Fix logging and notifyOnLogging to work with Plotly.newPlot calls

[archmoj]

Fixes #4555 thanks to #4555 (comment).

TODO:

• test this from python
• add a jasmine test

@plotly/plotly_js

[nicolaskruchten]

If we're concerned about this refactoring shaking something loose at runtime that causes silent/subtle failures, could we try adding some extra runtime checks at the top of all the function definitions that we're changing which error out if the first argument is not as expected, and then run the test suite? That might flush out some extra silent failures? We wouldn't commit this code into master but as an extra one-time safety check?

Otherwise, would this be a safer change if we added an argument instead of changing the first one?

[archmoj]

If we're concerned about this refactoring shaking something loose at runtime that causes silent/subtle failures, could we try adding some extra runtime checks at the top of all the function definitions that we're changing which error out if the first argument is not as expected, and then run the test suite? That might flush out some extra silent failures? We wouldn't commit this code into master but as an extra one-time safety check?

Otherwise, would this be a safer change if we added an argument instead of changing the first one?

@nicolaskruchten thanks very much for the feedback.

Adding this new gd parameter to the end of function arguments is not very helpful; while those functions may be called without full set of arguments.

On the other hand checking the first argument in the test looks like a good idea; although in the end we may not catch anything new since those tests are passing already. Anyway I could give it a try.

[archmoj]

Notes by @alexcjohnson:

I’m a bit uncomfortable with the logging PR - seems like it’s passing gd essentially everywhere, when there are really only a few places we call log/warn/error.

One thing we could do to make it easier is if you have access to any DOM node, allow the log functions to search its parents for a _context

And some of them, like randstr, we could probably just get rid of the log message, turn it into throwing an error.

[archmoj]

Notes by @alexcjohnson:

And some of them, like randstr, we could probably just get rid of the log message, turn it into throwing an error.

Good call. Done in bddd218.

[archmoj]

Notes by @alexcjohnson:

One thing we could do to make it easier is if you have access to any DOM node, allow the log functions to search its parents

Searching DOM elements for gd is little hacky (in my opinion) and could add to the complexity of this PR.