Force Qt validator to use C locale.

[anntzer]

PR for #6082: Cannot interactively edit axes limits using Qt5 backend

I found that this was a locale issue: I am on a French locale, where the decimal separator is ",".
On Qt4 (http://doc.qt.io/qt-4.8/qdoublevalidator.html),

QDoubleValidator uses its locale() to interpret the number. For example, in the German locale, "1,234" will be accepted as the fractional number 1.234. In Arabic locales, QDoubleValidator will accept Arabic digits.
In addition, QDoubleValidator is always guaranteed to accept a number formatted according to the "C" locale.

On Qt5 (http://doc.qt.io/qt-5/qdoublevalidator.html), the second provision has been removed.

Thus, on the French locale, "1.23" is considered an invalid input and rejected by QDoubleValidator. In fact, I can input instead "1,23", but this causes an exception later when Python tries to convert this to a float. (To reproduce this, set the locale appropriately and edit the axes parameters of a default plot.)

Given that matplotlib is not really localized anyways (except for axes.formatter.use_locale), this patch sets the validator's locale to C in all cases. Note that despite what the Qt docs indicate, this does not seem to prevent input of localized ("1,23") values, which still lead to an exception.

[mdboom]

Should we do this for ints as well?

Also, are there issues for dates? There, I think ISO dates is probably the only thing that's internationally understood.

[anntzer]

There isn't anywhere in the UI right now where an int is required as input (AFAICT) but I guess that if there was then something bad could happen with the use of "," as thousands separator in English.
Dates seem completely not handled by the property editor right now, say e.g. examples/api/date_demo.py: it would be nice if the editor provided a calendar widget or similar to input the x limits, but right now it just gives something that looks like a number of days since year 0.

[anntzer]

Any chance this can be backported to 1.5.x? It's a pretty annoying issue...

[tacaswell]

backported to v1.5.x as c1cd4f2