We’ve reached another minor milestone. Match-case statements are now supported including some rules for how variables are allowed to be used in branching code. This also means that Plasma is now Turing complete, although it is far from being expressive.

Match statements attempt to match an expression against a number of patterns. Currently the only patterns that are supported are integers, variable names and underscore (a wild-card pattern). Cases are tried in order. Later the compiler may warn about cases that are never executed (because an earlier case matches).

    match (n) {
      0 -> {
        beer = "There's no beer!"
      1 -> {
        beer = "There's only one beer"
      m -> {
        beer = "There are " ++ int_to_string(m) ++ " bottles of beer"

If a variable is produced by at least one case, and used outside the match statement then it must be produced in all cases that fall through. If a variable is produced on at least one case, and not used outside the match, then it will be named apart by the compiler, allowing it to have a different type from variables with the same name in other cases. These rules have been implemented however the error messages will abort compilation. Friendlier errors will be implemented later.

If it is possible that no case matches, and the other cases produce values used after the match, then that is also an error. However, until types are properly implemented, we can’t detect this error.

Cases may contain a return statement (or otherwise not fall through, if that happens they do not need to produce any variables. Currently the compiler only supports either all cases falling through, or none falling through. I plan to implement this later, probably by transforming the code after the match into a separate function which is called in the cases that fall through. This will also have implications for type checking due to the order of compilation passes, so it will have to wait until then.