Closed patrick-nicholson closed 7 years ago
Probably due to the ordering of the messages. The kernel has to send the idle status message after sending all of the output to the notebook frontend.
Is that a metakernel issue then?
I think the async file reader code is a more likely suspect. It's happening asynchronously from the execute request completing.
While toying with the fix for #26, I confirmed this is indeed due to iopub messages with display_output / stream output arriving after the kernel goes idle for the cell code that generated them. Will need to work on a synchronization strategy to fix.
When executing many cells at once, the occasional long-running (by execution time) cell has its output delivered to the next cell. This seems like it happens most often with minute-runtime-level cell followed by a millisecond-runtime-level cell.
I am not able to reproduce an example on demand, it just happens when it happens.