Issue with feedback lines

[swilliamson7]

@wsmoses This is the bug we saw when I include the feedback lines which effectively just share things such as model integration time, NaN checks, etc.

feedback_error.txt

[wsmoses]

Yeah this one requires some minimization, if possible. Can you also post the source code reproducer?

…

On Mon, Sep 11, 2023 at 1:26 PM Sarah Williamson ***@***.***> wrote: @wsmoses <https://github.com/wsmoses> This is the bug we saw when I include the feedback lines which effectively just share things such as model integration time, NaN checks, etc. feedback_error.txt <https://github.com/EnzymeAD/Enzyme.jl/files/12579250/feedback_error.txt> — Reply to this email directly, view it on GitHub <#1045>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAJTUXGCHWGMFEIHQUO3PPDXZ5JTXANCNFSM6AAAAAA4TXRBHY> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[swilliamson7]

I'll need to check if I have a standalone script that can produce this, but at the very least can show you the code tomorrow otherwise? I'll work on making a file I can post here in any case (assuming I don't have one)

[wsmoses]

@swilliamson7 any chance you have a reproducer for this?

[swilliamson7]

I'll create a minimal example

[wsmoses]

bumping this @swilliamson7

[swilliamson7]

Sorry for the delay! I minimized a bit, and can try working on it more later

Running this script will trigger Enzyme bugs seemingly related to lines that just provide feedback on the integration or initialize netcdf output

[wsmoses]

@swilliamson7 does this still err?

[swilliamson7]

Last I checked yes

[wsmoses]

@swilliamson7 I don't see a file at the linked script. Do you have something that still errs [we also since fixed a bunch of things]

[swilliamson7]

Yeah things have been moved around since May. I think I still get this bug, but have just been omitting the lines of code that cause it. We can either (a) leave this open and in the coming days I'll create a new script that leads to it or (b) close this issue and I'll reopen whenever I get around to looking at it more