Compiler performed name resolution of literals in type signatures but would silently fail to report any problems.
This meant that wrong names or forgotten imports would sneak in to stdlib.
This change introduces 2 main changes:
1) failed name resolutions are appended in `TypeNames` pass
2) `GatherDiagnostics` pass also collects and reports failures from type
signatures IR
Updated stdlib so that it passes given the correct gatekeepers in place.
Implements https://www.pivotaltracker.com/story/show/184032869
# Important Notes
- Currently we get failures in Full joins on Postgres which show a more serious problem - amending equality to ensure that `[NULL = NULL] == True` breaks hash/merge based indexing - so such joins will be extremely inefficient. All our joins currently rely on this notion of equality which will mean all of our DB joins will be extremely inefficient.
- We need to find a solution that will support nulls and still work OK with indices (but after exploring a few approaches: `COALESCE(a = b, a IS NULL AND b is NULL)`, `a IS NOT DISTINCT FROM b`, `(a = b) OR (a IS NULL AND b is NULL)`; all of which did not work (they all result in `ERROR: FULL JOIN is only supported with merge-joinable or hash-joinable join conditions`) I'm less certain that it is possible. Alternatively, we may need to change the NULL semantics to align it with SQL - this seems like likely the simpler solution, allowing us to generate simple, reliable SQL - the NULL=NULL solution will be cornering us into nasty workarounds very dependent on the particular backend.
`runtime-with-instruments` project sets `-Dgraalvm.locatorDisabled=true` that disables the discovery of available polyglot languages (installed with `gu`). On the other hand, enabling locator makes polyglot languages available, but also makes the program classes and the test classes loaded with different classloaders. This way we're unable to use `EnsoContext` in tests to observe internal context state (there is an exception when you try to cast to `EnsoContext`).
The solution is to move tests with enabled polyglot support, but disabled `EnsoContext` introspection to a separate project.
PR adds a flag to `Text` implementation tracking whether it is in a FCD normal form. Then this information can be used in the `Normalizer.compare` method.
| Benchmark name | Old (ms) | With flag (ms)
| --- | --- | ---
| Unicode very short | 40.29 | 40.04
| Unicode medium | 9.07 | 1.99
| Unicode big - random | 115.39 | 0.35
| Unicode big - early difference | 107.02 | 0.54
| Unicode big - late difference | 749.81 | 94.73
| ASCII very short | 28.13 | 31.13
| ASCII medium | 4.58 | 2.26
| ASCII big - random | 42.68 | 0.26
| ASCII big - early difference | 30.91 | 0.32
| ASCII big - late difference | 66.29 | 42.72
Full benchmark output.
[bench_old.txt](https://github.com/enso-org/enso/files/10325202/bench_old.txt)
[bench_new.txt](https://github.com/enso-org/enso/files/10325201/bench_new.txt)
`NestedPatternMatch` pass desugared complex patterns in a very inefficient way resulting in an exponential generation of the number of `case` IR (and Truffle) nodes. Every failed nested pattern would copy all the remaining patterns of the original case expression, in a desugared form. While the execution itself of such deeply nested `case` expression might not have many problems, the time spent in compilation phases certainly was a blocker.
This change desugars deeply nested into individual cases with a fallthrough logic. However the fallthrough logic is implemented directly in Truffle nodes, rather than via IR. That way we can generate much simpler IR for nested patterns.
Consider a simple case of
```
case x of
Cons (Cons a b) Nil -> a + b
Cons a Nil -> a
_ -> 0
```
Before the change, the compiler would generate rather large IR even for those two patterns:
```
case x of
Cons w y -> case w of
Cons a b -> case y of
Nil -> a + b
_ -> case x of
Cons a z -> case z of
Nil -> a
_ -> case x of
_ -> 0
_ -> 0
_ -> case x of
Cons a z -> case z of
Nil -> a
_ -> case x of
_ -> 0
_ -> 0
Cons a z -> case z of
Nil -> a
_ -> case x of
_ -> 0
_ -> 0
```
Now we generate simple patterns with fallthrough semantics and no catch-all branches:
```
case x of
Cons w y -> case w of
Cons a b -> case y of ## fallthrough on failed match ##
Nil -> a + b ## fallthrough on failed match ##
Cons a z -> case z of
Nil -> a ## fallthrough on failed match ##
_ -> 0
```
# Important Notes
If you wonder how much does it improve, then @radeusgd's example in https://www.pivotaltracker.com/story/show/183971366/comments/234688327 used to take at least 8 minutes to compile and run.
Now it takes 5 seconds from cold start.
Also, the example in the benchmark includes compilation time on purpose (that was the main culprit of the slowdown).
For the old implementation I had to kill it after 15 minutes as it still wouldn't finish a single compilation.
Now it runs 2 seconds or less.
Bonus points: This PR will also fix problem reported in https://www.pivotaltracker.com/story/show/184071954 (duplicate errors for nested patterns)
Nested type declaration shall yield a syntax error
# Important Notes
Now the Radek's sample:
```
type Foo
type Bar
main = 42
```
yields
```bash
$ enso --run test.enso
In module test:
Compiler encountered errors:
test.enso[2:9-2:16]: Unexpected declaration in the body of a type.
Aborting due to 1 errors and 0 warnings.
Execution finished with an error: Compilation aborted due to errors.
```
`Any.==` is a builtin method. The semantics is the same as it used to be, except that we no longer assume `x == y` iff `Meta.is_same_object x y`, which used to be the case and caused failures in table tests.
# Important Notes
Measurements from `EqualsBenchmarks` shows that the performance of `Any.==` for recursive atoms increased by roughly 20%, and the performance for primitive types stays roughly the same.
First part of fixing `Text.to_text`.
- add: `pretty` method for pretty printing.
- update: make `Text.to_text` conversion identity for Text
In the next iterations `to_text` will be gradually replaced with `to Text` conversion once the related issues with conversions are fixed.
Implements `getMetaObject` and related messages from Truffle interop for Enso values and types. Turns `Meta.is_a` into builtin and re-uses the same functionality.
# Important Notes
Adds `ValueGenerator` testing infrastructure to provide unified access to special Enso values and builtin types that can be reused by other tests, not just `MetaIsATest` and `MetaObjectTest`.
Use JavaScript to parse and serialise to JSON. Parses to native Enso object.
- `.to_json` now returns a `Text` of the JSON.
- Json methods now `parse`, `stringify` and `from_pairs`.
- New `JSON_Object` representing a JavaScript Object.
- `.to_js_object` allows for types to custom serialize. Returning a `JS_Object`.
- Default JSON format for Atom now has a `type` and `constructor` property (or method to call for as needed to deserialise).
- Removed `.into` support for now.
- Added JSON File Format and SPI to allow `Data.read` to work.
- Added `Data.fetch` API for easy Web download.
- Default visualization for JS Object trunctes, and made Vector default truncate children too.
Fixes defect where types with no constructor crashed on `to_json` (e.g. `Matching_Mode.Last.to_json`.
Adjusted default visualisation for Vector, so it doesn't serialise an array of arrays forever.
Likewise, JS_Object default visualisation is truncated to a small subset.
New convention:
- `.get` returns `Nothing` if a key or index is not present. Takes an `other` argument allowing control of default.
- `.at` error if key or index is not present.
- `Nothing` gains a `get` method allowing for easy propagation.
This removes the special handling of polyglot exceptions and allows matching on Java exceptions in the same way as for any other types.
`Polyglot_Error`, `Panic.catch_java` and `Panic.catch_primitive` are gone
The change mostly deals with the backslash of removing `Polyglot_Error` and two `Panic` methods.
`Panic.catch` was implemented as a builtin instead of delegating to `Panic.catch_primitive` builtin that is now gone.
This fixes https://www.pivotaltracker.com/story/show/182844611
Benchmark to compare _curried and lambda_ based function invocations and a fix to make _curried_ invocation (at least) as fast as the _lambda_ one. Allows us to use _curried invocations_ in standard library again without loosing any speed.
# Important Notes
Execute as:
```
sbt:runtime> benchOnly CurriedFunctionBenchmarks
```
Prior to subsequent bugfixes in this PR the benchmark results were:
- `averageCurried` runs in 0.290 ms
- `averageLambda` runs in 0.122 ms
e.g. _curried invocations_ is more than twice slow. That confirms our findings from the `Array_Proxy` vector benchmarks. The problem is that _function object is not compilation final_. After fixing it we have following results:
- `averageCurried` runs in 0.102 ms
- `averageLambda` runs in 0.111 ms
e.g. both operations are of similar complexity.
- Implemented https://www.pivotaltracker.com/story/show/183913276
- Refactored MultiValueIndex and MultiValueKeys to be more type-safe and more direct about using ordered or unordered maps.
- Added performance tests ensuring we use an efficient algorithm for the joins (the tests will fail for a full O(N*M) scan).
- Removed some duplicate code in the Table library.
- Added optional coloring of test results in terminal to make failures easier to spot.
- Aligned `compare_to` so returns `Type_Error` if `that` is wrong type for `Text`, `Ordering` and `Duration`.
- Add `empty_object`, `empty_array`. `get_or_else`, `at`, `field_names` and `length` to `Json`.
- Fix `Json` serialisation of NaN and Infinity (to "null").
- Added `length`, `at` and `to_vector` to Pair (allowing it to be treated as a Vector).
- Added `running_fold` to the `Vector` and `Range`.
- Added `first` and `last` to the `Vector.Builder`.
- Allow `order_by` to take a single `Sort_Column` or have a mix of `Text` and `Sort_Column.Name` in a `Vector`.
- Allow `select_columns_helper` to take a `Text` value. Allows for a single field in group_by in cross_tab.
- Added `Patch` and `Custom` to HTTP_Method.
- Added running `Statistic` calculation and moved more of the logic from Java to Enso. Performance seems similar to pure Java version now.
Fighting with _too many messages being delivered_ I wrote a test that dumps information about `AvoidIdInstrumentationTag` - every node that has `AvoidIdInstrumentationTag` is excluded from the instrumentation. However, when I look at the output for
```
from Standard.Base import all
import Standard.Visualization
run n = 0.up_to n . map i-> 1.noise * i
```
I see that `1.noise` didn't have the tag. Now there is [AvoidIdInstrumentationTagTest.java](https://github.com/enso-org/enso/pull/3973/files#diff-32cd9240bda2bfe0e5904695ced008daba86fefb3d137ac401997f4265fa50eb) which can be used to collect all programs where _too many messages is being delivered_. Just add a program, identify _isLambda_ and verify all nodes are properly tagged.
```
from Standard.Base import all
main =
value = 1
case value of
1 -> IO.println "one"
2 -> IO.println "two"
-1 -> IO.println "minus one"
_ -> IO.println "other"
```
# Important Notes
Had to write new `ExecCompilerTest` as comparing the AST with old compiler produced miserable results - the old AST wasn't ready for negative constants in `case of` at all.
Using lambda instead of higher order function.
# Important Notes
Running:
```
sbt:runtime> benchOnly VectorBenchmarks.averageOverArrayProxy
```
speeds up from `0.038 ms/op` to `0.016 ms/op` on my computer. Which seems good enough.
Fixes bug in visualization of host polyglot values - `ExpressionNode` is only wrapped once Chrome inspector instrument is attached to the context. With this fix, when chromeinspector is attached (`enso --run --inspect ...`), all the host values are reinterpreted as text - the assumption is invalidated. But when running as language server, nothing is wrapped.
Allow arbitrary expression evaluation in the chromeinspector console. Moreover, allow modifications of any variable in any stack frame.
# Important Notes
- Implement inline parsing in `EnsoLanguage.parse(InlineParsingRequest)`.
- Debugging experience is affected by this [bug in Truffle](https://github.com/oracle/graal/issues/5513), which causes NPEs when a host object gets into chromeinspector. I tried to implement a workaround, but it does not work all the time. Nevertheless, it should not matter that much - if there is a NPE in the debugger, you can just ignore it, as it should be concealed in the debugger and should not be propagted outside. See comments in the `docs/debugger`.
* Sequence literal (Vector) should preserve warnings
When Vector was created via a sequence literal, we simply dropped any
associated any warnings associated with it.
This change propagates Warnings during the creation of the Vector.
Ideally, it would be sufficient to propagate warnings from the
individual elements to the underlying storage but doesn't go well with
`Vector.fromArray`.
* update changelog
* Array-like structures preserver warnings
Added a WarningsLibrary that exposes `hasWarnings` and `getWarnings`
messages. That way we can have a single storage that defines how to
extract warnings from an Array and the others just delegate to it.
This simplifies logic added to sequence literals to handle warnings.
* Ensure polyglot method calls are warning-free
Since warnings are no longer automatically extracted from Array-like
structures, we delay the operation until an actual polyglot method call
is performed.
Discovered a bug in `Warning.detach_selected_warnings` which was missing
any usage or tests.
* nits
* Support multi-dimensional Vectors with warnings
* Propagate warnings from case branches
* nit
* Propagate all vector warnings when reading element
Previously, accessing an element of an Array-like structure would only
return warnings of that element or of the structure itself.
Now, accessing an element also returns warnings from all its elements as
well.
- Moved `to_default_visualization_data` to `Standard.Visualization`.
- Remove the use of `is_a` in favour of case statements.
- Stop exporting Standard.Base.Error.Common.
- Separate errors to own files.
- Change constructors to be called `Error`.
- Rename `Caught_Panic.Caught_Panic_Data` -> `Caught_Panic.Panic`.
- Rename `Project_Description.Project_Description_Data` ->`Project_Description.Value`
- Rename `Regex_Matcher.Regex_Matcher_Data` -> `Regex_Matcher.Value` (can't come up with anything better!).
- Rename `Range.Value` -> `Range.Between`.
- Rename `Interval.Value` -> `Interval.Between`.
- Rename `Column.Column_Data` -> `Column.Value`.
- Rename `Table.Table_Data` -> `Table.Value`.
- Align all the Error types in Table.
- Removed GEO Json bits from Table.
- `Json.to_table` doesn't have the GEO bits anymore.
- Added `Json.geo_json_to_table` to add the functions back in.
# Important Notes
No more exports from anywhere but Main!
No more `_Data` constructors!
- Moved `Any`, `Error` and `Panic` to `Standard.Base`.
- Separated `Json` and `Range` extensions into own modules.
- Tidied `Case`, `Case_Sensitivity`, `Encoding`, `Matching`, `Regex_Matcher`, `Span`, `Text_Matcher`, `Text_Ordering` and `Text_Sub_Range` in `Standard.Base.Data.Text`.
- Tidied `Standard.Base.Data.Text.Extensions` and stopped it re-exporting anything.
- Tidied `Regex_Mode`. Renamed `Option` to `Regex_Option` and added type to export.
- Tidied up `Regex` space.
- Tidied up `Meta` space.
- Remove `Matching` from export.
- Moved `Standard.Base.Data.Boolean` to `Standard.Base.Boolean`.
# Important Notes
- Moved `to_json` and `to_default_visualization_data` from base types to extension methods.
Converting `Integer.parse` into a builtin and making sure it can parse big values like `100!`. Adding `locale` parameter to `Locale.parse` and making sure it parses `32,5` as `32.5` double in Czech locale.
# Important Notes
Note that one cannot
```
import Standard.Table as Table_Module
```
because of the 2-component name restriction that gets desugared to `Standard.Table.Main` and we have to write
```
import Standard.Table.Main as Table_Module
```
in a few places. Once we move `Json.to_table` extension this can be improved.
Disabling `musl` as it isn't capable to load dynamic library.
# Important Notes
With this change it is possible to:
```
$ sbt bootstrap
$ sbt engine-runner/buildNativeImage
$ ./runner --run ./engine/runner/src/test/resources/Factorial.enso 3
6
$ ./runner --run ./engine/runner/src/test/resources/Factorial.enso 4
24
$ ./runner --run ./engine/runner/src/test/resources/Factorial.enso 100
93326215443944152681699238856266700490715968264381621468592963895217599993229915608941463976156518286253697920827223758251185210916864000000000000000000000000
```
Is it OK, @radeusgd to disable `musl`? If not, we would have to find a way to link the parser in statically, not dynamically.
It appears that when were doing
`from XYZ import all`
the module `XYZ` was also being taken into account during name resolution.
This was unfortunate and became problematic when one had a type with the same name defined in it.
During pattern matching one could not simply do
```
from XYZ import all
...
case ... of
_ : XYZ -> ...
```
since the compiler would complain that we try to pattern match on a type but give it a module.
The module is now excluded from the name resolution, when importing everything from the module.
It appears that this "feature" was used in a number of our tests, so they had to be adapted.
This fixes task 4 in https://www.pivotaltracker.com/story/show/183833055
By default all polyglot symbols that have been imported were always exported. This means that importing a module that had some polyglot imports brought them into the scope automatically. This didn't follow our desired semantics.
Fixes task 3 in https://www.pivotaltracker.com/story/show/183833055.
- Export all for `Problem_Behavior` (allowing for Report_Warning, Report_Error and Ignore to be trivially used).
- Renamed `Range.Range_Data` to `Range.Value` moved to using `up_to` wherever possible.
- Reviewed `Function`, `IO`, `Polyglot`, `Random`, `Runtime`, `System`.
- `File` now published as type. Some static methods moved to `Data` others into type. Removed `read_bytes` static.
- New `Data` module for reading input data in one place (e.g. `Data.read_file`) will add `Data.connect` later.
- Added `Random` module to the exports.
- Move static methods into `Warning` type and exporting the type not the module.
# Important Notes
- Sorted a few imports into order (ordering by direct import in project, then by from import in project then polyglot and finally self imports).
Upgrading to GraalVM 22.3.0.
# Important Notes
- Removed all deprecated `FrameSlot`, and replaced them with frame indexes - integers.
- Add more information to `AliasAnalysis` so that it also gathers these indexes.
- Add quick build mode option to `native-image` as default for non-release builds
- `graaljs` and `native-image` should now be downloaded via `gu` automatically, as dependencies.
- Remove `engine-runner-native` project - native image is now build straight from `engine-runner`.
- We used to have `engine-runner-native` without `sqldf` in classpath as a workaround for an internal native image bug.
- Fixed chrome inspector integration, such that it shows values of local variables both for current stack frame and caller stack frames.
- There are still many issues with the debugging in general, for example, when there is a polyglot value among local variables, a `NullPointerException` is thrown and no values are displayed.
- Removed some deprecated `native-image` options
- Remove some deprecated Truffle API method calls.
Previously, when exporting the same module multiple times only the first statement would count and the rest would be discarded by the compiler.
This change allows for multiple exports of the same module e.g.,
```
export project.F1
from project.F1 export foo
```
Multiple exports may however lead to conflicts when combined with hiding names. Added logic in `ImportResolver` to detect such scenarios.
This fixes https://www.pivotaltracker.com/n/projects/2539304/stories/183092447
# Important Notes
Added a bunch of scenarios to simulate pos and neg results.
RuntimeStdlibTest now checks that names in type signatures are qualified (aka poor man's typechecker)
Will be merged after the stdlib tidying by James is complete.
Fixes `NullPointerException` in `TreeToIr` by detecting the case and raising `UnexpectedExpression` syntax error.
# Important Notes
This PR prevents the `NullPointerException`. It doesn't change the meaning of `x.` (followed by space) `length`. It just detects when `Tree.OprApp.getRhs() == null` and treats that as an error.
- Moved static methods into `Locale` type. Publishing type not module.
- Stop publishing `Nil` and `Cons` from `List`.
- Tidied up `Json` and merged static in to type. Sorted out various type signatures which used a `Constructor`. Now exporting type and extensions.
- Tidied up `Noise` and merge `Generator` into file. Export type not module.
- Moved static method of `Map` into type. Publishing type not module.
# Important Notes
- Move `Text.compare_to` into `Text`.
- Move `Text.to_json` into `Json`.
* Tidy Bound and Interval.
* Fix Interval tests.
* Fix Interval tests.
* Restructure Index_Sub_Range to new Type/Statics.
* Adjust for Vector exported as a type and static methods on it.
* Tidy Maybe.
* Fix issue with Line_Ending_Style.
* Revert Filter_Condition change.
Fix benchmark test issue.
Tidy imports on Index_Sub_Range.
* Revert Filter_Condition change.
Fix benchmark test issue.
Tidy imports on Index_Sub_Range.
* Can't export constructors unless exported from type in module.
* Fix failing tests.
