Steps to reproduce
- From an open note, open a new note, then another (in the same pane)
- Go back in history
The second note is displayed
Sometimes, the first note is displayed – i.e., the third note replaced the second note in history. This most often happens when I open today’s daily note for the first time, though not every day. (It’s also possible that it happens other times that I’m not noticing: the daily note scenario is one where I commonly navigate forward/back a lot and know exactly what note should appear when I go backward.)
It happens just often enough to be vexing, and has been happening since the 0.10.x series, but I never reported it because I could never figure out how to reproduce it on purpose.
Now, however, I have been able to figure out what is happening by tracing through the code in the debugger, so I can at least report the cause, as I will describe below.
- Operating system: Windows
- Obsidian version: 0.12.3
Here’s what’s happening, as far as I can tell:
- In the middle of setViewState(), after the view has updated its current file but before updateStateResult() is invoked, the 1-second setInterval for updateHistory fires, requesting a history save for the active leaf.
- updateHistory calls recordHistory with a flag indicating the current state should be replaced – which causes the presence of note 2 to be deleted from history, replaced with an entry for note 3
- setViewState resumes execution, and eventually gets around to calling updateStateResult(), which calls recordHistory with a flag indicating a new state should be added to the history for note 3
- recordHistory() detects that the state to be added to history is identical to the current state in history (saved in step 2), and does nothing. The net result is that note 3 has replaced note 2 in the history, leaving a history of [note 1, note 3] instead of [note 1, note 2, note 3] as should be the case.
I’m planning to add a workaround for this in the pane-relief plugin, by turning a replaceState into a pushState if an attempt is made to change the current file for the active leaf. But it would probably be a good idea for this to be fixed in Obsidian as well.
ISTM that one way to do that would be to perform a pushState at the beginning of setViewState, before any
await can happen. The state to be pushed is known, after all, as is whether the push is needed. At the end of the operation, a request to replaceState could then be issued, which would fix any discrepancy between the state passed to setViewState and the view’s final state. And there’d be no way for the setInterval of updateHistory to interfere, since it would just update the already-pushed state, and the final replaceState would be a no-op, as it is now.