wilkerson <brianwilkerson@google.com>
This brings IKG recompilation time from 20-30 ms range to 5-10 ms range,
i.e. 3-4 times faster.
Of course performance of the system in general starts depending now
on how how many libraries we need to send to VM, and how fast the client
of IKG can serialize the returned libraries.
1. Leaf library.
Compilation: 10 ms
Returned libraries: 8
Serialization time: 1 ms.
2. A deep library.
Compilation: 10 ms
Returned libraries: 222
Serialization time: 60 ms.
R=paulberry@google.com, sigmund@google.com
Bug:
Change-Id: I5eab99c0bd7916d605696696d5e570c5fa6f1978
Reviewed-on: https://dart-review.googlesource.com/21900
Commit-Queue: Konstantin Shcheglov <scheglov@google.com>
Reviewed-by: Sigmund Cherem <sigmund@google.com>
This just makes the name more appropriate since now that the default incremental
generator is not using kernel-driver altogether.
Change-Id: Ic7357dbb4eaad482f20249dd13ba3a49a39322d3
Reviewed-on: https://dart-review.googlesource.com/22160
Commit-Queue: Sigmund Cherem <sigmund@google.com>
Reviewed-by: Jonas Termansen <sortie@google.com>
The changes in this CL include:
* handle interface targets also on this dispatch (which happens due
to mixins)
* preserve class members if interface targets refer to them
* remove references to shaken elements in `Library.additionalExports`
* support handling of typedefs
Change-Id: Id8df6fbf5c3428ae9c9f8b11ee9bf80c35a7a091
Reviewed-on: https://dart-review.googlesource.com/21640
Reviewed-by: Jens Johansen <jensj@google.com>
Commit-Queue: Martin Kustermann <kustermann@google.com>
Running a test via pkg/testing where the command line starts with "--"
would throw a RangeError (it's asking for a range from 0 to -1).
Having the "--" somewhere in the middle of the arguments list would not
throw, but instead "eat" an argument.
Bug:
Change-Id: Ie13052a378746c17b343dff7704002a6929088bf
Reviewed-on: https://dart-review.googlesource.com/21920
Reviewed-by: Peter von der Ahé <ahe@google.com>
Commit-Queue: Jens Johansen <jensj@google.com>
This CL adds a cleanup method to pkg/testing.
This logically allows one to have a "finally" step in a chain to perform
cleanup even upon step failure.
Bug:
Change-Id: I934e97354d733df32f186ebfe370d0307515bebb
Reviewed-on: https://dart-review.googlesource.com/21980
Commit-Queue: Jens Johansen <jensj@google.com>
Reviewed-by: Peter von der Ahé <ahe@google.com>
Eliminate uses of the comment syntax in lib/async. This is mostly
done simply by turning explicit (but commented) casts into implicit
casts to avoid regressing dart2js performance.
Bug:
Change-Id: I286c7bffbf609b5e2fa24b9d48b90876c2c45a9b
Reviewed-on: https://dart-review.googlesource.com/21566
Reviewed-by: Paul Berry <paulberry@google.com>
Reviewed-by: Stephen Adams <sra@google.com>
* kernel-service.dart is moved to pkg/vm/bin and renamed to
kernel_service.dart
* LICENSE, pubspec.yaml and analysis_options.yaml added to pkg/vm.
* Kernel isolate name is corrected to be independent of URI passed
in --dfe VM option.
Change-Id: I5b7b860297314e240e557af80913e7ac3e6324ad
Reviewed-on: https://dart-review.googlesource.com/21260
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Sigmund Cherem <sigmund@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Previously, a ShadowSyntheticExpression was its own kind of Expression
unrelated to any of the Expression types defined in the kernel. It
responded to visit requests by forwarding the visit request to the
desugared expression. This had the advantage that for serialization
purposes, ShadowSyntheticExpressions would automatically disappear.
But it had the disadvantage that if a ShadowSyntheticExpression ever
needed to be replaced with another node (e.g. because an implicit
downcast was needed), the _ChildReplacer would get confused by the
visitor forwarding and fail to replace the node properly.
This CL fixes the problem by changing ShadowSyntheticExpression to a
Let node of the form `let _ = null in desugaredExpression`. Since
this is a genuine kernel expression, it no longer needs special
visiting semantics, and node replacement works properly.
In most circumstances, we replace the ShadowSyntheticExpression with
its desugared equivalent during type inference, so the final
serialized kernel representation is unchanged. However, in a few rare
circumstances involving error recovery, a ShadowSyntheticExpression
remains in the tree. Semantically this shouldn't be a problem, since
the behavior of `let _ = null in x` is the same as the behavior of
`x`. Nonetheless, I hope to get rid of these rare circumstances in
follow up CLs.
Change-Id: Iacf8c0028c424ac9ef98ef8302680e289bf7a017
Reviewed-on: https://dart-review.googlesource.com/21571
Reviewed-by: Konstantin Shcheglov <scheglov@google.com>
Commit-Queue: Paul Berry <paulberry@google.com>
Currently we produce 'partial dills' for the VM when compiling through
kernel, i.e. we only include the user-code and whatever canonical names
the user refers to. That's perfectly fine for the VM as it always loads
the platform.dill file too.
We cannot currently print it though.
(Technically we should be able to concatenate it with platform.dill and
print that).
This CL allows us to print these partial dills without concatenating
anything first.
Bug:
Change-Id: I7d5f12332e16623d02531ad46e2dc103222f59e9
Reviewed-on: https://dart-review.googlesource.com/19002
Commit-Queue: Jens Johansen <jensj@google.com>
Reviewed-by: Dmitry Stefantsov <dmitryas@google.com>
Johnni Winther