open-source-silicon.dev

Found a bug on cace.

When run an electrical parameter with lots of simulations, if I save the results in a `results.txt` file and then reopen it with `cace` I expect the generated data to be shown but that's not the case.
The data exists if I open the file, but `cace` is not able to recognize it.

Also, if I run another test after running the one with multiple simulations, those results will be lost when `cace` refresh the new results.

That's strange! Unfortunately, I don't have time to look into this right now because I'm working on restructuring CACE. Maybe the restructuring will fix it, if not I can take a look at it later.

<@U016EM8L91B> Can you see if this problem could be fixed manually? Maybe adding or renaming an entry on `testbench` or `results`.

<@U03B029DKU5>:  The "results" sections have only one entry "status: done", but no numerical results.  I don't know why the numerical results are not making it into the saved file.  I'll see if I can duplicate the result.

<@U03B029DKU5> (<@U02G0CLTGTH>):  I tracked down the problem to the presence of the `status: done` entries in `results`.  The `status: done` is normally associated with plots.  But apparently it fools CACE into thinking the parameter result is a plot, which prevents it from writing the numerical results to the file.

There are multiple issues here;  the main one is that the results get lost when "`status`" is present in the results dictionary.  But there are other issues, too, such as that the numerical results from the testbenches exist but are inaccessible because the "pass/fail" button isn't created.  The button provides access to the testbench data, so the button should be created when "testbenches" data exist rather than when "results" data exist.  The results can always be regenerated on demand if needed.

The root cause of this is that *any time a plot is made* the `status: done` entry is created, and then messes up everything.  Until we fix this, _don't look at any graphs of data for parameters that are not intentionally made to be plots_ between the time they are simulated and the time that the output file is made.  Run the simulations, save the resulting file, _then_ look at plots.
We'll get this fixed ASAP.

Mmm, so it's not just the .txt with a bad format that can be fixed manually... At least I can extract the values from the file and process them externally
Thanks <@U016EM8L91B> <@U02G0CLTGTH>.

<@U03B029DKU5>:  I fixed the problem but it's sitting as a pull request on the CACE repository.  I could potentially go ahead and merge it, and hopefully it won't cause Leo issues.
However, the main problem for you right now is that once you look at a plot, it overwrites the results and they are gone, and (for now) you will have to re-simulate to get them back.

I'm not in a (total) hurry so please don't merge it without leo review. I think it's not necessary to redo the simulations because I can reformat the fields to be valid in `cace`.

It looks good to me and does not interfere with my work :+1:️
It's merged now and the new version is available on PyPI. Note that the command line arguments are a little different since we switched to argparse.