REF: lazy load all tools and hopefully improve performance

[klauer]

Description

• All suite tools are now lazily loaded
• Miscellaneous performance fixes as I found that:
  • We call load_best_template twice (due to a setattr in a property and then an explicit call)
  • We call setVisible on displays and then move them into a new layout - effectively doing a show/hide/show event under the hood
• Preliminary results indicate a pretty big speed-up, but I'd like to run more tests before making any claims about percentages

Motivation and Context

• Speed is important
• Tools are rarely used but heavy

How Has This Been Tested?

• Interactively
• TODO: based on another branch because I was trying to fix that and lazily did not check out master again

Where Has This Been Documented?

This PR text

Screenshots (if appropriate):

[klauer]

There may be some bugs introduced - or made easier to hit - with this branch. Need to investigate:

$ devtyphos at2l0
[2022-05-26 11:59:51] - INFO - Loading 'at2l0' ...
[2022-05-26 11:59:55] - INFO - Loading Tools ...
[2022-05-26 11:59:56] - INFO - Adding devices ...
Segmentation fault

[ZLLentz]

We call load_best_template twice (due to a setattr in a property and then an explicit call)

This explains A LOT of the oddities I was seeing a while back while fixing other internals

[ZLLentz]

The code here looks good/normal- should I give this a test drive combined with your pydm PR?

[klauer]

The PyDM PR is only necessary if you want to run the profiler, of course. I think it'd be a good idea to have you take a look at as well, if you can spare the time 👍

I'd be curious if you can reproduce any startup core dumps, as now that I'm trying to cause them I'm failing (of course!).

[ZLLentz]

I'm not able to produce any issues at all with this so far This definitely makes the 1-device suites feel noticeably faster to load, the multi-device suites still feel pretty sluggish

[ZLLentz]

(Of course, I've been running with your pydm patch the entire time RE: startup core dumps)

[klauer]

Two additional commits here - relatively minor, and one could be removed after a discussion.

The --exit-after was added so I could test typhos initialization in a loop and try to get it to crash

[ZLLentz]

I think it's ok to leave --exit-after in- might replace it later if we get a better metric for when to close the screen for benchmarking