fdtable

fdtable

[Peter Stephenson]

In 2.6-beta17, fdtable[] is a char array, and there are tests like

		    if (fdtable[i] < 0)
			fdtable[i]--;

for which gcc on IRIX is complaining about limited range of data type.
Shouldn't fdtable[] be `signed char'? (Yes.)

*** Src/globals.h.sc	Tue May  7 10:09:50 1996
--- Src/globals.h	Tue May  7 10:10:13 1996
***************
*** 397,403 ****
   * table is not used.  A table element is set by movefd and cleard *
   * by zclose.                                                      */
  
! EXTERN char fdtable[OPEN_MAX];
  
  /* input fd from the coprocess */
  
--- 397,403 ----
   * table is not used.  A table element is set by movefd and cleard *
   * by zclose.                                                      */
  
! EXTERN signed char fdtable[OPEN_MAX];
  
  /* input fd from the coprocess */
  
-- 
Peter Stephenson <pws@ifh.de>       Tel: +49 33762 77366
WWW:  http://www.ifh.de/~pws/       Fax: +49 33762 77330
Deutches Electronen-Synchrotron --- Institut fuer Hochenergiephysik Zeuthen
DESY-IfH, 15735 Zeuthen, Germany.

Re: fdtable

[Zefram]

>In 2.6-beta17, fdtable[] is a char array, and there are tests like
>
>		    if (fdtable[i] < 0)
>			fdtable[i]--;
>
>for which gcc on IRIX is complaining about limited range of data type.
>Shouldn't fdtable[] be `signed char'? (Yes.)

Careful!  The `signed' keyword does not exist in K&R C.  We should
either avoid its usage altogether, or do a feature test for it.  I
think all K&R compilers have char signed by default, so this should
work.

-zefram

Re: fdtable

[Peter Stephenson]

> >Shouldn't fdtable[] be `signed char'? (Yes.)
> 
> Careful!  The `signed' keyword does not exist in K&R C.  We should
> either avoid its usage altogether, or do a feature test for it.  I
> think all K&R compilers have char signed by default, so this should
> work.

OK, here's a configure version.  I haven't rerun autoconf; that needs
to be done for this to take effect.

By the way, the previous test in configure, for casting signed to
unsigned, isn't being cached properly.  I think that's because of the
name of the variable.

*** config.h.in.sig	Tue May  7 15:42:38 1996
--- config.h.in	Tue May  7 15:57:52 1996
***************
*** 96,101 ****
--- 96,104 ----
  /* Define to 1 if compiler incorrectly cast signed to unsigned */
  #undef BROKEN_SIGNED_TO_UNSIGNED_CASTING
  
+ /* Define to 1 if compiler handles `signed' keyword */
+ #undef HAVE_SIGNED_KEYWORD
+ 
  /* Define if your system defines TIOCGWINSZ in sys/ioctl.h.  */
  #undef GWINSZ_IN_SYS_IOCTL
  
*** configure.in.sig	Fri May  3 20:54:33 1996
--- configure.in	Tue May  7 16:00:39 1996
***************
*** 154,159 ****
--- 154,167 ----
    AC_DEFINE(BROKEN_SIGNED_TO_UNSIGNED_CASTING)
  fi
  
+ dnl  Checking if complier handles `signed' keyword
+ AC_CACHE_CHECK(if the signed keyword is handled, zsh_cv_signed_keyword,
+ [AC_TRY_COMPILE(, [signed char foo; foo = -1;],
+ zsh_cv_signed_keyword=yes, zsh_cv_signed_keyword=no)])
+ if test $zsh_cv_signed_keyword = yes; then
+   AC_DEFINE(HAVE_SIGNED_KEYWORD)
+ fi
+ 
  dnl ------------------
  dnl CHECK FOR PROGRAMS
  dnl ------------------
*** Src/globals.h.sig	Tue May  7 15:43:49 1996
--- Src/globals.h	Tue May  7 15:56:39 1996
***************
*** 397,403 ****
--- 397,408 ----
   * table is not used.  A table element is set by movefd and cleard *
   * by zclose.                                                      */
  
+ #ifdef HAVE_SIGNED_KEYWORD
+ EXTERN signed char fdtable[OPEN_MAX];
+ #else
+ /* If `signed' does not exist, we assume char's are signed anyway. */
  EXTERN char fdtable[OPEN_MAX];
+ #endif
  
  /* input fd from the coprocess */
  
-- 
Peter Stephenson <pws@ifh.de>       Tel: +49 33762 77366
WWW:  http://www.ifh.de/~pws/       Fax: +49 33762 77330
Deutches Electronen-Synchrotron --- Institut fuer Hochenergiephysik Zeuthen
DESY-IfH, 15735 Zeuthen, Germany.

Re: fdtable

[Peter Stephenson]

I wrote:
> Shouldn't fdtable[] be `signed char'? (Yes.)

Zefram just pointed out the simplest thing to do was simply to define
signed to nothing if it isn't supported.  Here's the third attempt.
I'd be obliged if somebody using a K&R compiler could have a go at it.
(I did try it by altering the cached value to `no'.)

*** config.h.in.sig	Tue May  7 15:42:38 1996
--- config.h.in	Tue May  7 16:43:33 1996
***************
*** 96,101 ****
--- 96,104 ----
  /* Define to 1 if compiler incorrectly cast signed to unsigned */
  #undef BROKEN_SIGNED_TO_UNSIGNED_CASTING
  
+ /* Define to nothing if compiler does not handle `signed' keyword */
+ #undef signed
+ 
  /* Define if your system defines TIOCGWINSZ in sys/ioctl.h.  */
  #undef GWINSZ_IN_SYS_IOCTL
  
*** configure.in.sig	Fri May  3 20:54:33 1996
--- configure.in	Tue May  7 16:40:11 1996
***************
*** 154,159 ****
--- 154,167 ----
    AC_DEFINE(BROKEN_SIGNED_TO_UNSIGNED_CASTING)
  fi
  
+ dnl  Checking if complier handles `signed' keyword
+ AC_CACHE_CHECK(if the signed keyword is handled, zsh_cv_signed_keyword,
+ [AC_TRY_COMPILE(, [signed char foo; foo = -1;],
+ zsh_cv_signed_keyword=yes, zsh_cv_signed_keyword=no)])
+ if test $zsh_cv_signed_keyword = no; then
+   AC_DEFINE(signed, )
+ fi
+ 
  dnl ------------------
  dnl CHECK FOR PROGRAMS
  dnl ------------------
*** Src/globals.h.sig	Tue May  7 15:43:49 1996
--- Src/globals.h	Tue May  7 16:41:01 1996
***************
*** 397,403 ****
   * table is not used.  A table element is set by movefd and cleard *
   * by zclose.                                                      */
  
! EXTERN char fdtable[OPEN_MAX];
  
  /* input fd from the coprocess */
  
--- 397,403 ----
   * table is not used.  A table element is set by movefd and cleard *
   * by zclose.                                                      */
  
! EXTERN signed char fdtable[OPEN_MAX];
  
  /* input fd from the coprocess */
  
-- 
Peter Stephenson <pws@ifh.de>       Tel: +49 33762 77366
WWW:  http://www.ifh.de/~pws/       Fax: +49 33762 77330
Deutches Electronen-Synchrotron --- Institut fuer Hochenergiephysik Zeuthen
DESY-IfH, 15735 Zeuthen, Germany.

Re: fdtable

[Zoltan Hidvegi]

[-- Attachment #1: Type: application/pgp, Size: 4388 bytes --]

Macros (Was: fdtable)

[Bas V. de Bakker]

Peter Stephenson <pws@ifh.de> writes:
  
> + /* Define to nothing if compiler does not handle `signed' keyword */
> + #undef signed
> + 

There's already another fix by Zoltan, so the following remark no
longer applies to this case, but I'd like to make a general remark
here:

When reading other people's code, I've regularly been bitten by the
fact that the code was not what it looked like, due to a #define in
one of the so many header files that was (directly or indirectly)
included.  The potential for errors is much less if one keeps all
macros uppercase.  So create a SIGNED macro that is either defined to
signed or to nothing, like EXTERN.

The same is true of things like

#define dirent direct
#define setpgrp setpgid

which are currently used.

Bas.

Re: Macros (Was: fdtable)

[Zefram]

>When reading other people's code, I've regularly been bitten by the
>fact that the code was not what it looked like, due to a #define in
>one of the so many header files that was (directly or indirectly)
>included.  The potential for errors is much less if one keeps all
>macros uppercase.  So create a SIGNED macro that is either defined to
>signed or to nothing, like EXTERN.

In the case of signed, I disagree.  The signed keyword usually has no
effect at all, so #defining it to nothing makes almost no difference to
its meaning.  const, similarly, is purely a modifier, and #defining it
to nothing makes no difference to the meaning of the code, provided
it's legal to start with.  The EXTERN macro is uppercase because it
doesn't simply mean extern; it means extern in most source files, but
it is different in one.

-zefram