bawkee / mdcai Goto Github PK

Implement the LaTex renderer. Relevant: xoofx/markdig#259

When selecting through message edits (without doing anything else in the convo), save should be triggered as with any other change.

When starting a new convo, only the first two messages get saved. To save others you'd have to go away and go back to the convo to get it saved completely.

Does not happen with existing conversations.

As the title says, instead of recovering and trying again, app throws an error when a possibly benign markdown parsing error happens.

User won't be able to proceed with the conversation and has to restart the app to continue.

Instead, app should log the error and continue trying. Sometimes, the AI will produce garbled markdown or just simply incomplete, causing some edge case bugs in the markdown renderer (which does have its own recovery but it can fail too).

When, for any reason completion generation response or stream goes wrong, an error message (sometimes cryptic) will show up and the user won't be able to proceed further until restarting the app for that given conversation.

This should be fixed such that a friendly error shows up and the user can try again, because usually these are network or load related issues.

When going to the Settings panel via the navigation pane (left bottom button), the back button should pick it up and allow the user to go back.

Instead of the right-click, a button would be ideal to sit next to the hamburger menu (in which case it has to support all 3 visual states).

If the very first conversation that the user selects contains more than 28 messages total, the app will hang and eventuall crash. The crash only happens if the "long" conversation is first to be selected. Going to the settings first for example will prevent the crash.

Workaround is to select anything else first.

NOTE: Does not happen when debugging in VS, only when running without the debugger attached (probably causes a sufficient enough delay not to trigger this).

There is a problem with WinUI cold startup layout logic causing a stack overflow in the main thread somewhere deep in the PropertyChanged handlers inside RxUI, related to the pivot view being used to switch views. The 28 messages is probably just a threshold on the given machine that makes the UI wait long enough for the bug to be triggered.

When category is deleted, it magically reappears when the app is restarted.