This shares a cache for mapping Elements to their public LibraryElements across each snippet producer, and also skips snippet producers that produce snippets that won't match any typed prefix.
Change-Id: I6b64b3c55f1030a5eaa7ca1afdcd6c416e4baa08
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/273962
Commit-Queue: Brian Wilkerson <brianwilkerson@google.com>
Reviewed-by: Brian Wilkerson <brianwilkerson@google.com>
When asking for all references of a member it starts by finding all
supertypes for the class, then for each supertype (including itself)
(that has a member with the same name as the one we're querying for)
it searches for all subtypes.
We recall that when getting all subtypes, it really gets all direct
subtypes, then for each of those, their direct subtypes etc.
Naturally, then, with getting all subtypes for all supertypes we
will often ask for all direct subtypes of the same class several times.
In fact, it's O(n^2):
As an example, if asking for references to `foo` on this
(in a file in analyzer, with 8 workspaces):
```
class X0 { void foo() { print('hello'); } }
class X1 extends X0 { void foo() { print('hello'); } }
class X2 extends X1 { void foo() { print('hello'); } }
[...]
class X149 extends X148 { void foo() { print('hello'); } }
```
we ask for subtypes 90600 (150 classes * 151 / 2 * 8 workspaces) times.
This CL stops that from happening and in the example from above we
"only" asks for subtypes 1200 (150 classes * 8 workspaces) times.
For the `newFile` from `AbstractSingleUnitTest` example, for instance,
we go from 28,264 asks to 17,432 asks (a ~38% reduction).
Non-first runtimes when asking for references:
Before:
0:00:03.074853
0:00:03.021881
0:00:03.034707
0:00:03.115596
0:00:03.032574
After:
0:00:02.223978
0:00:02.149937
0:00:02.150236
0:00:02.104704
0:00:02.175859
Difference at 95.0% confidence
-0.894979 +/- 0.060283
-29.2867% +/- 1.97266%
(Student's t, pooled s = 0.0413338)
Change-Id: Id792a595e74de01c7186ab1263c38728f051f603
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/272623
Commit-Queue: Jens Johansen <jensj@google.com>
Reviewed-by: Konstantin Shcheglov <scheglov@google.com>
Reviewed-by: Brian Wilkerson <brianwilkerson@google.com>
TL;DR:
56c416430e vs this CL:
textDocument/implementation (non-first runs):
```
Difference at 95.0% confidence
-3.6465 +/- 0.307413
-82.3689% +/- 6.94398%
(Student's t, pooled s = 0.238983)
```
I.e. from ~4.4s to ~0.78s on non-first runs.
textDocument/references (non-first runs):
```
Difference at 95.0% confidence
-33.1856 +/- 0.789512
-91.4283% +/- 2.17515%
(Student's t, pooled s = 0.613768)
```
I.e. from ~36.2s to ~3.11s on non-first runs.
Raw data below.
Details:
When having multiple workspace folders open, sometimes the drivers
representing these folders will have overlapping files.
For instance in my "analyzer" workspace there are a total of ~7000
unique files but if just adding each drives known files there’s
40,000+ files.
When asking for all direct subtypes of a class the a
`_FilesReferencingNameTask` does a lookup in a set for every known file
of that drive, meaning that in practice, often the same set is queried
for the same thing many times.
This CL instead assigns each `Drive` the files it was to look at,
each set will therefore only be querues once per type.
This takes the number of operations (per type) from `N*M` where `N` and
`M` are number of drives and number of files respectively (as each drive
could know all the same files) to just `M`, i.e. number of files.
This CL takes the runtime for asking for implementations of 'ToJsonable'
from package:analysis_server/lsp_protocol/protocol_special.dart in my
analyzer workspace (with 8 workspace folders / drivers) from ~4.4 seconds
to ~0.8 seconds (see below for details).
It also takes a call for "find all references" ("textDocument/references")
on method "newFile" in class "AbstractSingleUnitTest" in
"analysis_server/test/abstract_single_unit.dart" from ~36 seconds
to ~3.2 seconds (see below for details).
Raw data:
HEAD as of 56c416430e
textDocument/implementation
===========================
(notice the first query taking a long time)
(restart analyzer process)
Got answer to query in 0:00:28.921389
Got answer to query in 0:00:04.270255
Got answer to query in 0:00:04.248489
(restart analyzer process)
Got answer to query in 0:00:08.319562
Got answer to query in 0:00:04.232582
Got answer to query in 0:00:04.102793
(restart analyzer process)
Got answer to query in 0:00:09.186371
Got answer to query in 0:00:04.867653
Got answer to query in 0:00:04.840468
textDocument/references
=======================
(notice the first query taking slightly longer)
(restart analyzer process)
Got answer to query in 0:00:41.916106
Got answer to query in 0:00:36.619501
Got answer to query in 0:00:36.502156
(restart analyzer process)
Got answer to query in 0:00:40.353819
Got answer to query in 0:00:37.709369
Got answer to query in 0:00:36.086530
(restart analyzer process)
Got answer to query in 0:00:40.052997
Got answer to query in 0:00:35.273826
Got answer to query in 0:00:35.589722
WITH CL
textDocument/implementation
===========================
(notice the first query taking a long time)
(restart analyzer process)
Got answer to query in 0:00:11.902654
Got answer to query in 0:00:00.742324
Got answer to query in 0:00:00.730748
(restart analyzer process)
Got answer to query in 0:00:04.966440
Got answer to query in 0:00:00.795087
Got answer to query in 0:00:00.820526
(restart analyzer process)
Got answer to query in 0:00:04.838173
Got answer to query in 0:00:00.810417
Got answer to query in 0:00:00.784124
textDocument/references
=======================
(notice the first query taking a long time)
(restart analyzer process)
Got answer to query in 0:00:07.632822
Got answer to query in 0:00:03.098794
Got answer to query in 0:00:03.008570
(restart analyzer process)
Got answer to query in 0:00:07.690932
Got answer to query in 0:00:03.062484
Got answer to query in 0:00:03.088793
(restart analyzer process)
Got answer to query in 0:00:07.786218
Got answer to query in 0:00:03.160477
Got answer to query in 0:00:03.248359
Change-Id: I18a3cbd870f62e87937f92942460baccf8bce088
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/272560
Reviewed-by: Brian Wilkerson <brianwilkerson@google.com>
Commit-Queue: Jens Johansen <jensj@google.com>
The edit regions should not overlap and the fixes are independent.
- Add `canBeAppliedInBulk` and `canBeAppliedToFile`.
- Add a BulkFixProcessorTest with multiple fixes in the same file.
- Add a FixKind for fixing implicit call tearofffs across the file.
Change-Id: I14260e4082fcd47c3a50514416b1c15b2c7e12c2
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/271540
Reviewed-by: Oleh Prypin <oprypin@google.com>
Reviewed-by: Konstantin Shcheglov <scheglov@google.com>
Commit-Queue: Nate Bosch <nbosch@google.com>
Asking for implementations of 'ToJsonable' from
package:analysis_server/lsp_protocol/protocol_special.dart takes
8+ seconds both on first request and on subsequent requests.
Turns out most of that is spent on getting line info data for each
result via `server.getLineInfo`.
This CL rewrites how the search is performed, getting all subtypes
directly via `searchAllSubtypes`, from which one can cheaply get the
line info.
On my machine, the call that always took 8+ seconds before now takes
<2 seconds (at least on subsequent runs).
Notice that the time also depends on how much is open by the editor and
that calls from my editor with many folders open still takes ~6 seconds.
This CL furthermore adds more performance measurements that's available
via the "Analysis Server Diagnostics" page "Timing" for
"textDocument/implementation" requests.
Change-Id: I2e5ef20238901fb4734cc182aeb53adfc96f79b6
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/272360
Commit-Queue: Jens Johansen <jensj@google.com>
Reviewed-by: Konstantin Shcheglov <scheglov@google.com>
Ahmed Ashour