12 years ago, I was working on a PHP library to parse XMLRPC requests, and I ran into it's numeric type slipperiness in a painful way. I was (naively) using PHP's eval to parse a decimal numeric string.

I noticed what seemed like random numeric errors in a random subset of RPCs. It turned out, after what seemed like eons of investigation, that some clients were stringifying decimal integers with leading zeros, which made PHP interpret them as octal numbers, and hence the random seeming errors.

Having come from a strongly-typed-language background (C/C++/Java), that was a major learning experience about the caveats of weakly typed languages.