Binfmtc – binfmt_misc C scripting interface | Hacker News Reader

Binfmtc – binfmt_misc C scripting interface

netfort.gr.jp

103 pointstodsacerdoti28 comments

You can already do this without using binfmt ...

  #if 0
  gcc "$0" -o "$@".out && exec ./"$@".out
  #endif
  #include <stdio.h>
  int main () { printf ("hello, world\n"); return 0; }

Usage:

  $ chmod +x print.c
  $ ./print.c
  hello, world

(Could be better to use a temporary file though.)

There's a similar cute trick for compiled OCaml scripts that we use with nbdkit: https://libguestfs.org/nbdkit-cc-plugin.3.html#Using-this-pl...

Compiler errors won’t cause as many funny consequences with

  gcc "$0" -o "$@".out && exec ./"$@".out || exit $?   # I'd use ${0%.c} not $@

Love this trick too, but the difference, as far as I understand, is that it only works with a Bourne(-compatible) shell, whereas shebangs or binfmt_misc also work with exec().

You can also #embed the compiler binary, and execve it to much the same effect as binfmtc. I explored that trick for an IOCC entry that was never submitted because it ended up far too readable.

Oh this is neat. Took me a bit.

The shell treats the first line as a comment. It executes the second line, which eventually exec's the binary so the rest of the file do not matter to the shell.

And the compiler treats the first line as a preprocessor directive, so it ignores the second line.

I initially misread/mistook the first line for a shebang.

The use of $@ doesn't look right to me.

In the trivial case exposed here where there are no additional arguments to pass to the .c program, the shell executes

  gcc "print.c" -o .out && exec ./.out

and it works "by chance".

In a more complex scenario where print.c expects some parameters, it won't work. For example,

  ./print.c a b c

will result in the shell trying to invoke

  gcc "print.c" -o "a" "b" "c".out && exec ./"a" "b" "c".out

which makes no sense.

Are you sure you didn't intend $0 instead of $@ ?

It's true, that's a mistake!

OTOH we're trying to write self-compiling executable C scripts, so the safety, correctness and good sense ships sailed a while back.

Is there a way to do this and have the shell remove the temporary file after exec?

There are many, no doubt

   #if 0
   cc -static -pipe -xc "$0"||exit 100
   exec ./a.out
   #endif
   int puts(const char *);
   int unlink(const char *);
   int main(int argc,char *argv[]){ 
   puts("it works"); 
   unlink(argv[0]);
   return 0; 
   }

Yes, but it's not worth it. It's better to forget gcc and use tcc instead, which has the -run flag to compile and run without creating any intermediate file. It's also much quicker than gcc.

Or even

    /*bin/true ; exec tcc -run "$0" "$@" # */

I'd have expected this to need a hashbang (#!/bin/sh) at the beginning. Why doesn't it?

Because your shell will execute anything that looks like a text file & has +x bit set as a shell script.

Similar project of mine from a long while ago: https://git.zx2c4.com/cscript/about/

Amazing; when I tried something similar I used a "#!" line pointing to a C compiler + runner of sorts (https://github.com/radiospiel/jit). https://git.zx2c4.com/cscript/about/ is also following that approach.

What is the benefit of registering an extension via binfmt_misc?

It's still valid C source code? Is the #! sequence valid C?

There's also tcc, which has a shebang compatible extension to allow it to be used by scripts.

This is doable entirely without a Linux-specific binfmt-misc hack.

https://rosettacode.org/wiki/Multiline_shebang#C

This is a neat hack, but the whole file is not a valid C program.

You mean it consists of a C program, plus non-C cruft to get it running?

Isn't that already legitimized by configure scripts, compiler command lines and Make files?

of course it is legitimized. I just meant that the selling point of binfmtc is that the same .c file can either be compiled by any C compiler, or executed directly.

We could perhaps split it into a two-file system where a "foo.sh" containing certain boilerplate will execute a "foo.c" that doesn't require any special conventions at all.

"foo.sh" could be identical for any .c file, and so we could symbolically link them all to a common file.

Of course, neither that file nor symlinks to it require a .sh suffix.

The contents might look like this:

  #!/bin/sh
  c_file=${0%.sh}.c
  x_file=${0%.sh}.bin

  # If cached executable file is missing, or out of date w.r.t.
  # the source file, compile it:

  if [ "$c_file" -nt "$x_file" ]; then
    cc $CFLAGS "$c_file" -o "$x_file"
  fi

  # Run cached executable file
  $x_file

I did a simple hack for doing the same thing from inside a C program for my book:

https://github.com/codr7/hacktical-c/tree/main/dynamic

was surprised that "sudo apt install binfmtc" works out of the box on my box (linux mint) and i can do the magic just as described here

(2006)

C is still my first love. You can hold the whole language in your head, and it’s fast. Yes there are footguns but it’s a libertarian programming language - you’re responsible for what you build. No hand holding.

I like that too, but the problem is that C doesn't keep its end of the deal. No hand holding, but make what you are doing transparent. It used to be the case back in the 80s, but not anymore. Not with our optimizing compilers and oodles of UB and spec subtleties and implicit actions.

So lovely

Reply on news.ycombinator.com