re: Problem with 11/785 running 4.3 BSD

[wolfgang@MGM.MIT.EDU (Wolfgang Rupprecht)]

Ha! looks like BDS4.3 really loses on this count.
I just discovered this the hard way 4 days ago.

The bug is that the Lsave() code closes all the open FILE * _iob[]
structs.  Problem is, these AREN'T all contiguous in the BSD4.3 _iob[]
layout.  This causes havoc when flose() is called on some stacked
return addresses, etc.

(Side note: this exposes a second bug. Even though free() gets pissed
off at the memory it just got handed and calls FEerror(), something
still corrupts the malloc arena, and make_simple_string dies the big
SEGV death.)

A quick ~portable hack fix is just to close (iobs < 20).  The correct
fix probably involves BSD4.3 specific code. (In discussions with Ken
Rimey on this matter, he made an excellent suggestion to call _fwalk()
with a function that calls fclose on the _iob *only* if the _iob isn't
stderr.)

Diffs that I use here at mgm (vax750 w. bsd4.3) follow.

*** /usr/local/okcl/c/unixsave.c-dist	Thu Jun  4 03:07:55 1987
--- /usr/local/okcl/c/unixsave.c	Mon Nov 16 11:30:52 1987
***************
*** 308,313 ****
--- 308,322 ----
  	fclose(save);
  }
  
+ #ifdef BSD4_3
+ fclose_notstderr(iop)
+ FILE *iop;
+ {
+      if (iop != stderr)
+ 	  fclose (iop);
+ }
+ #endif
+ 
  Lsave()
  {
  	char filename[256];
***************
*** 318,323 ****
--- 327,335 ----
  /*
  	_cleanup();
  */
+ #ifdef BSD4_3
+ 	_fwalk (fclose_notstderr);
+ #else
  	{
  		FILE *p;
  		int nfile;
***************
*** 330,335 ****
--- 342,348 ----
  		for (p = &_iob[3];  p < &_iob[nfile];  p++)
  			fclose(p);
  	}
+ #endif
  	memory_save(kcl_self, filename);
  /*
  	_exit(0);


---
Wolfgang Rupprecht	UUCP: mit-eddie!mgm.mit.edu!wolfgang
				(or) mirror!mit-mgm!wolfgang
			ARPA: wolfgang@mgm.mit.edu (IP addr 18.82.0.114)