Interfacing directly with POSIX File I/O


In March 2004 there was a lengthy discussion between Krishna Myneni (of kForth fame) and me. Krishna had been introducing an IOCTL word to his system - trying to get other Forth system makers to do something as well as it will prove very useful on the application level. That is true - but there are a number of problems attached.

The first problem is that no forth system maker uses the posix integer file descriptors as a file reference on the Forth application level. Instead it is a wrapper around whatever is being used internally, for PFE that is usually the stdio file handle (called FILE). The second and perhaps bigger problem: the commands to IOCTL are usually defined as C preprocessor macros and they value differs from system to system. Therefore the bitvalues must be exported as named constants to the forth application level.

Other Systems

So, what is it like on other languages - we can easily compare with Perl, Python and PHP here which all provide a POSIX module for some more direct interaction with the underlying operating system. Basically we see that the best thing is to have a separate module that does not overload the native "open" and "file" semantics - especially the "open" word name should be left intact for applications to use them as their own method name.

In perl the functions not in extra module but they are builtin as sysopen, dbmopen and opendir alongside the "open" function for the native file handles of perl. In python there is a definite module concept that allows to put the function as (matching a name like OS-OPEN in Forth? no, I don`t like that either). In PHP, there are many open functions from their respective modules like bzopen, dba_open, dbmopen, dio_open, fopen, fsockopen, opendir, popen, zip_open.

Consequently, it would be sys-ioctl? Well, perhaps it should be more forthish in its name, where we do not abbreviate as much as this one. Perhaps the "io*" has something to it, what about OPEN-IO / WRITE-IO / CLOSE-IO / SET-IO ? Oh, by the way, there are actually two ioctl names, ioctl and fcntl, that can be used for the module name for system interaction. Which would also need to export as many #defs as possible


Most operations about files can be achieved as well through the Forth filehandle wrappers. For some special circumstances one might consider the addition of "fileno" (which is a call in C stdio) that returns the underlying file descriptor of the operating system. However we find that in reality the task of access to the posix file descriptors comes from the need to use a SOCKET.

the socket interface is somewhat distinct from posix device files. As far as I am concerned, I'd vote for a forthish module like OPEN-SOCKET, LISTEN-SOCKET etc. The reason for the posix way had been always to use the socket descriptor handle as a means to a select() call - however a similarity to that would need to wrapped in highlevel forth anyway simply for the FILE-EXT interface, and also since win32 has a waitobject functionality truly different than unix select() - in fact the win32 sockets are not filehandle (on some versions of windows).

have a look at gforth/unix/socket.fs which imports the libc names with building an open-socket around it that does actually return a forth file-ext descriptor. That is surely a good means to it, so that a forth SOCKET should be used with the forth WRITE-FILE calls and not with SYSWRITE. If one wants to export the raw interface then it should be again with a sys-prefix like SYSSOCKET, to allow to wrap the actual open-socket/write-socket calls to either a file-ext based system or the sys-ext based.