After application of Daishi Harada's patch on 0.85, I tried to use it with
SciPy core SVN from yesterday and get rather strange results:

x = scipy.array( [16.5]*10 )
y = scipy.array( [19.5]*10 )
w = scipy.arange(10)

# This plots a line at 16.0, not 16.5!
pylab.plot( w, x )

# However this works perfectly with a blue box 0-9 x 16.5x19.5
f = pylab.figure()
a = f.add_subplot( 1, 1, 1 )
a.fill( scipy.concatenate((w,w[::-1])), scipy.concatenate((x,y[::-1])) )
f.canvas.draw()

Further, the legend boxes are too large, usually obscuring most of the picture
with a polygonal patch (such as the one produced by the fill above) or at
least a half without any polygonal patches.

I'm sorry you're having troubles with the patch.
Unfortunately, I can't seem to recreate your problem.
I realize "works for me" isn't a particularly useful
response, but I'm afraid that's the best I can do
for now - and I'll be away again for Thanksgiving
until next week.

FWIW, I'm using the CVS matplotlib with the wx backend,
and the new scipy w/atlas.

d

···

On Nov 23, 2005, at 12:20 PM, Ravikiran Rajagopal wrote:

After application of Daishi Harada's patch on 0.85, I tried to use it with
SciPy core SVN from yesterday and get rather strange results:

The problem is that I have both Numeric[1] and the new scipy installed. When
both are installed, the Numeric headers are picked up by default during scipy
compilation. By forcing them to pick up the scipy headers, the problem was
resolved. But it is a rather strange coincidence that the error results in
truncated numbers rather than something drastic.

Ravi

[1] Reason: It will be a while before all my Numeric-based libraries are
ported to the new scipy; porting is trivial, but regression testing takes
time, especially given that the new scipy C API is not quite stable yet and
that I haven't yet ported boost.python.numeric to scipy.

···

On Wednesday 23 November 2005 19:51, daishi@...43... wrote:

Unfortunately, I can't seem to recreate your problem.