libpng error when plotting in notebook after importing tensorflow

[bnaul]

Environment info

Operating System: CentOS Linux 7 x86_64

Tensorflow version

0.8.0rc0 (from binary)

Steps to reproduce

1. Fresh conda environment: conda create -y -n plt_tf python=3.4 jupyter notebook matplotlib numpy && source activate plt_tf
2. pip install --upgrade https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.8.0rc0-cp34-cp34m-linux_x86_64.whl
3. Launch a notebook server and execute the following:

import tensorflow as tf
import matplotlib.pyplot as plt
%matplotlib inline

plt.plot([1,2], [1,2])

Logs or other output that would be helpful

Traceback:

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
/home/bnaul/miniconda3/envs/plt_tf/lib/python3.5/site-packages/IPython/core/formatters.py in __call__(self, obj)
    337                 pass
    338             else:
--> 339                 return printer(obj)
    340             # Finally look for special method names
    341             method = _safe_get_formatter_method(obj, self.print_method)

/home/bnaul/miniconda3/envs/plt_tf/lib/python3.5/site-packages/IPython/core/pylabtools.py in <lambda>(fig)
    224 
    225     if 'png' in formats:
--> 226         png_formatter.for_type(Figure, lambda fig: print_figure(fig, 'png', **kwargs))
    227     if 'retina' in formats or 'png2x' in formats:
    228         png_formatter.for_type(Figure, lambda fig: retina_figure(fig, **kwargs))

/home/bnaul/miniconda3/envs/plt_tf/lib/python3.5/site-packages/IPython/core/pylabtools.py in print_figure(fig, fmt, bbox_inches, **kwargs)
    115 
    116     bytes_io = BytesIO()
--> 117     fig.canvas.print_figure(bytes_io, **kw)
    118     data = bytes_io.getvalue()
    119     if fmt == 'svg':

/home/bnaul/miniconda3/envs/plt_tf/lib/python3.5/site-packages/matplotlib/backend_bases.py in print_figure(self, filename, dpi, facecolor, edgecolor, orientation, format, **kwargs)
   2178                     orientation=orientation,
   2179                     dryrun=True,
-> 2180                     **kwargs)
   2181                 renderer = self.figure._cachedRenderer
   2182                 bbox_inches = self.figure.get_tightbbox(renderer)

/home/bnaul/miniconda3/envs/plt_tf/lib/python3.5/site-packages/matplotlib/backends/backend_agg.py in print_png(self, filename_or_obj, *args, **kwargs)
    535             close = False
    536         try:
--> 537             _png.write_png(renderer._renderer, filename_or_obj, self.figure.dpi)
    538         finally:
    539             if close:

RuntimeError: Could not create write struct

Server log:

libpng warning: Application was compiled with png.h from libpng-1.6.17
libpng warning: Application  is  running with png.c from libpng-1.2.53
libpng error: Incompatible libpng version in application and library

Workaround

Importing tensorflow after matplotlib prevents this from happening.

[bnaul]

Similar to #1924?

[mrry]

Since this looks like the same issue as #1924, let's close this issue for now and continue the discussion over there.

[JensenQi]

put the "import pylab" before "import tensorflow as tf" and it may help
so, the code likes this

import pylab
import tensorflow as tf

blabla....