* Add parsing keyword function calls inside pipelines
Co-authored-by: Adam Chalmers <adam.chalmers@zoo.dev>
* Add three point circle stdlib function
* Generate new documentation
* Fix 20:20 for the circle three point test
* Convert to using keyword arguments
* Wtf yo
* Remove unused structure
* Use the new simulation tests
* Regenerate documentation
---------
Co-authored-by: Jonathan Tran <jonnytran@gmail.com>
Co-authored-by: Adam Chalmers <adam.chalmers@zoo.dev>
* Add support for float numbers in rem() arguments
* Update docs
* Rename to prevent name collision in docs
* Update docs after rename to NumberArg
* Fix parameter types to follow naming convention
* Change doc comment description to not refer to floating point
* Update docs after NumberArg doc change
* Change function to be more condensed
* Change to not try to preserve ints
* Update docs to use f64
* A snapshot a day keeps the bugs away! 📷🐛 (OS: namespace-profile-macos-8-cores)
* Trigger CI
---------
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Implement some basic cache program generation
A bit ago, @jessfraz added the ability to control reexecution from the
executor. When we did this, it used the digest to determine if there
was a code change rather than a non-code change (whitespace, comment,
etc).
This allows the creation of a new program to be run without clearing the
scene. This, in conjunction with being able to delete Engine objects by
ID will allow us to do some clever stuff when incrementally executing
a program.
I'm still working on something a bit more advanced, but a good first
step to derisk some of the caching behavior here fully is to implement a
basic "changed program" stub.
This process the ast programs (old and new) if it doesn't exactly match.
This would have been a complete refresh before this commit.
1) Check all overlapping top-level statements of the body of the new and
old AST and ensure they all match.
- If this is true, this means that one of the two AST programs has more
elements then the other, and they all otherwise match (addition or
deletion of the end of the program). We continue to #2 in this
case.
- If this is false, we have a meaingful difference in a section of
overlapping code. This will result in a cache miss and rebuild
the scene. We short-cut here and the scene is rebuilt.
2) Check the lengths of the two bodies.
- If they're the same, we shouldn't have even been called. We will
short-cut with a noop cache return (no clear, no program).
- if the old ast is longer, we've removed instructions from the
program. We can't delete things now, so this will result in a cache
miss and rebuild the scene. We short-cut here and the scene is
rebuilt.
- If the new ast is longer, we have an insertion of code at the end.
3) construct a new program using only the new elements from the new
ast, and return a `CacheResult` that *does not clear the scene*.
This means nothing will be rebuilt, and only a new object will polp
onto the scene. This is the first case where we diverge with
existing behavior.
Signed-off-by: Paul R. Tagliamonte <paul@zoo.dev>
* KCL: Show beautiful Miette errors when a KCL example test fails
Background: KCL example tests are generated from the stdlib KCL examples in the `#[stdlib]` macro in derive-docs.
Problem: When these tests fail, they output a really unhelpful error message like Kcl(Semantic(KclErrorDetails { source_ranges: [156, 160, 0], message: "Expected a sketch but found array" } )).
Solution: Use miette. Now the errors highlight the KCL code that failed and show exactly what went wrong, on which line, presenting nice diagnostics that look like cargo/rustc output.
* Update helix snapshots
Part of #4600
KCL functions can declare one special argument that doesn't require a label on its parameter when called.
This PR will default that arg to % (the current pipeline) if not given.
* Parse annotations
Signed-off-by: Nick Cameron <nrc@ncameron.org>
* Propagate settings from annotations to exec_state
Signed-off-by: Nick Cameron <nrc@ncameron.org>
---------
Signed-off-by: Nick Cameron <nrc@ncameron.org>
* redoing bool logic impl on latest main
* adding snapshot tests (removing .new)
* removing accidental change smh:(
* accepting client side scene snapshot
* accepting png snapshot and triggering ci
* A snapshot a day keeps the bugs away! 📷🐛 (OS: ubuntu-latest-8-cores)
* accepting png again?
* A snapshot a day keeps the bugs away! 📷🐛 (OS: windows-latest-8-cores)
* accepting grid visibility snapshot
* A snapshot a day keeps the bugs away! 📷🐛 (OS: windows-latest-8-cores)
* accepting png snapshot
* A snapshot a day keeps the bugs away! 📷🐛 (OS: windows-latest-8-cores)
* accepting png snapshot
* A snapshot a day keeps the bugs away! 📷🐛 (OS: ubuntu-latest-8-cores)
* accepting png snapshot
* rerunning simtest creation to get ops.snap files
---------
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Jonathan Tran <jonnytran@gmail.com>
* Allow underscores but only for un-referenced names
Signed-off-by: Nick Cameron <nrc@ncameron.org>
* Support numeric suffixes for UoM types
Signed-off-by: Nick Cameron <nrc@ncameron.org>
* UoM type arguments
Signed-off-by: Nick Cameron <nrc@ncameron.org>
* warnings -> non-fatal errors
Signed-off-by: Nick Cameron <nrc@ncameron.org>
* type ascription
Signed-off-by: Nick Cameron <nrc@ncameron.org>
---------
Signed-off-by: Nick Cameron <nrc@ncameron.org>
* Add operations tracking for the timeline
* Change to only track certain stdlib functions as operations
* Update gen files
* Add operations to simulation snapshot tests
* Add tracking of positional function calls
* Fix generated field names to be camel case in TS
* Fix generated TS field names to match and better docs
* Fix order of ops with patternTransform
* Fix sweep to be included
* Add new expected test outputs
* Add tracking for startSketchOn
* Update ops output to include startSketchOn
* Fix serde field name
* Fix output field name
* Add tracking of operations that fail
* Add snapshots of operations even when there's a KCL execution error
* Add ops output for error executions
* Add operations output to executor error
* Update op source ranges
* Remove tracking of circle() and polygon() since they're not needed
* Update output without circle and polygon
* Fix to track patternCircular3d and patternLinear3d
* Remove tracking for mirror2d
* Update ops output
* Fix to track the correct source range of function definitions
---------
Co-authored-by: Frank Noirot <frank@zoo.dev>
When engine merged their big extrude ID bugfix,
they changed the response for extrudes on face.
Basically there's no longer a start face for
something you extrude from a face, because there
just isn't. There's a hole in a previous face,
it's just a gap.
Co-authored-by: Jonathan Tran <jonnytran@gmail.com>
* Bump wasm-bindgen from 0.2.91 to 0.2.99 and wasm-bindgen-futures from 0.4.44 to 0.4.49
* Upgrade in the kcl crate also
* Update web-sys version constraint to match lock
Remove non code from Digests
@jessfraz and I talked it over; for the time being we're going to remove
comments from the AST digest. We already exclude source position, so
this is just increasing the degree to which we're going to ignore things
that are not germane to execution.
Before, we'd digest *some* but not all of the comments in the AST.
Silly, I know, right?
So, this code:
```
firstSketch = startSketchOn('XY')
|> startProfileAt([-12, 12], %)
|> line([-24, 0], %) // my thing
|> close(%)
|> extrude(6, %)
```
Would digest differently than:
```
firstSketch = startSketchOn('XY')
|> startProfileAt([-12, 12], %)
|> line([-24, 0], %)
|> close(%)
|> extrude(6, %)
```
Which is wrong. We've fully divested of hashing code comments, so this
will now hash to be the same. Hooray.
* More Walk cleanup
- The `Node` type contained two enums by mistake. Those have been
removed.
- Export the `Visitor` and `Visitable` traits, as I start to migrate
stuff to them.
- Add a wrapper to pull the `digest` off the node without doing a
`match` elsewhere.
* update camera controls to Zoo
* update e2e and initial settings
* update types and camera controls ts
* update mod.rs test
* update test, test locally
* Reapply "More aggressive using of cache on engine settings changes (#4691)" (#4729)
This reverts commit 3f1f40eeba.
* Add a utility to get all the current values from the settings object
* Use an XState selector to get the latest settings snapshot for WASM
---------
Co-authored-by: Frank Noirot <frank@kittycad.io>
Part of https://github.com/KittyCAD/modeling-app/issues/4600
You can now call a user-defined function via keyword args. E.g.
```
fn increment(@x) {
return x + 1
}
fn add(@x, delta) {
return x + delta
}
two = increment(1)
three = add(1, delta: 2)
```
* Add test with tag parameter
* Fix to not define TagIdentifiers pointing to nothing
* Add fixed test output
* Rename function to be clearer
* Remove optional param so that unparse works
* Fix to not allow redefining tags
* Fix to only ever define tags in memory when used with stdlib functions
* Fix to not define local variables when non-literal tag declarator is used
* Add removing mutability
* Change function signature since it isn't falliable
* Refactor TokenStream (and some minor changes to Token)
Signed-off-by: Nick Cameron <nrc@ncameron.org>
* Tidy up lexer tests
Signed-off-by: Nick Cameron <nrc@ncameron.org>
---------
Signed-off-by: Nick Cameron <nrc@ncameron.org>
# Problem
Before this PR, our formatter reformats
```
squares_out = reduce(arr, 0, fn (i, squares) {
return 1
})
```
to
```
squares_out = reduce(arr, 0, (i, squares) {
return 1
})
```
i.e. it removes the `fn` keyword from the closure. This keyword is required, so, our formatter turned working code into invalid code.
# Cause
When this closure parameter is formatted, the ExprContext is ::Decl, so `Expr::recast` skips adding the `fn` keyword. The reason it's ::Decl is because the `squares_out = ` declaration sets it, and no subsequent call sets the context to something else.
# Solution
When recasting a call expression, set the context for every argument to `ExprContext::Other`.
* start a rework of the walk module
I'd like to have the code explicitly recurse in the visitor rather than
implicitly. This allows the calling code to build up an idea of the
depth of each node, or skip parts of the AST entirely.
i'm going to rebuild this enough to get the old API to work before
looking to merge this.
I've also discovered a number of new AST types that were just papered
over and/or excluded entirely -- I'm going to have to go through and
make sure that both Digest and the walker are still current, or if we
silently added types or, more likely, fields getting ignored.
