I have found a reference to the issue of QLabel resizing it's parent object here which sounds like the same problem the 3c61684 was trying to fix. If the error isn't seen any longer perhaps a Qt4 update has 'fixed' it, will have a check (+ with PyQt5, PySide, etc.).

If it's still reproducible it should be fixable by applying correct/additional size policies to the widgets.

I can't replicate the horizontal expansion bug from 3c61684 in latest PyQt5 (Qt 5.4.1) and PyQt4, PySide (Qt 4.8.6) on MacOS X. I'll check Windows/Linux tomorrow in case of platform weirdness.

An alternative to manual line-wrapping is .setWordWrap(True) on the label. In combination with setting a 12pt font this gives three lines of info in the toolbar (enough for the mode, x and y to sit in). I also tried using QFontMetrics to auto-ellipsis longer text, but that doesn't play nicely with wrapped text.

I can confirm the old expanding window bug no longer occurs on PyQt4 or PyQt5 (on Qt 4.8.6 and Qt 5.4.1) on Windows. So I think it is safe to revert this change. However, the result is still kind of ugly, so I think it might be worth trying to improve that while we're looking here. There are two options:

1. Keep the wrapped text, but with additional removal of whitespace + shrink the font so 3 lines fit in the space.
2. Set the text on a single line and elide so rather than disappearing the text is chopped(...)

The code for wrapping and stripping is as follows:

def set_message(self, s):
    self.message.emit(s)
    if self.coordinates:
        l = s.strip().split(',')
        l[-1:] = l[-1].split()
        self.locLabel.setText('\n'.join(l))

The use of .split() with no parameter allows us to drop multiple whitespace. The font setting (~line 618) is:

        f = self.locLabel.font()
        f.setPointSize(f.pointSize()-1)   # Or set to 12pt directly
        self.locLabel.setFont(f)

The result for this is:

The text-eliding option can be done with:

def set_message(self, s):
    self.message.emit(s)
    if self.coordinates:
        m = QtGui.QFontMetrics(self.locLabel.font())
        s = m.elidedText(s, QtCore.Qt.ElideRight, self.locLabel.width())
        self.locLabel.setText(s)

This looks like:

@anntzer what do you think to either of these?

I think shrinking is by far nicer than elision, but

1. on KDE, this is seems to be not enough -- the text is still higher than the toolbar. Can someone confirm this? A Windows tester would be useful too.
2. the "zoom rect" text (magnifying glass) gets incorrectly broken over two lines (most practical (short of passing lines of strings along) would probably be to use a tab internally as a separator).

@mfitzp Not sure how much work should go into this as MEP22 should completely revolutionise this, ensuring that the toolbar goes on a completely separate line so as to leave more room for tools. so I feel inclined to say that we should fix the example file itself rather than Qt.

@tacaswell can you clarify the timetable for MEP22? I.e. will the rcParam controlling its use change straight away, or make the default rcParam switch-over at a later day. If at a later date, will we actively promote that people switch to MEP22? I ask because I guess the answers will decide whether we update the embedding docs to use MEP22 now, or at that later date.

@mdboom and @efiring need to weigh in on the time line. We definitely need
the new toolbar working with all backends before it becomes the default and
should probably go through a cycle of it being optional to get more through
real-life testing before it is made the default.

I know it is annoyingly conservative, but we can't break existing user code.

On Tue, Apr 14, 2015 at 11:35 AM OceanWolf notifications@github.com wrote:

@mfitzp https://github.com/mfitzp Not sure how much work should go into
this as MEP22 should completely revolutionise this, ensuring that the
toolbar goes on a completely separate line so as to leave more room for
tools. so I feel inclined to say that we should fix the example file itself
rather than Qt.

@tacaswell https://github.com/tacaswell can you clarify the timetable
for MEP22? I.e. will the rcParam controlling its use change straight away,
or make the default rcParam switch-over at a later day. If at a later date,
will we actively promote that people switch to MEP22? I ask because I guess
the answers will decide whether we update the embedding docs to use MEP22
now, or at that later date.

—
Reply to this email directly or view it on GitHub
#4329 (comment)
.

@tacaswell No problem, just wondered when should we change the embedding examples?

For example I now see we didn't convert https://github.com/matplotlib/matplotlib/blob/master/examples/user_interfaces/embedding_in_gtk3_panzoom.py during #3652.

Anyway, from your comment I see whatever happens it won't get made default straight away, so that just leaves the question do we want to encourage people to switch early, i.e. by changing the embedding examples as the backends?

Basically I see it as, without MEP22 the embed example for QT should get rewritten to follow the pattern FigureManagerQT which works and uses a pattern 99% similar to the one that MEP22 adopts. However if it looks like the example will get rewritten again before it makes it to a release, then I would suggest saving it and only doing it the once (only color release between MEP22 release).

(P.S. Definitely agree on getting all backends converted first, but for now I work on the assumption that we will get that done for the next point release)

is this PR finished? are there outstanding issues?

I agree with @mfitzp that the result is a bit ugly but it is certainly more usable than the status quo, and less ugly than getting "zoom rect" broken into two lines.

@anntzer which option do you think we should use, do we need any changes to this PR? As I said above the whole toolbar will get a complete overhaul soon, so this will only act as a short-term fix for those using the current soon to become old toolbar...

Keep in mind "soon" will likely be the next point release after 2.0. Which
will likely be a year from now. Getting this in today if it doesn't break
anything and is an improvement is a good thing.

On Mon, Jul 13, 2015 at 4:16 PM, OceanWolf notifications@github.com wrote:

@anntzer https://github.com/anntzer which option do you think we should
use, do we need any changes to this PR? As I said above the whole toolbar
will get a complete overhaul soon, so this will only act as a short-term
fix for those using the current soon to become old toolbar...

—
Reply to this email directly or view it on GitHub
#4329 (comment)
.

We should err on the side of short term fixes here. I suspect we are not
going to get mep 27 in for 1.5 (due to failures on the review side of
things), it will go out in 2.1 in April.

On Mon, Jul 13, 2015, 4:16 PM OceanWolf notifications@github.com wrote:

@anntzer https://github.com/anntzer which option do you think we should
use, do we need any changes to this PR? As I said above the whole toolbar
will get a complete overhaul soon, so this will only act as a short-term
fix for those using the current soon to become old toolbar...

—
Reply to this email directly or view it on GitHub
#4329 (comment)
.

I'll stick with the original PR, which is less ugly imho...

Just had a thought. How does this look now with the new feature showing the image value on mouseover of an image?

Kindly asking whether this should go to 1.5 as a short term fix (perhaps it's too late now?) or just be subsumed by MEP27 (feel free to close this and reopen a similar issue on MEP27's side if this is the case).

I don't think we need to wait on MEP27 for this. Have you had a chance to investigate @WeatherGod's question?

How does this look now with the new feature showing the image value on mouseover of an image?

With the old code, the x, y and image values are still cropped and half visible below the "pan/zoom" text. With the PR, the text can sometimes be too long to fit in a single line, depending on the width of the window (I guess it was already the issue before but it didn't appear with the default size used in the embedding_in_qt4_wtoolbar example), in which case no text appears, but this can easily be fixed by making the window wider (which is easier and probably more justifiable from an UI POV than making the toolbar thicker).

Thanks!

Backported to v1.5.x as 50f373f