Require numpy 1 due to incompatibility with Streamlit

[tkoskela]

Update bokeh version to work with latest numpy

Closes #334

[giordano]

Looks good to me, but I don't know why we were pinning an exact (and quite old) version. Latest version at the moment is 3.5.0.

[tkoskela]

I actually meant to use >= not ==, let me fix

[tkoskela]

Ah, there was a reason for that specific version. This could be a problem

StreamlitAPIException: Streamlit only supports Bokeh version 2.4.3, but you have version 3.5.0 installed. Please run pip install --force-reinstall --no-deps bokeh==2.4.3 to install the correct version.

[pineapple-cat]

Yes, we need to use Bokeh v2.4.3 so it doesn't break Streamlit. Can we just cap the numpy version below 2.0.0 for now?

[pineapple-cat]

I think the test failure that's happening now is the same one as in the line plot PR, but I can't quite tell what's causing it at a glance. Maybe there's another library we need to version cap – pandas perhaps?

[tkoskela]

It's failing on

    assert all(c in EXPECTED_FIELDS for c in df.columns.tolist())

On a quick look through the perflog the fields look ok. Maybe something has changed in the spac spec dict?

[pineapple-cat]

If it helps, I think the change causing this may have taken place 28 - 29 May. Using the line plot PR as reference again, all tests were passing on the 28th, but when I pushed some inconsequential changes on the 29th this specific test started failing.

[tkoskela]

The field version has changed to reframe version (I assume in a new version of ReFrame). I'm running 4.6.2. I updated it in the test and now it passes. I noticed we don't have whitespace in other field names. Is that going to cause any issues @pineapple-cat?

Edit: I should probably update the required version of ReFrame with this.

[pineapple-cat]

Of course, I forgot about ReFrame field name updates; thank you for finding the issue.

[pineapple-cat]

The presence of whitespace in "reframe version" bothers me on a psychological level, but I don't think there's anything in our config handler that explicitly can't handle it. We may want to replace all whitespace with an underscore to ensure consistency in label handling for later stages, but it's not strictly necessary. I just ran some tests with whitespace in field names and the resulting graphs look fine.

[tkoskela]

This change seems to have happened in ReFrame 4.6.1. Are we happy to update the requirement to that?

[giordano]

I think so. Better to be explicit about the actually supported versions.