Log scale tick labels are overlapping

[mxmlnkn]

• Linux hypatia-pc 4.5.0-1-amd64 Fix autofmt_xdate() when using in conjunction with twinx() #1 SMP Debian 4.5.1-1 (2016-04-14) x86_64 GNU/Linux

  lsb_release -a
      No LSB modules are available.
      Distributor ID: Debian
      Description:    Debian GNU/Linux 8.4 (jessie)
      Release:    8.4
      Codename:   jessie
  

• matplotlib 1.5.2~rc2-1

• Python 2.7.11+

• installed using: sudo apt-get install -t sid python matplotlibg

Here is an example where the problem appears:

from matplotlib.pyplot import *
figure( figsize=[4,3] )
subplot( 111, xscale='log' )
plot( [1e-20,1e5], [1e-16,10] )
show()

For lack of time the following image wasn't reproduced with the above example, but came from another problem:

The right side shows the problem and the left side is my workaround which calculates some distance for a given amount of labels I want (here I wanted 5 labels or less). So it set every exponent, but every sixth integer exponent, i.e. -16,-10,-4.

It would be cool if it was fully automatic like with linear scales.

[mxmlnkn]

This bug also happens with the linear scale:

from matplotlib.pyplot import *
figure( figsize=[4,3] )
plot( [0.1310,0.1315], [0,1] )
tight_layout()
show()

Note that without tight_layout the +1.309e1 won't be inside the canvas:

I would say this is still another bug.

I would just be nice if the formatter could do something like:

while exists i: labelpos[i] + width( text(label[i]) ) + spacing >= labelpos[i+1]:
    reduce number of labels by one

[tacaswell]

Can you try the 2.0.0b1 which implements something like that?

[anntzer]

2.0 solves the problem for linear scale, but not for log-scale (the original issue): the LogLocator class does not use the ticks-limiting logic present in MaxNLocator. (Apparently neither do SymmetricalLogLocator or LogitLocator.)

[mxmlnkn]

@tacaswell : The only difference in 2.0b is that the default label font size is smaller (or maybe it just uses a different backend), resulting in the above linear scale example to actually work. But adding some 0s, I can still reproduce the problem even with linear scales, unlike @anntzer said.

logarithmic scale (minimal example from 1st post):

linear scale (minimal example from 2nd post):

linear scale even more extreme (This may be an extreme example, but it could still be solved by reducing the number of tick labels by one):

Above done with:

from matplotlib.pyplot import *
figure( figsize=[4,3] )
plot( [0.131000,0.131005], [0,1] )
tight_layout()
show()

print matplotlib.__version__
    2.0.0b1

[anntzer]

Indeed, 2.0 just uses a heuristic where the aspect ratio of an xlabel is no more than 1x3 (by the way, I just realized that this would be way too pessimistic (too optimistic for ylabels) if the labels are rotated...).

As discussed in #5494, it is not so easy to handle this automatically because the tick labels are rendered after the tick positions are selected.

[tacaswell]

Linked with #7001