Swap subarrays of a numpy array

Question

First a bit of context.

I am trying to draw into a Gdk (actually pygtk) pixbuf with Cairo (actually pycairo). My original code looked like this:

import cairo as C
import gtk.gdk as GG

FMT = C.FORMAT_ARGB32
CSP = GG.COLORSPACE_RGB

st = C.ImageSurface.format_stride_for_width(FMT, zw)
surf = C.ImageSurface(FMT, zw, zh)
c = C.Context(surf)

# draw into c here ...

pixels = surf.get_data()
return GG.pixbuf_new_from_data(pixels, CSP, True, 8, zw, zh, st)

For a time, it looked like this would just work, until I tried to draw colors instead of just black text. It turns out the two libraries disagree on byte order, like so:

# !!! FIXME !!! cairo ARGB surfaces and gdk pixbufs disagree on bytesex:
# cairo: LSB --> B G R A <-- MSB
# gdk:   LSB --> R G B A <-- MSB
# !!! WTF !!!

The result is (after blitting from the pixbuf to screen) an image with red and blue channels swapped :-(

So, if I keep using pixbufs (as I want to) I need to postprocess the data, swapping every 1st byte with every 3rd. I can do that in plain python:

def changesex(data):
    i0 = 0
    i2 = 2
    l = len(data)
    while i2 < l:
        temp = data[i0]
        data[i0] = data[i2]
        data[i2] = temp
        i0 += 4
        i2 += 4

but I thought using numpy might be faster if there is a built-in operator for this written in C. Something like this:

a = np.frombuffer(data, dtype=np.uint8)
a.shape = (len(data) / 4, 4)
temp = a[:,0]
a[:,0] = a[:,2]
a[:,2] = temp

Am I just dreaming?

Answer 1

Noting a your initial (N,4) array of B,G,R,A, you could get a (N,4) representation as R,G,B,A rather simply using some advanced indexing:

a = np.frombuffer(data, dtype=np.uint8)
a.shape = (-1, 4)
rearranged = a[:,[2,1,0,3]]