suspect error in regex.c

suspect error in regex.c

[Francesco Potorti`]

I don't know how regex.c works, but this looks strange to me (I
discovered it because of a warning issued by MSVC on a report I got
about etags, which uses regex.c):

== src/regex.c ==

1837:typedef unsigned regnum_t;

in regex_compile:
2245:  regnum_t regnum = 0;
2847: 		COMPILE_STACK_TOP.regnum = shy ? -regnum : regnum;

Notice that an unsigned value is negated, then assigned to an unsigned
field.

later in regex_compile:
2897:		regnum_t this_group_regnum;
2906:		this_group_regnum = COMPILE_STACK_TOP.regnum;
2914:		if (this_group_regnum <= MAX_REGNUM && this_group_regnum > 0)

That is, the same unsigned value is compared against MAX_REGNUM and 0.
In principle, the code could be correct, but it does sound strange to me
that such a particular programming practice is not even commented.
Looks suspiciously like a bug to me.

Re: suspect error in regex.c

[Richard Stallman]

Does this patch work right?


*** regex.c.~1.179.~	Thu Sep  5 23:00:55 2002
--- regex.c	Sat Sep  7 15:15:41 2002
***************
*** 1834,1840 ****
  
  /* But patterns can have more than `MAX_REGNUM' registers.  We just
     ignore the excess.  */
! typedef unsigned regnum_t;
  
  
  /* Macros for the compile stack.  */
--- 1834,1840 ----
  
  /* But patterns can have more than `MAX_REGNUM' registers.  We just
     ignore the excess.  */
! typedef int regnum_t;
  
  
  /* Macros for the compile stack.  */

Re: suspect error in regex.c

[Francesco Potorti`]

   Does this patch work right?
   
   *** regex.c.~1.179.~	Thu Sep  5 23:00:55 2002
   ! typedef unsigned regnum_t;
...
   ! typedef int regnum_t;

It works for etags.  Did not try with emacs.