Starting java.exe from ash/dash breaks command line arguments with spaces

[dmik]

Given this hello.java:

public class hello
{
  public static void main(String args[])
  {
    System.out.println("hello world!");
    for (int i = 0; i < args.length; ++i)
      System.out.println("arg["+i+"]=["+args[i]+"]");
  }
}

and this SH script:

D:/Tools/openjdk6_b27_sdk_os2_ga5-20130217/bin/java.exe -cp . hello "1 2" 3

you get this output if the script is run by bash:

hello world!
arg[0]=[1 2]
arg[1]=[3]

and this output if it's run by ash/dash:

hello world!
arg[0]=[1]
arg[1]=[2]
arg[2]=[3]

Apparently, in the second case spaces are handled incorrectly.

[dmik]

Note that if you take a normal C test like this:

#define OS2EMX_PLAIN_CHAR
#define INCL_BASE
#include <os2.h>

#include <alloca.h>
#include <sys/stat.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <string.h>

int main(int argc, const char **argv)
{
  printf ("Hello, world!\nHello, Universe!\n");

  PPIB ppib;
  DosGetInfoBlocks(NULL, &ppib);
  char *s = ppib->pib_pchcmd;
  while (*s)
  {
    printf ("pchcmd [%s]\n", s);
    s += strlen(s) + 1;
  }

  int i;
  for (i = 0; i < argc; ++i)
    printf ("arg[%d]=[%s]\n", i, argv[i]);
  return 0;
}

and run this script:

hello.exe "1 2" 3

the result are as follows for bash:

Hello, world!
Hello, Universe!
pchcmd [hello.exe]
pchcmd ["1 2" 3]
arg[0]=[hello.exe]
arg[1]=[1 2]
arg[2]=[3]

and for ash/dash:

Hello, world!
Hello, Universe!
pchcmd [../hello.exe]
pchcmd [kLIBC]
pchcmd [ 1 2]
pchcmd [ 3]
arg[0]=[../hello.exe]
arg[1]=[1 2]
arg[2]=[3]

You may see that the raw command line (pib_pchcmd) differs because kLIBC applies special processing to it to improve performance when handling arguments in exec/spawn and fork for kLIBC-based executables, but the result you see through the standard argv vector are identical (as they should be).

So my bet is that our Java launchers try to play with the raw command line and don't take all the details into account (I recall that I did something in that area now I need to find the place).

[dmik]

Okay, I got why it breaks. We actually go Windows way a lot there due to heavy use of Odin and we in particular use Odin's RegisterLxExe() and __argcA/__argvA which are of course set directly right from the raw command line. There are hacks that work around KLIBC specifics but they are not perfect and don't properly process everything. We should fix it in Odin I guess. Sigh.

[dmik]

I fixed this in Odin, see ​http://trac.netlabs.org/odin32/changeset/22131. Now Java processes arguments with spaces and with quotes in them correctly. In other words, running the above snippet from dash/ash produces exactly the same results as from the old, EMX-based bash.

[dmik]

The DLLs from the test Odin build are here (just in case if you want to test w/o rebuilding Odin): ​http://rpm.netlabs.org/test/tod1.zip.