Unix-shell programs are designed to read the first line of any “command file,” looking therein for a #!shebang™ line. If they find one, they fork() the specified command, providing “the rest of the input-file” to the subprocess.

I doubt that. In modern unixes, the shell simply fork()s and then wait()s, and, in the child process, invokes exec(script,arguments). The kernel, not the shell, detects what kind of file is about to be executed, and runs the proper loader/interpreter. For dynamically linked ELF binaries, the proper loader name is embedded in the ELF file (something like /lib/ld-linux.so). For other file formats, a similar logic is used. For some file formats, a hard-coded or configurable helper program is invoked. For scripts that start with #!, the interpreter is the first "word" (block of non-whitespace characters) after #!. It may have an optional argument (everything up to end of line).

Shells usually have some fallback code that handles a failing exec() for a text file, as required by POSIX.

Links:

• History and various implementations
• binfmt_script in Linux
• exec implementation in Linux - note that exec.c uses the various binfmt_*.c modules to allow different executable formats.
• bash execute_cmd - relevant code is in shell_execve().
• busybox ash - relevant code is in tryexec() and shellexec().

Alexander