New plan to [fix the `sbt` build](https://www.pivotaltracker.com/n/projects/2539304/stories/182209126) and its annoying:
```
log.error(
"Truffle Instrumentation is not up to date, " +
"which will lead to runtime errors\n" +
"Fixes have been applied to ensure consistent Instrumentation state, " +
"but compilation has to be triggered again.\n" +
"Please re-run the previous command.\n" +
"(If this for some reason fails, " +
s"please do a clean build of the $projectName project)"
)
```
When it is hard to fix `sbt` incremental compilation, let's restructure our project sources so that each `@TruffleInstrument` and `@TruffleLanguage` registration is in individual compilation unit. Each such unit is either going to be compiled or not going to be compiled as a batch - that will eliminate the `sbt` incremental compilation issues without addressing them in `sbt` itself.
fa2cf6a33ec4a5b2e3370e1b22c2b5f712286a75 is the first step - it introduces `IdExecutionService` and moves all the `IdExecutionInstrument` API up to that interface. The rest of the `runtime` project then depends only on `IdExecutionService`. Such refactoring allows us to move the `IdExecutionInstrument` out of `runtime` project into independent compilation unit.
Auto-generate all builtin methods for builtin `File` type from method signatures.
Similarly, for `ManagedResource` and `Warning`.
Additionally, support for specializations for overloaded and non-overloaded methods is added.
Coverage can be tracked by the number of hard-coded builtin classes that are now deleted.
## Important notes
Notice how `type File` now lacks `prim_file` field and we were able to get rid off all of those
propagating method calls without writing a single builtin node class.
Similarly `ManagedResource` and `Warning` are now builtins and `Prim_Warnings` stub is now gone.
### Pull Request Description
Using the new tooling (#3491), I investigated the **performance / compile-time tradeoff** of different codegen options for release mode builds. By scripting the testing procedure, I was able to explore many possible combinations of options, which is important because their interactions (on both application performance and build time) are complex. I found **two candidate profiles** that offer specific advantages over the current `release` settings (`baseline`):
- `thin16`: Supports incremental compiles in 1/3 the time of `baseline` in common cases. Application runs about 2% slower than `baseline`.
- `fat1-O4`: Application performs 13% better than `baseline`. Compile time is almost 3x `baseline`, and non-incremental.
(See key in first chart for the settings defining these profiles.)
We can build faster or run faster, though not in the same build. Because the effect sizes are large enough to be impactful to developer and user experience, respectively, I think we should consider having it both ways. We could **split the `release` profile** into two profiles to serve different purposes:
- `release`: A profile that supports fast developer iteration, while offering realistic performance.
- `production`: A maximally-optimized profile, for nightly builds and actual releases.
Since `wasm-pack` doesn't currently support custom profiles (rustwasm/wasm-pack#1111), we can't use a Cargo profile for `production`; however, we can implement our own profile by overriding rustc flags.
### Performance details
As you can see, `thin16` is slightly slower than `baseline`; `fat1-O4` is dramatically faster.
<details>
<summary>Methodology (click to show)</summary>
I developed a procedure for benchmarking "whole application" performance, using the new "open project" workflow (which opens the IDE and loads a complex project), and some statistical analysis to account for variance. To gather this data:
Build the application with profiling:
`./run.sh ide build --profiling-level=debug`
Run the `open_project` workflow repeatedly:
`for i in $(seq 0 9); do dist/ide/linux-unpacked/enso --entry-point profile --workflow open_project --save-profile open_project_thin16_${i}.json; done`
For each profile recorded, take the new `total_self_time` output of the `intervals` tool; gather into CSV:
`echo $(for i in $(seq 0 9); do target/rust/debug/intervals < open_project_thin16_${i}.json | tail -n1 | awk '{print $2}'; do`
(Note that the output of intervals should not be considered stable; this command may need modification in the future. Eventually it would be nice to support formatted outputs...)
The data is ready to graph. I used the `boxplot` method of the [seaborn](https://seaborn.pydata.org/index.html) package, in order to show the distribution of data.
</details>
#### Build times
In the case of changing a file in `enso-prelude`, with the current `baseline` settings rebuilding takes over 3 minutes. With the `thin16` settings, the same rebuild completes in 40 seconds.
(To gather this data on different hardware or in the future, just run the new `bench-build.sh` script for each case to be measured.)
Implemented the `order_by` function with support for all modes of operation.
Added support for case insensitive natural order.
# Important Notes
- Improved MultiValueIndex/Key to not create loads of arrays.
- Adjusted HashCode for MultiValueKey to have a simple algorithm.
- Added Text_Utils.compare_normalized_ignoring_case to allow for case insensitive comparisons.
- Fixed issues with ObjectComparator and added some unit tests for it.
[ci no changelog needed]
This PR implements a new helper for the future Component Browser - `component_group::multi::Wrapper`. It propagates FRP events from multiple component groups and ensures that only a single component group is focused at all times.
See the updated component group demo scene (console logs shows propagated FRP events from all component groups):
https://user-images.githubusercontent.com/6566674/172359141-8ea6f1ba-e357-4c1b-852a-adb4d5207e03.mp4
- Fixed a `define_endpoints_2!` macro. FRP endpoints for `focus` events weren't connected properly.
- List View now uses an overlay shape to catch mouse events, it allows much easier implementation of `is_header_selected` in the component group.
Drop `Core` implementation (replacement for IR) as it (sadly) looks increasingly
unlikely this effort will be continued. Also, it heavily relies
on implicits which increases some compilation time (~1sec from `clean`)
Related to https://www.pivotaltracker.com/story/show/182359029
This PR introduces a new structure delivered by Searcher Controller: The Component List.
The Component list is hierarchical, and its structure reflects how the components are displayed in Component Browser - only "submodule" section at this point, other sections will be covered in next tasks.
This does not introduce anything visual; the structures are tested in unit tests.
- Change dev profile settings. Improves build performance; will not affect anything else. Details below.
- Introduce script for benchmarking various incremental builds. Usage is explained in the script comments.
- Add a line to `intervals` showing total main-thread CPU work logged in a profile; this can be used to compare the results of optimizations (I'll be starting a discussion informed by that data separately; this change just enables the tooling to report it).
Fix a memory leak introduced by #3451. Found this to be part of the cause of [the buffer leaks](https://www.pivotaltracker.com/story/show/181943457/comments/231558434).
The problem with this use of `define_endpoints_2` is,
- The FRP network contains (strong) references to the `Model`.
- The `Model` has a strong reference to `api::Private`, which owns the FRP network.
Thus we never free the `Model`.
Define some workflows for batch-mode profiling.
Implemented:
- collapse nodes
- create node
- enter collapsed node
- new project
- open visualization
They can currently be built and run with a command like:
`./run.sh ide build --profiling-level=debug && dist/ide/linux-unpacked/enso --entry-point profile --workflow create_node --save-profile out.json`
And the data can be displayed with:
`dist/ide/linux-unpacked/enso --entry-point profiling_run_graph --load-profile out.json`
Demo of recording and viewing a profile with a command-line one-liner:
https://user-images.githubusercontent.com/1047859/169954795-2d9520ca-84f9-45d2-b83a-5063ebe6f718.mp4
See: https://www.pivotaltracker.com/story/show/182195399.
# Important Notes
- When defining workflows, two helpers are enough to allow us to tell when the action is really done: `Fixture::compile_new_shaders`, and `Fixture::backend_execution`. Often, it is appropriate to await both, but it depends on the task.
- The shader compiler is now driven by a `Controller`; while the `Compiler` is reset if context is lost, the `Controller`'s state survives context loss.
- A new `--load-profile` option supports specifying a profile by path when running `profiling_run_graph`.
- Drop the `with_same_start` profiler interface; we ended up preferring a child profiler convention, and this interface was not implemented compatibly with the stricter data model we've had since the introduction of `profiler::data`.
- Fix the noisy `rustfmt` output.
This change introduces a custom LogManager for console that allows for
excluding certain log messages. The primarily reason for introducing
such LogManager/Appender is to stop issuing hundreds of pointless
warnings coming from the analyzing compiler (wrapper around javac) for
classes that are being generated by annotation processors.
The output looks like this:
```
[info] Cannot install GraalVM MBean due to Failed to load org.graalvm.nativebridge.jni.JNIExceptionWrapperEntryPoints
[info] compiling 129 Scala sources and 395 Java sources to /home/hubert/work/repos/enso/enso/engine/runtime/target/scala-2.13/classes ...
[warn] Unexpected javac output: warning: File for type 'org.enso.interpreter.runtime.type.ConstantsGen' created in the last round will not be subject to annotation processing.
[warn] 1 warning.
[info] [Use -Dgraal.LogFile=<path> to redirect Graal log output to a file.]
[info] Cannot install GraalVM MBean due to Failed to load org.graalvm.nativebridge.jni.JNIExceptionWrapperEntryPoints
[info] foojavac Filer
[warn] Could not determine source for class org.enso.interpreter.node.expression.builtin.number.decimal.CeilMethodGen
[warn] Could not determine source for class org.enso.interpreter.node.expression.builtin.resource.TakeNodeGen
[warn] Could not determine source for class org.enso.interpreter.node.expression.builtin.error.ThrowErrorMethodGen
[warn] Could not determine source for class org.enso.interpreter.node.expression.builtin.number.smallInteger.MultiplyMethodGen
[warn] Could not determine source for class org.enso.interpreter.node.expression.builtin.warning.GetWarningsNodeGen
[warn] Could not determine source for class org.enso.interpreter.node.expression.builtin.number.smallInteger.BitAndMethodGen
[warn] Could not determine source for class org.enso.interpreter.node.expression.builtin.error.ErrorToTextNodeGen
[warn] Could not determine source for class org.enso.interpreter.node.expression.builtin.warning.GetValueMethodGen
[warn] Could not determine source for class org.enso.interpreter.runtime.callable.atom.AtomGen$MethodDispatchLibraryExports$Cached
....
```
The output now has over 500 of those and there will be more. Much more
(generated by our and Truffle processors).
There is no way to tell SBT that those are OK. One could potentially
think of splitting compilation into 3 stages (Java processors, Java and
Scala) but that will already complicate the non-trivial build definition
and we may still end up with the initial problem.
This is a fix to make it possible to get reasonable feedback from
compilation without scrolling mutliple screens *every single time*.
Also fixed a spurious warning in javac processor complaining about
creating files in the last round.
Related to https://www.pivotaltracker.com/story/show/182138198
- Merge the two approaches and makes them consistent
- Add warning support into Reader
# Important Notes
- Added support for JUnit format XML generation on tests. Use `ENSO_TEST_JUNIT_DIR`
* The bash entry point was renamed `run.sh` -> `run`. Thanks to that `./run` works both on Linux and Windows with PowerShell (sadly not on CMD).
* Everyone's favorite checks for WASM size and program versions are back. These can be disabled through `--wasm-size-limit=0` and `--skip-version-check` respectively. WASM size limit is stored in `build-config.yaml`.
* Improved diagnostics for case when downloaded CI run artifact archive cannot be extracted.
* Added GH API authentication to the build script calls on CI. This should fix the macOS build failures that were occurring from time to time. (Actually they were due to runner being GitHub-hosted, not really an OS-specific issue by itself.)
* If the GH API Personal Access Token is provided, it will be validated. Later on it is difficult to say, whether fail was caused by wrong PAT or other issue.
* Renamed `clean` to `git-clean` as per suggestion to reduce risk of user accidently deleting unstaged work.
* Whitelisting dependabot from changelog checks, so PRs created by it are mergeable.
* Fixing issue where wasm-pack-action (third party) randomly failed to recognize the latest version of wasm-pack (macOS runners), leading to failed builds.
* Build logs can be filtered using `ENSO_BUILD_LOG` environment variable. See https://docs.rs/tracing-subscriber/0.3.11/tracing_subscriber/struct.EnvFilter.html#directives for the supported syntax.
* Improve help for ci-run source, to make clear that PAT token is required and what scope is expected there.
Also, JS parts were updated with some cleanups and fixes following the changes made when introducing the build script.
`interpreter-dsl` should only attempt to run explicitly specified
processors. That way, even if the generated
`META-INF/services/javax.annotation.processing.Processor` is present,
it does not attempt to apply those processors on itself.
This change makes errors related to
```
[warn] Unexpected javac output: error: Bad service configuration file, or
exception thrown while constructing Processor object:
javax.annotation.processing.Processor: Provider org.enso.interpreter.dsl....
```
a thing of the past. This was supper annoying when switching branches and
required to either clean the project or remove the file by hand.
Related to https://www.pivotaltracker.com/story/show/182297597
- Added new `Statistic`s: Covariance, Pearson, Spearman, R Squared
- Added `covariance_matrix` function
- Added `pearson_correlation` function to compute correlation matrix
- Added `rank_data` and Rank_Method type to create rankings of a Vector
- Added `spearman_correlation` function to compute Spearman Rank correlation matrix
# Important Notes
- Added `Panic.throw_wrapped_if_error` and `Panic.handle_wrapped_dataflow_error` to help with errors within a loop.
- Removed `Array.set_at` use from `Table.Vector_Builder`
There are two dirty flags in layers: depth_order_dirty and element_depth_order_dirty - one marking changed in Layer, second marking change in one of sublayers. The depth_order_dirty has a proper callback for setting element_depth_order_dirty of its parent. However, the latter did not propagate up.
I fixed it by adding callback for element_depth_order_dirty which sets the depth_order_dirty of the parent.
# Important Notes
* The question to @wdanilo : is it possible, that I can propagate dirty directly to element_depth_order_dirty, without setting depth_order_dirty? As far as I understand the code, it would also work (and we would omit some unnecessary updates).
* I tried to leave some logs, but I don't feel how to do that: the tooling I used was very specific, only the concrete ids of symbols and layers were logged, and I don't know how to generalize it.
This PR contains minimal integration with new engine's method and an integration test printing the method's return value. It was written as a part of https://www.pivotaltracker.com/story/show/181743571
# Important Notes
The test requires 2022.1.1-nightly.2022-04-26 engine version or later.
Add a method in `SuggestionDatabase` allowing to find a suggestion entry by a fully qualified path, working faster than a linear search through all the entries in the `SuggestionDatabase`.
https://www.pivotaltracker.com/story/show/181852566
# Important Notes
- **Testing:** when testing the PR with the newest currently available nightly version of the Engine, you may observe the following warnings in the Chrome JS Developer Console (the numerical values may differ):
index.ts?ab16:289 WARN app/gui/src/model/suggestion_database.rs:61 An existing suggestion entry id at Standard.Base.Nothing.Nothing.is_nothing was overwritten with 768.
index.ts?ab16:289 WARN app/gui/src/model/suggestion_database.rs:61 An existing suggestion entry id at Standard.Base.Data.Numbers.Integer.up_to was overwritten with 936.
index.ts?ab16:289 WARN app/gui/src/model/suggestion_database.rs:61 An existing suggestion entry id at Standard.Base.Data.Numbers.Integer.down_to was overwritten with 937.
index.ts?ab16:289 WARN app/gui/src/model/suggestion_database.rs:61 An existing suggestion entry id at Standard.Base.Data.Text.Text.== was overwritten with 971.
This is a result of bugs in the Standard Library. Those are planned to be addressed by:
- #3480 short-term (to fix the specific bugs currently present in the Standard Library),
- https://www.pivotaltracker.com/story/show/182283983 long-term (to improve the Engine such that it disallows introducing this category of bugs in the Standard Library in the future).
(For more details, see also: https://discord.com/channels/401396655599124480/978929754138877962.)
As a result of the bugs mentioned above, the Engine is responding with some invalid replies. In case of such invalid replies, warnings are emitted in the JS Dev Console. Other than the warnings, the code is expected to work correctly with the Engine for the cases when the Engine returns correct data.
- A `HashMapTree` was used for storing the map. A quick back-of-the-envelope estimation of memory usage of a simpler alternative (a one-level `HashMap`) would put the alternative at ~1MB (~10k entities × averaged ~100 bytes per entity path), which was considered too much in a discussion with @farmaazon.
- The `HashMapTree::remove` method deletes a whole subtree of a `HashMapTree`. An alternative removal method was implemented for use in `suggestion_database::QualifiedNameToIdMap` to better match `HashMap` entry removal semantics.
- In case of path collisions, a warning is emitted.
- Paths are treated case-sensitively.
- The new method of the `SuggestionDatabase` type is currently only used in unit tests. This matches the explicit requirement in the Task's description.
- JS Developer Console logging was enabled for the `tracing` package. The `WARN` level was picked to match the default level enabled in the "old" logging infrastructure.
[ci no changelog needed]
@radeusgd discovered that no formatting was being applied to std-bits projects.
This was caused by the fact that `enso` project didn't aggregate them. Compilation and
packaging still worked because one relied on the output of some tasks but
```
sbt> javafmtAll
```
didn't apply it to `std-bits`.
# Important Notes
Apart from `build.sbt` no manual changes were made.
This PR extends the Component Group Entry with icon and option to highlight the text. Here the convert has highlighted "con".
https://user-images.githubusercontent.com/3919101/169046537-4f8b823c-322e-40dc-8abb-24d1d7092341.mp4
### Important Notes
Although this PR includes effort for adjusting Component Group style to better reflect the design, it is not entirely finished: the selection still works badly and will be fixed in another PR.
