Date: Tue, 19 Feb 2013 06:52:34 -0800 From: mdf@FreeBSD.org To: Borja Marcos <borjam@sarenet.es> Cc: freebsd-hackers@freebsd.org Subject: Re: Stupid question about integer sizes Message-ID: <CAMBSHm86GB1No3Zc97KYhDnH9nkUEuhawJu82WwScJE%2BqvS9Bg@mail.gmail.com> In-Reply-To: <77B93E04-9254-4EF1-8BDB-ED9214CEC2BC@sarenet.es> References: <77B93E04-9254-4EF1-8BDB-ED9214CEC2BC@sarenet.es>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, Feb 19, 2013 at 5:11 AM, Borja Marcos <borjam@sarenet.es> wrote: > > Hello, > > I'm really sorry if this is a stupid question, but as far as I know, u_int64_t defined in /usr/include/sys/types.h should *always* be > a 64 bit unsigned integer, right? > > Seems there's a bug (or I need more and stronger coffee). Compiling a program on a 64 bit system with -m32 gets the 64 bit integer types wrong. > > % cat tachin.c > #include <sys/types.h> > #include <stdio.h> > > > main() > { > printf("sizeof uint64_t = %d\n", sizeof(uint64_t)); > printf("sizeof u_int64_t = %d\n", sizeof(u_int64_t)); > } > > > > uname -a > FreeBSD splunk 9.1-RELEASE FreeBSD 9.1-RELEASE #14: Wed Jan 23 17:24:05 CET 2013 root@splunk:/usr/obj/usr/src/sys/SPLUNK amd64 > > % gcc -o tachin tachin.c > % ./tachin > sizeof uint64_t = 8 > sizeof u_int64_t = 8 > > % ./tachin > sizeof uint64_t = 4 <======= WRONG!! > sizeof u_int64_t = 4 <======= WRONG!! > > The same happens with clang. > > % clang -m32 -o tachin tachin.c > tachin.c:5:1: warning: type specifier missing, defaults to 'int' [-Wimplicit-int] > main() > ^~~~ > 1 warning generated. > % ./tachin > sizeof uint64_t = 4 <======= WRONG!! > sizeof u_int64_t = 4 <======= WRONG!! > > > if I do the same on a i386 system (8.2-RELEASE, but it should be irrelevant) the u_int64 types have the correct size. > > %gcc -o tachin tachin.c > %./tachin > sizeof uint64_t = 8 > sizeof u_int64_t = 8 > > > > > > Am I missing anything? Seems like a too stupid problem to be a real bug. Sorry if I am wrong. > Last I knew -m32 still wasn't quite supported on 9.1. This is fixed in CURRENT. The problem in in the machine-dependent headers. sys/types.h includes sys/_types.h which includes machine/_types.h. But the machine alias can only point to one place; it's the amd64 headers since you're running the amd64 kernel. sys/amd64/include/_types.h defines __uint64_t as unsigned long, which is correct in 64-bit mode but wrong in 32-bit mode. On CURRENT there is a merge x86/_types.h which uses #ifdef guards to define __uint64_t as unsigned long or unsigned long long, depending on __LP64__, so that the size is correct on a 32-bit compiler invocation. Cheers, matthew
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAMBSHm86GB1No3Zc97KYhDnH9nkUEuhawJu82WwScJE%2BqvS9Bg>