[GUI2] Undoing a node deletion breaks AST ID assignment.

[Frizi]

After deleting a node and hitting undo, the node is no longer being evaluated. Reloading the project reveals, that the node metadata was also not properly saved. This is caused by the ID map is not being properly undone. The behavior observed so far:

• Multiple AST IDs ends up being assigned to the same range, and ranges generally are a little bit incorrect for the undone text ranges
• The affected node's expression AST ID (used for node ID) is not assigned properly, but somehow the local client still recognizes the node metadata unitl page refresh.

It appears that the Yjs Relative Positions don't behave as expected after an undo operation, since their representation in the undone transaction is not accurately representing the state after that undo is applied. Understanding that behavior needs more testing.

[enso-bot[bot]]

Keziah Wesley reports a new STANDUP for today (2023-10-26):

Progress: Finished #8073 in the morning. Debugging undo/IDs today. It should be finished by 2023-10-30.

Next Day: Next day I will be working on the #8134 task. Continue debugging and adding tests.

[kazcw]

The conflict resolution of Y.Text can break syntax trees.

• This can't easily be detected and repaired.
• To ensure conflict behavior is reasonable, we should change the representation we synchronize.
• Although sync is not a priority now, the canonical representation is a fundamental design choice, easier to change sooner than later.

An important design choice is the mapping between AST and text.

• The parser's current Spans cannot efficiently support edits.
• Y.Text has been providing a mutation-aware mapping in the form of RelativeOffsets.
  • They aren't perfect; they don't interact well with undo/redo.
  • But they are why we hadn't needed mutation-friendly source-code references yet.
• We should add edit-friendly references to Spans.

The new representation needs to support read/write via AST changes, the text editor, and external file changes; as well as correct conflict resolution, and undo/redo.

• We can achieve this by using block structure as the Rosetta stone.
  • codemirror supports line:column text locations natively.
  • The parser can include line:column information in its source code references.
  • We can synchronize a module as a sea of lines.

Sea of lines:

• A Y.Array of lines, but we don't care about the array's order.
• Each line has an ID, a parent, siblings, and Y.Text data.
• The only conflicts that aren't automatically resolved well are ordering, but we can recover:
  • If we end up with a dead reference within a block, we still know its position because we have before+after, so we can repair the links.
  • If two nodes are placed with the same before/after, we know they don't reference each other, so we just need to choose any relative order and repair the links.

[enso-bot[bot]]

Keziah Wesley reports a new STANDUP for the last Friday (2023-10-27):

Progress: Designing synchronization representation and new spans. It should be finished by 2023-10-30.

Next Day: Next day I will be working on the #8134 task. Implementing.

[enso-bot[bot]]

Keziah Wesley reports a new STANDUP for yesterday (2023-10-30):

Progress: Adding line:col to spans in parser, and refactoring parser to fit the more restrictive arithmetic properties of this metric. It should be finished by 2023-10-30.

Next Day: Next day I will be working on the #8134 task. Finish Rust parser changes

[enso-bot[bot]]

Keziah Wesley reports a new 🔴 DELAY for yesterday (2023-10-31):

Summary: There is 4 days delay in implementation of the [GUI2] Undoing a node deletion breaks AST ID assignment. (#8134) task. It will cause 0 days delay for the delivery of this weekly plan.

Delay Cause: Solution requires parser changes and replacing a complex component of the GUI.

[enso-bot[bot]]

Keziah Wesley reports a new STANDUP for yesterday (2023-10-31):

Progress: Finished implementing Rust parser changes. It should be finished by 2023-11-03.

Next Day: Next day I will be working on the #8134 task. Start implementing planned typescript datastructure change.

[enso-bot[bot]]

Keziah Wesley reports a new STANDUP for today (2023-11-01):

Progress: Implementing new edit/sync logic. It should be finished by 2023-11-03.

Next Day: Next day I will be working on the #8134 task. Continue work on new edit/sync.

[enso-bot[bot]]

Keziah Wesley reports a new STANDUP for yesterday (2023-11-02):

Progress: Refactoring APIs, decided we need deeper design changes. It should be finished by 2023-11-03.

Next Day: Next day I will be working on the #8134 task. Designing tree-based sync.

[enso-bot[bot]]

Keziah Wesley reports a new STANDUP for the last Friday (2023-11-03):

Progress: Discussed sync strategy. Designing node representation and edit process. It should be finished by 2023-11-03.

Next Day: Next day I will be working on the #8134 task. Finish design doc, start implementing.

[farmaazon]

There is no longer a problem with deleted node being not evaluated... but metadata are still gone. But that is now tracked by #9314