What architectural patterns does Pillow use to make image operations fast and flexible? Have you found the need to do any significant refactorings of the original code to make it compatible with modern uses and execution environments?

[aclark4life]

[wiredfool]

Pillow's core is a C extension, using (for the most part) standard looping constructs to perform image operations at that level. There haven't been any major refactorings at that layer, though there have been some exploration of parallelisim using SMID and OpenMP, as well as some algorithmic changes to the blur and resampling code to improve performance by improving cache friendliness.

[wiredfool]

Many of the bits in the C level are pluggable for the different image format types, either by macros or function pointers. This leads to small understandable chunks of code to handle the distinct cases. (for the most part, ymmv)