Re: scsh on GNU/Hurd

[Martin Gasbichler]

>>>>> "Niels" == Niels Möller <nisse@lysator.liu.se> writes:

Niels> Hi,
Niels> I've compiled scsh-0.5.3 on a current Hurd machine, which required

Why didn't you try the current version (0.6.2) of scsh? It's avaiable
from http://www.scsh.net/download.html and contains a lot more
features than 0.5.3.

Niels> some minor hacking. config.guess output on this machine was
Niels> "i386-unknown-gnu0.2". The steps needed were:

Niels> 1. I added

Niels>     *-*-gnu* )
Niels>       dir=gnu
Niels>       SCSH_ELF
Niels>     ;;

Niels>   at about line 295 in configure.in, just below the
Niels>   "*-*-linux* )"-clause.

Niels> 2. I copied the scsh/linux directory to a new directory scsh/gnu.

Niels> 3. In syscalls1.c, I added the lines

Niels>     /* MAXPATHLEN is not defined on the Hurd. */
Niels>     /* FIXME: Do this properly, and get rid of the static memory as well 
*/
Niels>     #ifndef MAXPATHLEN
Niels>     #define MAXPATHLEN 500
Niels>     #endif

Niels>   just before the definiton of scm_readlink,

Niels>     /* Read the symlink into static memory. Return NULL on error. */
    
Niels>     static char linkpath[MAXPATHLEN+1]; /*  Maybe unaligned. Not 
reentrant. */
    
Niels>     char const *scm_readlink(const char *path)
Niels>     {
Niels>       int retval = readlink(path, linkpath, MAXPATHLEN);
    
Niels>       return (char const *)
Niels>           ((retval == -1) ? NULL : ( linkpath[retval] = '\0', linkpath 
));
Niels>     }

Niels>   This is an ugly hack, but I haven't figured out how to fix
Niels>   scm_readlink properly.

The code for scm_readlink looks slightly differernt in 0.6.2. Freeing
the buffer is no problem there. Here is a patch that does essentially
the same as the GLib C example quoted by Marcus Brinkmann but has no
memory leaks:

Index: syscalls1.c
===================================================================
RCS file: /cvsroot/scsh/scsh-0.6/scsh/syscalls1.c,v
retrieving revision 1.31
diff -u -c -r1.31 syscalls1.c
*** syscalls1.c 11 Feb 2002 17:32:20 -0000      1.31
--- syscalls1.c 24 May 2002 08:01:32 -0000
***************
*** 189,195 ****
    
  
  /* Read the symlink. */
! 
  s48_value scsh_readlink(s48_value sch_path)
  {
    char linkpath[MAXPATHLEN+1];
--- 189,195 ----
    
  
  /* Read the symlink. */
! #ifdef MAXPATHLEN
  s48_value scsh_readlink(s48_value sch_path)
  {
    char linkpath[MAXPATHLEN+1];
***************
*** 202,207 ****
--- 202,238 ----
        return s48_enter_string(linkpath);
      }
  }
+ #else
+ s48_value scsh_readlink(s48_value sch_path)
+ {
+   char *linkpath;
+   int size;
+   int retval;
+   s48_value sch_sym_link_path = S48_UNSPECIFIC;
+   
+   for (size = 256;; size *=2){
+ 
+     linkpath = Malloc(char,size);
+     if (!linkpath)
+       s48_raise_os_error_1(errno, sch_path);
+ 
+     retval = readlink(s48_extract_string (sch_path), linkpath, size);
+ 
+     if (retval == -1){
+       free (linkpath);
+       s48_raise_os_error_1(errno, sch_path);
+     }
+     
+     if (retval < size){
+       sch_sym_link_path = s48_enter_substring (linkpath, retval);
+       free (linkpath);
+       return sch_sym_link_path;
+     }
+     free (linkpath);
+   }
+ }
+ #endif
+ 
  
  s48_value scsh_rename(s48_value sch_from, s48_value sch_to)
  {


Niels> 3. sigset_t seems to be a plain unsigned int, so I had to redefine the
Niels>    macros in scsh/gnu/sigset.h, as follows:

Niels>      /* On gnu, it seems a sigset_t is just an unsigned int. */
Niels>      #define make_sigset(maskp, hi, lo) \
Niels>          (*(maskp) = (unsigned long int) ((hi) << 24) | (lo))
     
Niels>      #define split_sigset(mask, hip, lop)\
Niels>          ((*(hip) = ((mask) >> 24) & 0xff),\
Niels>          (*(lop) = ((mask) & 0xffffff)))
    
These hacks are gone in the 0.6 series of scsh.

Niels> After these steps, both make and make check succeeded.

There is another small test suite, which can be invoked by

./go -lm scsh/test/test-packages.scm -o test-all -c "(test-all)"

in the top-level directory of scsh-0.6.2. 

I suspect some more adjustments are required in the scsh/gnu/
directory, but I don't know, how releated Hurd and Linux
are. doc/porting.txt provides some (maybe out-dated) information on
this topic.

Thanks for the report!

-- 
Martin