yangzhang75 opened a new issue, #8606:
URL: https://github.com/apache/texera/issues/8606
### Task Summary
`WorkflowEditorComponent` finds the element it builds its JointJS paper into
with a document-wide lookup:
```ts
// workflow-editor.component.ts, ngAfterViewInit
this.editor = document.getElementById("workflow-editor")!;
this.editorWrapper = document.getElementById("workflow-editor-wrapper")!;
...
this.paper = this.wrapper.attachMainJointPaper({ el: this.editor, ... });
```
The ids come from the component's own template (`<div
id="workflow-editor-wrapper"><div id="workflow-editor">`), so every instance of
this component renders elements carrying them. The lookup therefore does not
return *this* instance's container; it returns whichever is first in document
order.
That is fine as long as only one instance is ever in the page, which is the
case today. It stops being fine the moment two are, and two views already mount
this same component: the operator canvas mounts it as the canvas, and the Form
View mounts it as the read-only workflow preview.
**Consequence.** With the canvas <-> Form View switch routed rather than
reloaded (#8580), the two instances overlap for one tick: the arriving view
runs `ngAfterViewInit` while the departing view's DOM is still attached. The
arriving canvas then builds its paper into the *departing* view's container,
which is removed moments later. The canvas is left with an empty `<div
id="workflow-editor">`: no SVG, no cells, nothing to pan and nothing to click,
while the graph itself is untouched -- the Form View's preview goes on showing
it, including a run in progress, right up until the switch.
Measured in a browser at the moment the canvas comes back, with the switch
routed:
```
[GETBYID] workflow-editor: 2 in document, returned index 0
#workflow-editor box=1399x1000 svg=false cells=0
```
after resolving the container from the component's own host:
```
#workflow-editor box=1399x1000 svg=true cells=1
```
No run is needed: expanding the Form View's preview once and switching back
is enough. Watching a run in the preview simply makes it certain, because that
is when a reader expands the preview.
**Not reproducible on `main` as it stands.** The switch is still a full page
load there, so the two instances never coexist and the lookup is always right.
This is a latent defect that any in-process switch turns into a blank canvas,
and it is wrong on its own terms regardless: a component should not reach
outside itself for its own element, and two elements sharing an id in one
document is invalid HTML.
Proposed: resolve both elements from the component's own host (`ElementRef`,
already injected) instead of from `document`, with a test that a second
instance created while a decoy `#workflow-editor` sits earlier in the document
still builds its paper in its own container.
Two more document-wide lookups of the same id exist and are worth a look
while this is open, though neither leaves a stuck state:
`MiniMapComponent.updateNavigator` reads it to place the navigator rectangle
(self-corrects on the next pointer event) and `ReportGenerationService` reads
it to snapshot the canvas.
Found while browser-testing #8581.
### Task Type
- [x] Bug Fix
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
