This CL:
* Makes the standard filesystem more synchronous by default.
This makes reading files for normal compiles faster.
* Adds a more asynchronous version of `exists` and `readAsBytes`.
This potentially allows for reading files faster when in a context
where the files can actually be read in parallel. The client have
to make the choise though.
* Skips `Uri.base.resolveUri` when the uri has a scheme. This makes it
slightly faster in my tests, but does not remove any `.` and `..`.
Having those when having a scheme seems sort of weird though.
If this turns out to be a problem we can add it back.
* Re-orders what the standard filesystem checks in `exists` to assume files
which seems more likely to happen. This - for files - means less checks
and should make checking for existance faster. (This was noticed by Siggi).
Using the same measurement procedure as in
https://dart-review.googlesource.com/c/sdk/+/85442
timing how long fasta spends on *reading* files when compiling dart2js
(when compiling via the VM) I get a good speedup.
Note that the measurements are in microseconds:
Before this CL (run like `out/ReleaseX64/dart pkg/compiler/bin/dart2js.dart 2>&1 | grep "Read file (total)" | tail -n 1` 10 times):
Read file (total): 94678
Read file (total): 94232
Read file (total): 91726
Read file (total): 77472
Read file (total): 95582
Read file (total): 88113
Read file (total): 89245
Read file (total): 91575
Read file (total): 96291
Read file (total): 95730
With this CL (run like `out/ReleaseX64/dart pkg/compiler/bin/dart2js.dart 2>&1 | grep "Read file (total)" | tail -n 1` 10 times):
Read file (total): 68379
Read file (total): 69320
Read file (total): 72930
Read file (total): 69692
Read file (total): 68685
Read file (total): 73548
Read file (total): 64649
Read file (total): 71951
Read file (total): 73486
Read file (total): 70621
Difference at 95.0% confidence
-21138.3 +/- 4193
-23.111% +/- 4.5843%
(Student's t, pooled s = 4462.56)
Furthermore, using an internal benchmark (discussed in an email thread)
I get these numbers (note that the measurements are in milliseconds):
before:
4453
4249
4187
4190
4216
after:
2310
2379
2449
2467
2407
Difference at 95.0% confidence
-1856.6 +/- 131.443
-43.5924% +/- 3.08625%
(Student's t, pooled s = 90.1257)
Using the asynchronous versions of exists and readAsBytes and checking for
existance and reading in parallel on the same internal benchmark it gets to
1137
1185
1067
1089
1189
For completion, comparing to using "raw" `File` (again using the internal benchmark):
sequential standard
2377
2419
2485
2373
2431
sequential io_sync
2279
2182
2237
2312
2324
io_sync faster:
-150.2 +/- 76.3122
-6.21432% +/- 3.15731%
(Student's t, pooled s = 52.3245)
sequential standardasync
4865
4915
4753
5185
4760
sequential io_async
4746
4834
4989
4840
4891
No difference proven at 95.0% confidence
parallel standard
2526
2563
2652
2616
2685
parallel io_sync
2335
2184
2244
2341
2401
io_async faster
Difference at 95.0% confidence
-307.4 +/- 111.037
-11.785% +/- 4.25691%
(Student's t, pooled s = 76.1341)
parallel standardasync
1137
1185
1067
1089
1189
parallel io_async
1105
1088
1145
1048
1113
No difference proven at 95.0% confidence
TEST=Existing tests.
Change-Id: I8ba56ab0768df8672bcdb693782d3f1eec86b683
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/185101
Commit-Queue: Jens Johansen <jensj@google.com>
Reviewed-by: Johnni Winther <johnniwinther@google.com>
Reviewed-by: Gary Roumanis <grouma@google.com>
For instance one could pass something like
```
--wantErrorOnReload="Error: The getter 'axis' isn't defined for the class 'Constraints'" --noTryToDeleteEmptyFilesUpFront
```
to reproduce https://dart-review.googlesource.com/c/sdk/+/180341
(the --noTryToDeleteEmptyFilesUpFront part might not be neccessary,
but was at one point doing the reproduction / development of this).
Change-Id: I943b715c56f3e1103c6287ed9529aa29c0e74f07
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/180360
Reviewed-by: Johnni Winther <johnniwinther@google.com>
Commit-Queue: Jens Johansen <jensj@google.com>
A factory is saved as a procedure in kernel, and was thus also indexed
as a procedure for the purpose of experimental invalidation.
This caused trouble when having a "real" procedure and a factory with
the same name.
That is fixed by this CL by indexing factories ad constructors.
Fixes https://github.com/flutter/flutter/issues/74180
Change-Id: Ieef0421c29a61e26f2a7ee4bb4f196672afd8445
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/180148
Commit-Queue: Jens Johansen <jensj@google.com>
Reviewed-by: Johnni Winther <johnniwinther@google.com>
Before this CL, the map we use to "reuse" references when doing
experimental invalidation mapped from String to Reference.
In bug #44523 two abstract member-signatures have the same textual
(String) name meaning that via the lookup one would get the correct
one, one would get the wrong one, they would get the same one
which eventually causes a crash when trying to serialize.
This CL fixes the issue by mapping via the Name instead, which
basically - for private names - wraps the name and the library
which disambiguates it.
This CL also includes the reproduction of #44523.
Fixes#44523.
TEST=Mostly relying on existing test coverage.
Change-Id: Ib62ebca0b7f5092f0b8410d3c458663c3032eca1
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/177704
Commit-Queue: Jens Johansen <jensj@google.com>
Reviewed-by: Johnni Winther <johnniwinther@google.com>
Johnni Winther