Thanx for this How To! I'd been using an external method to gain access to Matplotlib and I knew I had to shove it into AT. Here are the modifications I did to it to integrate matplotlib. Four comments :

1) AFAIK, matplotlib's savefig does not support writing to a StringIO object ;( You have to use a temporary file.

2) Instead of resetting the mimetype in a separate line, you can force it in the setMyImage method (see code below).

3) I had to add ImageWidget to the from Products.Archetypes.public import ImageField line

4) Change PlotImage for MakePlot in the doc string

Also, for those for whom accessing the method via a URL is not all that clear (like me!), you create a new instance of your object, then go .../myObject/MakePlot and just go to .../muObject/base_view to see your image.

Thanx again!

Yves Moisan

from cStringIO import StringIO # import from the graphics package import matplotlib matplotlib.use(Agg) from pylab import *

class mpl (BaseContent): """ An example custom Archetype that generates a PNG graph that's viewed correctly by all browsers. Call PlotImage(), then look at base_view, or aim the browser directly at the URL of the myImage. """ # Add an ImageField to your schema # Read up on the deails of ImageField, it has nice features like scaling schema = BaseSchema.copy() + Schema(( ImageField(myImage, default_content_type=image/png, widget=ImageWidget(label="Image générée par Matplotlib",), ) )) def MakePlot (self,): """ Call this function (e.g., by URL from the browser) to draw the plot. After you call this function, you can see the plot on the base_view page, or directly by URL. """

#imageFile = StringIO() # show that the mimetype starts out as image/png #print "default content type", self.schema['myImage'].getContentType()

t = arange(0.0, 2.0, 0.01) s = sin(2pit) plot(t, s, linewidth=1.0)

xlabel(time (s)) ylabel(voltage (mV)) title(Graphique dans Archetypes) grid(True) #savefig(imageFile) savefig("C:\\temp\\test.png") fh = open("C:\\temp\\test.png", "rb") data = fh.read() self.setMyImage(data,mimetype=image/png) #self.schema['myImage'].setContentType(instance=self, value=image/png) fh.close() clf() close("all")

I cut and pasted from Notepad++ and apparently that screwed CR/LF. Sorry about that.

Following comments on the mpl site, pylab should not be used for a web application (like I showed in my reply to this How To). So I used the code in the webapp_demo.py file in the examples dir of mpl, which basically amounts to replacing the imports. Not using pylab did make a few things not work (e.g. I had to explicitly "import numarray as nx" insterad of "import numerix as nx", as shown in the webapp_demo.py file), but overall it wroks well. Last glitch is to remove the need to write the PNG to a temp file ...

On page 61 of the Matplotlib user's guide I found the following code:

import sys import matplotlib matplotlib.use(Agg) from pylab import * plot([1, 2, 3]) savefig(sys.stdout)

So it seems possible to get rid of the temp file... haven't tried myself, though.

Hope this helps.

Sascha

Sasha,

Since there seems we can't avoid using a temp file, I used the tempfile module to get sth cleaner. First I tried :

... canvas = FigureCanvasAgg(fig) tempplotfilename = tempfile.TemporaryFile(suffix=.png) canvas.print_figure(tempplotfilename.name, dpi=150) ...

but since I'm on windows this won't work because print_figure opens the file and the file is already open and "Whether the name can be used to open the file a second time, while the named temporary file is still open, varies across platforms (it can be so used on Unix; it cannot on Windows NT or later"

so I ended up with :

canvas = FigureCanvasAgg(fig) tempplotfilename = tempfile.mkstemp(suffix=.png) canvas.print_figure(tempplotfilename[1], dpi=150) data = os.read(tempplotfilename[0],os.fstat(tempplotfilename[0]).st_size) self.setLastPlot(data, mimetype=image/png) os.close(tempplotfilename[0]) os.unlink(tempplotfilename[1])

which transparently creates a temp file and cleans itself. Under Unix, TemporaryFile will make things easier.

Cheers!

Yves

Thanks to a tip from Nicolas Young on the MPL mailing list, my app now does no longer need a temp file.

Here's how it's being done:

canvas = FigureCanvasAgg(fig) canvas.draw() size = canvas.get_width_height() buf=canvas.tostring_rgb() im=PILImage.fromstring(RGB, size, buf, raw, RGB, 0, 1) imdata=StringIO() im.save(imdata, format)

imdata now contains the PNG image data that can be rendered on a page.

Hope this helps, Sascha

fig contains the current figure (e.g. fig=pylab.gca() or fig=Figure() etc.) and format is a string (e.g. PNG or JPG).

Sascha

Sascha,

I had to adapt your original recipe (e.g. no get_width_height method on a canvas and also getting imdata.getvalue() to get an image instead of a StringIO object in my browser window) but still couldn't get it to work. Here's what I did :

def pngPlot ... fig = Figure() canvas = FigureCanvasAgg(fig) ax = fig.add_subplot(111) # add a standard subplot y = val_pH c = ax.plot_date(dates, y, -) ax.set_title(self.Title()) ax.set_ylabel(pH) ax.set_xlabel(unicode ('unités', latin-1)) ax.xaxis.set_major_locator(days) ax.xaxis.set_major_formatter(DateFormatter(%d %B)) ax.xaxis.set_minor_locator(DayLocator()) canvas.draw() size = canvas.get_renderer().get_canvas_width_height() buf=canvas.tostring_rgb() im=Image.fromstring(RGB, size, buf, raw, RGB, 0, 1) imdata=StringIO() im.save(imdata, format=PNG) self.REQUEST.RESPONSE.setHeader(Pragma, no-cache) self.REQUEST.RESPONSE.setHeader(Content-Type, image/png) return imdata.getvalue()

In my template :

If I call pngPlot in the browser address bar, I can see it show up in IE and Firefox will prompt me to save the file. If I try and access it through a ZPT, Firefox (1.0+) will complain the file is malformed and crash altogether while IE will just display garbage instead of the image like :

‰PNG  IHDR
@ðþO*0 .....

Hello Yves,

are you sure that Image contains a PIL Image? I forgot to include the import line:

from PIL import image as PILImage

Otherwise you get a Zope image and that does not speak our language...

Hope this helps...

Sascha

Sascha,

Still no go. Same errors. Baffles me.

Thanx anyhow!

I just remember that I had problems when I was trying to include a legend on the Matplotlib chart. I got a ZODB error; I could not find the real cause for this. Maybe it was an MPL bug and it's something similar in your case. Why don't you try with a very simplistic MPL operation (e.g. plot([1,2,3])) and do nothing else.

HTH, Sascha

Perhaps when you use ZPT, it sends text/html as the mime type. Or, think about this: sometimes IE decides what kind of file it's getting by looking at the extension of the filename instead of by looking at the MIME type coming from the server. Can you change the name of the ZPT page to end with .png?

What happens is that both browsers don't behave as expected, although in different ways. The only occasion that it will work is if I save the resulting image (with or without going through a temp file I believe) to an AT ImageField. If I call context/lastPlot in a template and lastPlot is an ImageField to which I have saved a PNG, no problem. The original how to (way up on th thread list) mentioned that one had to tweak the mimetype for things to work, e.g. :

self.setLastPlot(data, mimetype=image/png)

Now if I decide not to update an AT field but instead to return a StringIO object (or a PIL image.getdata()) and explicitly set the mimetype of the response as in (let's say that's the end of the makePLlt function) :

self.REQUEST.RESPONSE.setHeader(Pragma, no-cache) self.REQUEST.RESPONSE.setHeader(Content-Type, image/png)

If I go in the browser address bar and call that function as in http://mysite.com/obj/makePlot, then IE will readily display the image, but Firefox will crash complaining the PNG is malformed. If, and it's the main use case I'm interested in, I'm trying to call in from a page template as in :

both IE and Firefox will display a garbled string. So it's as though the ZPT machinery somehow "forgot" to read in the mimetype that definitely is passed (as evidenced by the fact that IE recognizes it's a PNG when I call the function from the command line).

I tried a tal:define="plot context/makePlot", then calling later tal:content="structure plot", but to no avail. The question is this : why does the same image gets shown when it's a Zope image (e.g. lastPlot) and not when it's called dynamically. It's obviously a mimetype issue, but I don't know what more can be done at this point. Thanx for sharing your thoughts!

Thank you for this howto that was quite usefull. I managed to implement pychart and chartDirector libraries in my product type. I would like to have the image up-dated every time I get the object. I made new accessor for the image field and called the MakePlot() method before getting the image field. It seems that the standard mutator called in MakePlot() to up-date the field content calls somehow the accessor, the code enters therefore in an endless recursion loop. Do you have a solution for this problem?