Question

I have to render a very big image (>50.000² pixel) with cairo. To do this without running out of memory I render parts of the image (<1.000² pixel) one after another and merge them together later.

1. Create 1000×1000 Surface
2. Translate to position of the current part
3. Draw image (calling the drawing instructions using pycairo)
4. Render/Save image to file (cairo_surface_write_to_png)
5. Repeat with next part

Because cairos clipping algorithms are faster than my own, step three draws the whole image, even if only a part of it is visible. Most of the CPU is used in Step 3 (by python). Most of the memory is used in Step 4 (by cairo).

Is there a way to speed things up? Something like this?

1. Create 1000×1000 Surface
2. Draw image
3. Move everything to position of the current part
4. Render/Save image to file
5. Repeat 3 with next part

or

1. Create 50000×50000 Surface
2. Draw image
3. Render/Save only the current part of the image to file
4. Repeat 3 with next part

Answer 1

First of all, using C or Vala instead of Python will probably speed things up.

As for memory usage, I would try writing to my own stream, rather than a file (see write_to_png_stream). This could allow you to (I didn’t try this) control memory usage, assuming Cairo doesn’t call your function only once after everything’s done.