xargs: The command line for command is built up until it reaches a system-defined limit - What limit? [duplicate]

Question

I looked briefly at the source code of xargs in findutils 4.6.0 and saw that 'n' defines: args_per_exec But there seems to be no default value for args_per_exec in find utils

So where is the 'system-defined limit' defined?

It's no duplicate since the question is meant as explanation for the manpage entry, not the generic question for an arbitrary command length

Answer 1

The limit is sysconf(_SC_ARG_MAX), including the environment strings and 2048 bytes of headroom. That's the maximum length of the actual strings + the separating null bytes, not the number of arguments. This is from GNU xargs' source:

      /* IEEE Std 1003.1, 2003 specifies that the combined argument and
       * environment list shall not exceed {ARG_MAX}-2048 bytes.  It also
       * specifies that it shall be at least LINE_MAX.
       */
      long val;
#ifdef _SC_ARG_MAX
      val = sysconf (_SC_ARG_MAX);
      if (val > 0)
        {
          assert (val > XARGS_POSIX_HEADROOM);
          /* Note that val can in fact be greater than ARG_MAX
           * and bc_ctl.arg_max can also be greater than ARG_MAX.
           */
          bc_ctl.arg_max = smaller_of (bc_ctl.arg_max,
                                       (size_t)val-XARGS_POSIX_HEADROOM);

That is then clamped to 128k in bc_use_sensible_arg_max(), but which could be increased via the -s option:

void
bc_use_sensible_arg_max (struct buildcmd_control *ctl)
{
#ifdef DEFAULT_ARG_SIZE
  enum { arg_size = DEFAULT_ARG_SIZE };
#else
  enum { arg_size = (128u * 1024u) };
#endif

Both the limit calculation and the clamping are the same for find -exec ... {} +; the code from above is duplicating the logic from bc_get_arg_max() and bc_init_controlinfo(), the latter of which is called with a headroom argument of 2048 from both find/parser.c and xargs/xargs.c.

GNU xargs also has a --show-limits option.