On Thu, Feb 08, 2007 at 04:12:48PM +0000, Dave Mitchell wrote: > (Arguably a patten should store both plain and utf8 versions of each > exact string for quicker matching.) If it needs to. If this were done at compile time, rather than lazy caching, it gives the option: "8" bit string, but NULL for UTF-8 => string is actually purely 7 bit UTF-8 string, but NULL for 8 bit => string has code points above 255 However, I'm not sure if that plays nicely with the semantics that PL_encoding would like to offer. Nicholas Clark