Thursday, September 5, 2013

Q: Run C or C++ file as a script, my "almost" c-interpreter and c-scripting utility.

This question on stackoverflow that sparked my curiosity:
"So this is probably a long shot, but is there any way to run a C or C++ file as a script"

With gcc it is not so straight forward, but with the tiny C compiler tcc this is easy, e.g.:
File: hw.c
#!/usr/local/bin/tcc -run
#include <stdio.h>

int main() 
{
    printf("Hello, world!\n");
    return 0;
}
 
¢ Didn't someone (Rusty?) manage to get linux to boot completely from a source-code C-script, starting with only a tcc binary? ¢

In Algol68g (my weekend extreme sport language) this is "reasonably" easy... mostly because a "#comment#" is native to the Algol68 language. e.g. here is an implementations of the classic echo command.

File: echo.a68
#!/usr/bin/a68g --script #
# -*- coding: utf-8 -*- #
 
STRING ofs := "";
FOR i FROM 4 TO argc DO print((ofs, argv(i))); ofs:=" " OD

But how can we natively do this in C with gcc? ...

Simple shebangs can help with scripting, e.g. "#!/usr/bin/env python" at the top of a Python script will allow it to be run in a terminal as "./script.py".
The task Multiline shebang largely demonstrates how to use "shell" code in the shebang to compile and/or run source-code from a 3rd language.
However in this task Native shebang task we are go native. In the shebang, instead of running a shell, we call a binary-executable generated from the original native language, e.g. when using C with gcc "#!/usr/local/bin/script_gcc" to extract, compile and run the native "script" source code.
Other small innovations required of this Native shebang task:
  • Cache the executable in some appropriate place in a path, dependant on available write permissions.
  • Generate a new cached executable only when the source has been touched.
  • If a cached is available, then run this instead of regenerating a new executable.
Difficulties:
  • Naturally, some languages are not compiled. These languages are forced to use shebang executables from another language, eg "#!/usr/bin/env python" uses the C binaries /usr/bin/env and /usr/bin/python. If this is the case, then simply document the details of the case.
  • In a perfect world, the test file (e.g. echo.c) would still be a valid program, and would compile without error using the native compiler (e.g. gcc for text.c). The problem is that "#!" is syntactically incorrect on many languages, but in others it can be parsed as a comment.
  • The "test binary" should be exec-ed and hence retain the original Process identifier.
Test case:
  • Create a simple "script file" (in the same native language) called "echo" then use the "script" to output "Hello, world!"

C

File: script_gcc.c
#!/usr/local/bin/script_gcc.sh
/* Optional: this C code initially is-being/can-be boot strapped (compiled) using bash script_gcc.sh */
#include <errno.h>
#include <libgen.h>
#include <stdarg.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/stat.h>
#include <unistd.h>
 
/* the actual shebang for C target scripts is:
#!/usr/local/bin/script_gcc.c
*/
 
/* general readability constants */
typedef char /* const */ *STRING;
typedef enum{FALSE=0, TRUE=1} BOOL;
const STRING ENDCAT = NULL;
 
/* script_gcc.c specific constants */
#define DIALECT "c" /* or cpp */
const STRING
  CC="gcc",
  COPTS="-lm -x "DIALECT,
  IEXT="."DIALECT,
  OEXT=".out";
const BOOL OPT_CACHE = TRUE;
 
/* general utility procedured */
char strcat_out[BUFSIZ];
 
STRING STRCAT(STRING argv, ... ){
  va_list ap;
  va_start(ap, argv);
  STRING arg;
  strcat_out[0]='\0';
  for(arg=argv; arg != ENDCAT; arg=va_arg(ap, STRING)){
     strncat(strcat_out, arg, sizeof strcat_out);
  }
  va_end(ap);
  return strndup(strcat_out, sizeof strcat_out);
}
 
char itoa_out[BUFSIZ];
 
STRING itoa(int i){
  sprintf(itoa_out, "%d", i);
  return itoa_out;
}
 
time_t modtime(STRING filename){
  struct stat buf;
  if(stat(filename, &buf) != EXIT_SUCCESS)perror(filename);
  return buf.st_mtime;
}
 
/* script_gcc specific procedure */
BOOL compile(STRING srcpath, STRING binpath){
  int out;
  STRING compiler_command=STRCAT(CC, " ", COPTS, " -o ", binpath, " -", ENDCAT);
  FILE *src=fopen(srcpath, "r"),
       *compiler=popen(compiler_command, "w");
  char buf[BUFSIZ];
  BOOL shebang;
 
  for(shebang=TRUE; fgets(buf, sizeof buf, src); shebang=FALSE)
    if(!shebang)fwrite(buf, strlen(buf), 1, compiler);
 
  out=pclose(compiler);
  return out;
}
 
void main(int argc, STRING *argv, STRING *envp){
 
  STRING binpath,
         srcpath=argv[1],
         argv0_basename=STRCAT(basename((char*)srcpath /*, .DIALECT */), ENDCAT),
         *dirnamew, *dirnamex;
  argv++; /* shift */
 
/* Warning: current dir "." is in path, AND * /tmp directories are common/shared */
  STRING paths[] = {
    dirname(strdup(srcpath)), /* not sure why strdup is required? */
    STRCAT(getenv("HOME"), "/bin", ENDCAT),
    "/usr/local/bin",
    ".",
    STRCAT(getenv("HOME"), "/tmp", ENDCAT),
    getenv("HOME"),
    STRCAT(getenv("HOME"), "/Desktop", ENDCAT),
/*  "/tmp" ... a  bit of a security hole */
    ENDCAT
  };
 
  for(dirnamew = paths; *dirnamew; dirnamew++){
    if(access(*dirnamew, W_OK) == EXIT_SUCCESS) break;
  }
 
/* if a CACHEd copy is not to be kept, then fork a sub-process to unlink the .out file */
  if(OPT_CACHE == FALSE){
    binpath=STRCAT(*dirnamew, "/", argv0_basename, itoa(getpid()), OEXT, ENDCAT);
    if(compile(srcpath, binpath) == EXIT_SUCCESS){
      if(fork()){
        sleep(0.1); unlink(binpath);
      } else {
        execvp(binpath, argv);
      }
    }
  } else {
/* else a CACHEd copy is kept, so find it */
    time_t modtime_srcpath = modtime(srcpath);
    for(dirnamex = paths; *dirnamex; dirnamex++){
      binpath=STRCAT(*dirnamex, "/", argv0_basename, OEXT, ENDCAT);
      if((access(binpath, X_OK) == EXIT_SUCCESS) && (modtime(binpath) >= modtime_srcpath))
        execvp(binpath, argv);
    }
  }
 
  binpath=STRCAT(*dirnamew, "/", argv0_basename, OEXT, ENDCAT);
  if(compile(srcpath, binpath) == EXIT_SUCCESS)
    execvp(binpath, argv);
 
  perror(STRCAT(binpath, ": executable not available", ENDCAT));
  exit(errno);
}
Test Source File: echo.c
#!/usr/local/bin/script_gcc.c
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
 
int main(int argc, char **argv, char **envp){
  char ofs = '\0';
  for(argv++; *argv; argv++){
    if(ofs)putchar(ofs); else ofs=' ';
    fwrite(*argv, strlen(*argv), 1, stdout);
  }
  putchar('\n');
  exit(EXIT_SUCCESS);
}
Test Execution:
$ ./echo.c Hello, world!
Test Output:
Hello, world!

UNIX Shell

Works with: Bourne Again SHell
Note: this Native shebang task does not exactly apply to bash because bash is interpretive, but as a skeleton template the following script is an example of how compiled languages can implement the shebang. Also: this bash code can be used to automatically compile the C code in /usr/local/bin/script_gcc.c above.
File: script_gcc.sh
#!/bin/bash
 
# Actual shebang when using bash:
#!/usr/local/bin/script_gcc.sh
 
# Alternative shebang when using bash:
#!/bin/bash /usr/local/bin/script_gcc.sh
 
# CACHE=No # to turn off caching...
 
# Note: this shell should be re-written in actual C! :-)
 
DIALECT=c # or cpp
CC="gcc"
COPTS="-lm -x $DIALECT"
IEXT=.$DIALECT
OEXT=.out
 
ENOENT=2
 
srcpath="$1"; shift # => "$@"
#basename="$(basename "$srcpath" ."$DIALECT")"
basename="$(basename "$srcpath")"
 
# Warning: current dir "." is in path, AND */tmp directories are common/shared
paths="$(dirname "$srcpath")
$HOME/bin
/usr/local/bin
.
$HOME/tmp
$HOME
$HOME/Desktop"
#/tmp
 
while read dirnamew; do
  [ -w "$dirnamew" ] && break
done << end_here_is
$paths
end_here_is
 
compile(){
  sed -n '2,$p' "$srcpath" | "$CC" $COPTS -o "$binpath" -
}
 
if [ "'$CACHE'" = "'No'" ]; then
  binpath="$dirnamew/$basename-v$$$OEXT"
  if compile; then
    ( sleep 0.1; exec rm "$binpath" ) & exec "$binpath" "$@"
  fi
else
  while read dirnamex; do
    binpath="$dirnamex/$basename$OEXT"
    if [ -x "$binpath" -a "$binpath" -nt "$srcpath" ];
      then exec "$binpath" "$@"; fi
  done << end_here_is
$paths
end_here_is
 
  binpath="$dirnamew/$basename$OEXT"
  if compile; then exec "$binpath" "$@"; fi
 
  echo "$binpath: executable not available" 1>&2
  exit $ENOENT
fi
 
Test Source File: echo.c
#!/usr/local/bin/script_gcc.sh
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
 
int main(int argc, char **argv, char **envp){
  char ofs = '\0';
  for(argv++; *argv; argv++){
    if(ofs)putchar(ofs); else ofs=' ';
    fwrite(*argv, strlen(*argv), 1, stdout);
  }
  putchar('\n');
  exit(EXIT_SUCCESS);
}
Test Execution:
$ ./echo.c Hello, world!
Test Output:
Hello, world!
 
Here is a summary of the input script files, together with binary files generated:
$ ls -ltr hw.* echo.* /usr/local/bin
-rwxr-xr-x. 1 nevillednz nevillednz   193 Sep  6 13:17 hw.c
-rwxr-xr-x. 1 nevillednz nevillednz  4757 Sep  6 13:18 hw.c.out
-rwxr-xr-x. 1 nevillednz nevillednz   131 Sep  6 13:19 echo.a68
-rwxr-xr-x. 1 nevillednz nevillednz 12373 Sep  6 13:20 echo.so
-rwxr-xr-x. 1 nevillednz nevillednz   315 Sep  6 13:29 echo.c
-rwxr-xr-x. 1 nevillednz nevillednz  5128 Sep  6 13:30 echo.c.out

/usr/local/bin:
total 20
-rwxr-xr-x. 1 nevillednz nevillednz 1216 Sep  6 13:13 script_gcc.sh
-rwxr-xr-x. 1 nevillednz nevillednz 3431 Sep  6 13:15 script_gcc.c
-rwxr-xr-x. 1 nevillednz nevillednz 8564 Sep  6 13:16 script_gcc.c.out



 
One interesting thing to observe is that "script_gcc.c" is boot strapped by "script_gcc.sh", so as long as gcc is installed, you don't have to compile anything... not even the script_gcc.c "script"...  (But you do have to have the right permissions to start off with!)

 

No comments:

Post a Comment