Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!cs.utexas.edu!usc!samsung!munnari.oz.au!cluster!metro!natmlab.dap.csiro.au!ditsyda!evans
From: evans@ditsyda.oz (Bruce Evans)
Newsgroups: comp.os.minix
Subject: Re: More compilation problems...
Message-ID: <2439@ditsyda.oz>
Date: 1 Feb 90 18:14:10 GMT
References: <712@augean.OZ>
Reply-To: evans@ditsyda.oz (Bruce Evans)
Organization: CSIRO DIT Sydney, Australia
Lines: 49

In article <712@augean.OZ> cagney@chook.ua.oz (Andrew Cagney - aka Noid) writes:
>The prototype for getpwuid in <pwd.h> vis
>	struct passwd *getpwuid(int __uid)
>clashes with the actual code in getpwent.c
>	struct passwd *getpwuid(uid_t __uid)
>				^^^^^
>uid_t is declared unsigned short in <sys/types.h>.

Actually getpwent.c has

	struct passwd *getpwuid(__uid)
	uid_t __uid;		/* NB old-style definition */

and ANSI-according-to-gcc says this is OK. The int in the prototype says
that an int is passed and the uid_t in the definition says it is regarded
as a uid_t. Only with the new-style definition must the types match exactly.

Andrew must have this problem from using Turbo C, which apparently treats
all definitions as new-style (non-ANSI or old-ANSI?).

The prototype has to have an int because that is all old compilers know
how to pass. It gets worse with 32 bit ints. Then a uid_t in the prototype
would allow an ANSI compiler to pass a different number of bytes than an
old compiler. Unsigned doesn't work, because 32-bit ANSI compilers promote
a 16-bit unsigned short to an int.

I think the correct fix for Turbo C is to declare the argument to getpwuid
as an int.

There are other functions in the library which mis-match their prototypes
in the same way. E.g., chmod, creat and umask. (Un)fortunately, the problem
is invisible because the file with the prototype is not included.

>Finally a POSIX question, do programs that include <signal.h> also
>need to include <sys/types.h> (Kermit doesn't)?

Signal.h is required to be self-supporting by ANSI (I think :-)). It is
slightly broken by requiring pid_t from sys/types.h for the prototype to
kill(). However, kill() is a POSIX function and the prototype is only
visible when _POSIX_SOURCE is defined. It is not clear to me whether
defining _POSIX_SOURCE is allowed to break ANSI headers. (Kermit hasn't
heard of POSIX but the makefile has been modified for Minix to supply
the define.) I think Kermit *is* required to include sys/types.h and
don't like it.

Kermit is also broken by including pwd.h without sys/types.h. At least
pwd.h is owned by POSIX.
-- 
Bruce Evans		evans@ditsyda.oz.au