on the passed path...
For fexist() simply return TRUE if access()!=-1 && !isdir()
For fexistcase() *try* the access()/isdir() method first, then do the full
search.  Should make a reasonable difference.  Profileing suggests that
iniFileName() is quite slow due to the fexistcase() calls.  As the general
case is that the passed path is correct, this should speed up most calls
to fexist*().

path (to be used in place of mkdir).

fname against the filespec spec, not allowing * to match the path delimiter
if path is TRUE.

Consider doing the same for all .h files.
Consider the meaning of sanity.

present.

over-write backslash with forwardslash (rather than just appending it, creating
and invalid path string).

(* or ?) chars are present.  Doing so prevents re-writing of the path
parameter, so behaviour is different between the two.

This would break something like:
int munlink(char *glob)
{
        char    this[MAX_PATH+1];

        strcpy(this,glob);
        while(fexistcase(this)) {
                unlink(this);
                strcpy(this,glob);
        }
}

Does Synchronet rely on the old behaviour anywhere?

delimiter everywhere in the world.

filename is an absolute (no relative directories at all), or is a full (rooted)
path.

it doesn't return directory entries beginning with a period (e.g. '.' and '..')
unless the GLOB_PERIOD flag is specified.

filename) if passed a path that included both foward and back-slashes.

included in directory and '.' included in extensions.

path/filename (or NULL if no extension exists).

long filenames) on Win32.

Created IS_PATH_DELIM() macro (stole websrvr.c->is_slash macro).
Defines BSD _PATH_TMP and _PATH_DEVNULL macros for all platforms.

(truncates the slash before calling stat).

all platforms, removed realpath_r()