Fixes one failure in #8942 which caught a real issue: clicks at various panels were triggering many handlers at once, often unexpectedly - for example quick clicking at breadcrumbs (and automatic tests always click fast) we could also trigger getting out the current node.
Therefore, I added `stopPropagation` for all mouse events on "panel" level + where I think the event should be considered "handled" and no longer bother anyone. Also, to unify things, most actions are for `click` event.
Additionally I spotted and fixed some issues:
* When "clicking-off" Component Browser, it creates new node only if anything was actually typed in (no more dangling `operatorX.` nodes)
* Filtering now works for operators
* When, after opening CB with source node, user starts typing operator, we replace dot with space
* Fixed our shortcut handler, so it works properly with `click` event.
* Fixed problems with defocusing input in CB when clicking at links.
# Important Notes
I removed `PointerMain` binding for deselectAll, because it was triggered every time did the area selection.
- Fix a bug where longitude was used as both latitude and longitude, due to a typo....
# Important Notes
- As our Mapbox token is no longer valid, the map visualization no longer works, even with this fix. As a result, this cannot be tested properly.
- However, this can still be tested somewhat, by seeing that the error is no longer present in the repro:
- Open the `Colorado COVID` template
- Open the viz on this node (at the bottom right of graph. it may help to zoom out using Ctrl+Shift+A.):
![image](https://github.com/enso-org/enso/assets/4046547/54849e5f-8a53-4f5b-8bdf-f08fe8ca51f0)
- Compare the errors logged in the console between `develop`, and this branch. This branch should not have the error about `lat` (latitude) being out of range.
* Prevent duplication of external IDs in syncTree
* trim component browser accepted value
* use flush post for visualization data watcher
* do not create empty nodes when closing component browser
* Lint
---------
Co-authored-by: Kaz <kaz@lambdaverse.org>
- Fix the UI problems with our CodeMirror integration (Fixed view stability; Fixed a focus bug; Fixed errors caused by diagnostics range exceptions; Fixed linter invalidation--see https://discuss.codemirror.net/t/problem-trying-to-force-linting/5823; Implemented edit-coalescing for performance).
- Introduce an algorithm for applying text edits to an AST. Compared to the GUI1 approach, the new algorithm supports deeper identity-stability for expressions (which is important for subexpression metadata and Y.Js sync), as well as reordered-subtree identification.
- Enable the code editor.
Fixes#9049
There were still some stubs left in node creation code.
# Important Notes
I haven't added any tests, as unit testing the graph store is very difficult - it has much tangled logic there. I'm going to try to untangle it a bit, but in a separate PR.
Fixes#9008
Now, even if we inherit FunctionCall config from parent widget (e.g. drop down), we still ask for config of the current call and try to merge them (preferring the inherited parameters).
To guard us from flaky tests, CI will run every test three times and fail if _any_ of the run fails.
This way we hope most flakiness will be catch before merging PR.
Configured dashboard in the same way.
Closes#8932
Now we use a bit more robust mechanism for passing dynamic configuration down the widget tree inside dropdowns, no longer relying on the `label`s used for dropdown items.
Curiously, we still need to use a hotfix implemented earlier, as we won’t have info about the currently selected item otherwise. Highlight for the currently selected item is not crucial as proper dynamic config, so we can leave with the current solution in the meantime.
No visual changes to the IDE, apart from fixed highlight for currently selected item.
# Important Notes
Target branch: #8950, for easier testing.
When using `npm run lint`, there are a lot of warnings for the last couple of weeks, and it is really annoying because I keep skipping important warnings for my files.
Introduce `SourceDocument`, a reactive source code representation that can be synced from a `MutableModule`. `SourceDocument` replaces various logic for tracking source code and spans--most importantly, `ReactiveModule`. There is no longer any reactively-tracked `Module`, per-se: Changes to the `MutableModule` attached to the synchronized `ydoc` are pushed as Y.Js events to the `SourceDocument` and `GraphDb`, which are reactively tracked. This avoids a problem in the upcoming text-synchronization (next PR) that was caused by a reactive back channel bypassing the `GraphDb` and resulting in observation of inconsistent states.
Stacked on #8956. Part of #8238.
Changes in preparation for #8238 features.
# Important Notes
Changed edit APIs:
- **`graph.astModule` is deprecated.** It will be removed in my next PR.
- Prefer `graph.edit` to start and commit an edit.
- Use `graph.startEdit` / `graph.commitEdit` if the edit can't be confined to one scope.
Probably fixes the first point in #8942
My guess is that the mockExpressionUpdate may be not set in cases when the test runs before App mounting. To make sure all the setup is done, we wait for `App` widget to being mounted before proceeding.
# Important Notes
This PR also change the way we run test's server. Before it was always a production build, but this makes development iteration long. Now we test production build only on CI or when `PROD=true` env variable is set.
Fixes#8871
The issue was caused by invalid port registration. Because of the existing context switch expression (which is not visible in GUI), the port incorrectly considered itself to belong to another node. This happened because the port was only aware of the visible part of the node’s AST and considered it the whole node.
There are two fixes in this PR. Either of them fixes the issue, and they are both implemented for robustness:
1. We provide `nodeId` information to the widget tree, so it no longer assumes the node ID from AST.
2. The order of checks in `getPortNodeId` is swapped. Now we first search by AST, only then try to look up the port. It makes sense to me because the AST is a single root of truth, and we should only rely on registered ports if AST does not exist (which happens for unconnected ports).
- Synchronize Y.Js clients by AST (implements #8237).
- Before committing an edit, insert any parentheses-nodes needed for the concrete syntax to reflect tree structure (fixes#8884).
- Move `externalId` and all node metadata into a Y.Map owned by each `Ast`. This allows including metadata changes in an edit, enables Y.Js merging of changes to different metadata fields, and will enable the use of Y.Js objects in metadata. (Implements #8804.)
### Important Notes
- Metadata is now set and retrieved through accessors on the `Ast` objects.
- Since some metadata edits need to take effect in real time (e.g. node dragging), new lower-overhead APIs (`commitDirect`, `skipTreeRepair`) are provided for careful use in certain cases.
- The client is now bundled as ESM.
- The build script cleans up git-untracked generated files in an outdated location, which fixes lint errors related to `src/generated` that may occur when switching branches.
This is a prerequisite for adding a CI action for E2E tests.
- Fix E2E tests
- Remove visual regression testing (VRT) and associated fixtures (screenshots) for now
- Switch dashboard almost fully to Vite, from ESBuild, to match GUI2's build tooling.
- Add some new E2E tests:
- Creating assets
- Deleting assets
- Creating assets from the samples on the home page
- Sort assets
- Includes fixes for sorting:
- Group sorted assets by type again (regression) (see https://github.com/enso-org/cloud-v2/issues/554)
- Make sorting by title, case insensitive. This is because it is more intuitive for non-programmers if all items with uppercase names *aren't* separated from those with lowercase names - especially since the Windows FS is case-insensitive.
- Normalization of Unicode letters is *not* currently being done. It can potentially be added later.
- Double-clicking *anywhere* on a directory row now expands it. Previously it was only being expanded when double clicking
- Add recursive label adding/removal to mirror backend
- Note: The current implementation is not exactly the same as the backend's implementation.
- Fix https://github.com/enso-org/cloud-v2/issues/872
- Unset "saved project details" (for opening the last open project) if fetching it produces an error.
# Important Notes
- All tests pass. (run `npm run test:e2e` in `app/ide-desktop/lib/dashboard`)
- All `npm` commands should be run in `app/ide-desktop/lib/dashboard`. `dashboard:*` npm scripts have been removed from `app/ide-desktop` to prevent a mess.
- `npm run dev` confirmed to still work. Note that it has not been changed as it was already using Vite.
- `npm run build` now uses `vite build`. This has been tested using a local HTTP server that supports `404.html`.
- Other cases have been tested:
- `npm run test:e2e` works (all tests pass)
- `./run ide build` works
- `./run ide watch` works
- `./run ide2 build` works
- `./run gui watch` works
Closes#8823https://github.com/enso-org/enso/assets/6566674/966576ec-6507-401c-98d3-bd71c2ffc6b2
Adds a basic text widget for text literals.
### Important Notes
Several known restrictions:
- Separators would always be replaced with single quotation marks. All types of separators in Enso are supported though, and they would be correctly escaped if needed.
- Logic for widget selection probably needs refinement (works for text literals and `Text` types, but does not work for `Text | Integer`, for example)
- **(!)** There is a very annoying issue when the input field suddenly loses focus, closing the editing mode and discarding any changes. Debugging shows that it happens when we receive an engine update (and probably recreate the node component/widget tree (???)). It requires a separate investigation.
Adds a new test scenario that checks that types are correctly set and displayed after an expression update. Includes checking the string set on the node as well as showing the type label when hovering the output port of a node.
Part of #8518Closes#8822
The drop downs were missing, because WidgetFunction had a higher score than them and took precedence. The fix was to adjust scores, so drop down is before both argument name and function. Generally, any widget expecting to "overlap" existing code (and possibly further widgets) should go before WidgetFunction.
# Important Notes
* Extended mockExpressionUpdate to accept a subexpression of a node. Also, I removed it from GraphDb - I want this function to have as little impact on app code as possible.
Introduce new AST APIs. They will be used for synchronization in the next PR.
# Important Notes
- `Module.edit` now clones a YDoc containing all the module's data; the resulting `MutableModule` can be edited with mutation APIs.
- Every AST type now has a mutable subtype. These `MutableAst` types expose all editing operations that were previously in `MutableModule`, such as `replace`; they also provide field setters.
Fixes several points from #8745
* Brought back cursor pointer in full screen visualization
* Changed behavior of edge disconnected from one side: Esc will bring the connection back, while click will remove it (and start creating new node if the source was connected).
* When hovering connection, the "active" part (i.e. the part which will be the hanging connection after click) is in normal color, and the rest is dimmed.
[Screencast from 2024-01-19 09-06-48.webm](https://github.com/enso-org/enso/assets/3919101/6df28a9c-51b3-4f98-be10-a35275eac800)