FIX: make set_text(None) keep string empty instead of "None"

[jklymak]

PR Summary

As documented in #10391, legend.set_title() doesn't do sensible things with the default title=None (i.e it set the string of the legend title to the string "None").

This is kind of an API change in that if anyone was checking for the string "None" to see if the legend title was empty will now fail, but lets hope no one was doing that.

As pointed out below by @anntzer the problem is more wide spread: The updated PR now checks if s is None in Text.set_text and if it is does not do anything. Before it would set the string to "None" which is silly.

One could go further and insist that s is a string (so the user knows what they are doing), but that'll likely break lots of people assuming the string cast happens.

This still needs a test....

PR Checklist

• Has Pytest style unit tests
• Code is PEP 8 compliant
• API change noted

[anntzer]

There's a bunch of places where passing in None is interpreted as the string "None" (e.g. xlabel, title) and I would rather have them all fixed at once rather than in a piecemeal fashion.

[jklymak]

Yeah, OK, fair enough.

I'd propse in text.py:

def set_text(self, s):
        """
        Set the text string *s*

        It may contain newlines (``\\n``) or math in LaTeX syntax.

        ACCEPTS: string or anything printable with '%s' conversion.
        """
        if s is not None:
            self._text = '%s' % (s,)
            self.stale = True

[anntzer]

probably an API change note

[anntzer]

shouldn't you always clear the text (and mark as stale) when s is None?

[jklymak]

Oh, good question! Open to suggestions

[jklymak]

Set to empty string if None is passed.

[timhoffm]

There's a lot to say here:

• IMO '%s' % (s,) is an anti-pattern. It is identical to str(s).
• self._text is not defined in __init__. It only calls set_text(text). Something like Text(text=None).get_text() would fail with an attibute error in your code. It's good practice to declare your attributes in init. You should add self._text = '' to init.
• I would expect that set_text(None) and set_text('') are completely equivalent.
• Minor thing: Add a colon at the end of the sentence in the docstring.

So either always stale:

def set_text(s):
    self._text = str(s) if s is not None else ''
    self.stale = True

or stale only when there's really a change:

def set_text(s):
    if s is None:
        s = ''
    if s != self._text:
        self._text = s
        self.stale = True

[jklymak]

Great catch. You're right I didn't notice that self._text wasn't being set.

[jklymak]

@timhoffm not sure where you want the colon...

[timhoffm]

Sorry, full stop: "Set the text string s."

[jklymak]

Great. Fixed. Also, your logic was cleaner, so used that (second option).

[timhoffm]

assert not leg.get_title().get_text()

[QuLogic]

I don't think we should rely on boolean-ness here.

[timhoffm]

This is PEP-8 convention

For sequences, (strings, lists, tuples), use the fact that empty sequences are false.

But I'm ok, if you want to have an explicit comparison.

[timhoffm]

"anything printable with '%s' conversion" is technically correct, but feels a bit beside the point. This has nothing to do with printing. I don't really know a good alternative.

According to __str__ docs "anything with an "informal" string representation" would be right, but I'm afraid that users will not understand this.

Maybe "anything that can be formatted by '%s'" or "anything that can be converted by '%s'"?

[anntzer]

Actually any object has a __str__... (yes, you can technically make __str__ raise an exception but whatever).
The test should more be something like "any object, will be cast to a string using str" (don't care much for Py2 now especially if this goes to mpl3.0...).

[timhoffm]

which is equivalent to an empty string.

[jklymak]

@timhoffm python 2.7 gave unicode errors for str(s) instead of '%s' % s. Since I'd rather gouge my eyes out than mess around with py2/3 differences in unicode support, I reverted that suggestion. If it drives you nuts, a) feel free to suggest a fix or b) keep it in mind for 3.0.

[timhoffm]

@jklymak Ok, leaving the str(s) to py3 only then.

[jklymak]

@tacaswell so for 3.0 milestones PRs can we ignore failing 2.7 tests? If so, then I can implement @timhoffm's str(s) suggestion.

[timhoffm]

Approved, whatever the decision on str(s) will be.

[jklymak]

Thanks for your help @timhoffm

[jklymak]

OK, have gone back to the cast if this is milestones 3.0. Should pass the 3.x tests. Modified the docstring as per @anntzer comment above.

[phobson]

Looks great. I'll merge pending successful CI

[jklymak]

Just to be clear, 2.7 tests/doc build will fail. But if this is 3.0 those tests should be ignored, right @tacaswell?

[phobson]

Circle CI failure seems unrelated:

Encoding error:
'ascii' codec can't encode character u'\u2212' in position 0: ordinal not in range(128)
The full traceback has been saved in /tmp/sphinx-err-y7dr1d.log, if you want to report the issue to the developers.
Makefile:20: recipe for target 'html' failed
make: *** [html] Error 1
Exited with code 2

u'\u2212' is the "minus sign"

[jklymak]

it’s definitley because set_text calls str(s) now. If I take that out the docs build.

[story645]

@jklymak technically cat.to_str(value) was basically designed to iron out Py2/Py3 string differences particularly for labeling (I only call it in the Formmater). It's messy, and should probably be in cbook, but it'd possibly work.

[jklymak]

Well, I'll just leave it the way it was before I touched anything.... We can revisit for python 3...

[efiring]

This seemed to be ready to go, so I used "squash and merge". Let's see if that broke anything.

[RpiController]

Could someone also check to make sure that ax.get_legend().get_visible() is working properly?
It is always returning True for me.

[jklymak]

What does working properly mean in this context? How do you expect it to behave and why?

[RpiController]

After executing ax.get_legend().set_visible(False) I would like ax.get_legend().get_visible() to return False.
That seems like it would be proper to me. Currently it is always returning True no matter what.

Another issue that I think needs to be fixed is that after setting ax.get_legend().set_visible(False) and then back True again, all the specifics of my legend are lost (legend position, marker size, etc)

[jklymak]

Can you provide a MSCE that has this problem? I can't reproduce. If you do provide an example, please open a new issue...

If I do:

import matplotlib.pyplot as plt

fig, ax = plt.subplots()

ax.plot(range(10), label='Boo')
leg = ax.legend()
print(leg)
ax.get_legend().set_visible(False)
print(ax.get_legend().get_visible())
ax.get_legend().set_visible(True)
print(ax.get_legend().get_visible())
plt.show()

I get

Legend
False
True

and the legend appears as expected... If I don't set_visible(True) it remains hidden.

[RpiController]

I'm not using plt since my plot is embedded in Tkinter. My code is pretty complicated, it's python 2.7 on a RaspberryPi with matplotlib 2.1. I was hoping someone else would have run into the same issue and found an easy fix or workaround. If not, I will just live with it.

The other issue of losing the legend parameters between making visible and invisible seems like a more serious flaw but I don't really want to go through the hassle of simplifying my code to provide a MSCE and doing whatever is needed to open a new issue. I'm kind of a newb on here.

This is just my good deed of bringing it to someone's attention.

[jklymak]

The other issue of losing the legend parameters between making visible and invisible seems like a more serious flaw but I don't really want to go through the hassle of simplifying my code to provide a MSCE and doing whatever is needed to open a new issue. I'm kind of a newb on here.

The example above doesn't have that behaviour, so its pretty hard to chase down what the issue might be that you are having. If you figure it out, let us know.