> But I do like the idea of having a special prefix character. What
> characters do you think would be good? Of course, there are no
> characters which are invalid for pathname (other than NUL), so we
> still can't avoid messing up something that someone might be doing
> somewhere.
>
> Why not implement it as #?perl, and define a new binfmt interpreter
> which rewrites it as #!/usr/bin/perl (or whatever), and then calls the
> #! interpreter? Having binfmt interpreters do a bit of prework and then
> call another interpreter is how binfmt_java is implemented, after all.
>
> This way, we avoid the conflict altogether.
How about implementing some kind of "program interpreter" which
- analogous to what /lib/ld.so.1 does for ELF executables and libs -
the process of searching the path, security issues, /etc/ld.so.cache
style cache for faster searching the patch etc? This approach would
be more flexible than a kernelspace implementation though slighly
less performant.
Ralf