This was my question, so I looked at the source. It refers to a page on algorithms that claims that it is 8-bit clean, i.e., it preserves non-ascii characters but does not modify them. I then compiled a test program to verify this, because the bit twiddling in the source code looks a little nuanced. It IS 8-bit clean.

Although, if you're looking for something that converts non-ascii characters to their lower case equivalents, just use LibICU. This library looks nice for decoding things like HTTP which have plenty of ascii-only case insensitivity.

Since it is doing bit manipulation, then probably not. I mean, the toupper function is treating (4) 8 bit chars as one 32 bit uint. Then doing math... so, I assume that it will miss the finer points of I18N.

However, the base64/85 en/decoders should work just fine.

And in the same vein as the nginx optimizations suggested in a comment yesterday, it fails on arcitectures that don't support unaligned access.