Module Env.NameSource

Environment variable names. Windows has complicated semantics for environment variables. The retrieval functions are case insensitive, but it's "legal" for the environment block to contain entries which differ only by case. If environment variables are set entirely using CRT or Win32 API functions, then there isn't usually a problem, the issue arises when creating a program where the environment block is instead passed. In this model, it's very easy to end up with two bindings in the same block. When dealing with Windows programs, this will mostly be transparent, but it's a problem with Cygwin which actively allows "duplicate" entries which differ by case only and implements Posix semantics on top of this. The problem is constantly with us thanks to the use of PATH on Unix, and Path on Windows! opam tries to ensure that environment variables are looked up according to the OS semantics (so case insensitively on Windows) and OpamEnv goes to some trouble to ensure that updates to environment variables are case preserving (i.e. PATH+=foo gets transformed to Path+=foo if Path exists in the environment block).

Key to this is not accidentally treating environment variable names as strings, without using the appropriate comparison functions. Name.t represents environment variable names as private strings, providing comparison operators to handle them, and still allowing the possibility to coerce them to strings.