matplotlib-users

[Matplotlib-users] bug in saving figure with dashed lines

[Eric E. <ems...@ob...>]

Hi,
I just produced a plot with a curve dashed line:

when saving it into a png file it works fine.
But when saving it in a postscript the dashed line looks solid on most
of the plot.
This happens if there are too many points in the line:

x = arange(0,1.,0.0001)
y = sqrt(x)
plot(x,y,'k--')
## Png file is ok
savefig("toto.png")
## Eps file is NOT ok
savefig("toto.eps")

Any cure to that behaviour?

Thanks!!!
Eric

Re: [Matplotlib-users] bug in saving figure with dashed lines

[Jouni K S. <jk...@ik...>]

Eric Emsellem <ems...@ob...> writes:

> But when saving it in a postscript the dashed line looks solid on most
> of the plot.
> This happens if there are too many points in the line:

The draw_lines method in the PS backend divides the line into
subsequences of at most 50 points, and calls the "stroke" operator for
each subsequence, which in effect resets the dash offset. Thus if 50
consecutive points fit within a dash, the line appears solid.

I suppose the division into subsequences solves some memory or speed
problem with PS interpreters, so it should be retained. Then the
solution would be to compute the length X of each subsequence and do
something like "currentdash pop X setdash" after the "stroke".

-- 
Jouni

Re: [Matplotlib-users] bug in saving figure with dashed lines

[John H. <jdh...@ac...>]

>>>>> "Jouni" == Jouni K Seppanen <jk...@ik...> writes:
    Jouni> The draw_lines method in the PS backend divides the line
    Jouni> into subsequences of at most 50 points, and calls the
    Jouni> "stroke" operator for each subsequence, which in effect
    Jouni> resets the dash offset. Thus if 50 consecutive points fit
    Jouni> within a dash, the line appears solid.

50 seems awfully small to me.  There are maximum length path
limitations of some postscript interpreters which is why we batch
these up.  Darren, could we get away with a step of 1000 here rather
than 50?

JDH

Re: [Matplotlib-users] bug in saving figure with dashed lines

[Darren D. <dd...@co...>]

On Tuesday 01 August 2006 08:52, John Hunter wrote:
> >>>>> "Jouni" == Jouni K Seppanen <jk...@ik...> writes:
>
>     Jouni> The draw_lines method in the PS backend divides the line
>     Jouni> into subsequences of at most 50 points, and calls the
>     Jouni> "stroke" operator for each subsequence, which in effect
>     Jouni> resets the dash offset. Thus if 50 consecutive points fit
>     Jouni> within a dash, the line appears solid.
>
> 50 seems awfully small to me.  There are maximum length path
> limitations of some postscript interpreters which is why we batch
> these up.  Darren, could we get away with a step of 1000 here rather
> than 50?

Here are GGV's render times for a line containing 1e6 points:

division | time (s)
50		3
1000	9
100000	8
none	28

I bumped it up to 100,000 in svn 2644, which seems to have improved the dash 
problem.

Darren