latest beta + MinGW 3.1.0-1 = minor fix needed

MinGW 3.1.0-1 is currently the latest official MinGW release, therefore 
I tried to compile the latest R beta (rsync today). There is one minor 
problem when trying that:

gcc  -O2 -Wall -pedantic -I../../include -I../../gnuwin32 
-DHAVE_CONFIG_H  -c internet.c -o internet.o
In file included from internet.c:880:
sock.h:23: conflicting types for `ssize_t'
n:/MinGW/include/sys/types.h:119: previous declaration of `ssize_t'
make[3]: *** [internet.o] Error 1

(This is in src/modules/internet)

Obviously ssize_t was added to the system types in MinGW. The 
corresponding files:

MinGW: sys/types.h:

#ifndef _SSIZE_T_
#define _SSIZE_T_
typedef long _ssize_t;

R: src/modules/internet/sock.h

#ifdef Win32
typedef int ssize_t;
#endif

There are two possible fixes for this:
1) make sock.h use long instead of int.
2) use something like:
#if defined Win32 && !defined _SSIZE_T_

I don't think either fix is "better" in any way, but personally I'd go 
for the "long" (for Win32 I don't see any difference, but for Win64 fix 
2) may be less safe?).

Configuration:
MinGW 3.1.0-1
w32api 2.4

Simon

Some thoughts:
The first suggested fix seems to be forbidden during code freeze 
(might break other platforms, doesn't it?), the second one puts one 
more OS dependence into the sources.

Simon Urbanek wrote:

There are two possible fixes for this:
1) make sock.h use long instead of int.
2) use something like:
#if defined Win32 && !defined _SSIZE_T_

Simon,

it does not happen with MinGW's gcc-3.3.1 candidate release we (Duncan
and myself) use for compiling R base + packages.
The gcc-3.3.1 candidate is older (2003-08-07) than the official
release (2003-09-14, gcc-3.2.?) you are using, though.

Some thoughts:
The first suggested fix seems to be forbidden during code freeze
(might break other platforms, doesn't it?), the second one puts one
more OS dependence into the sources.
Another solution is to hope for an official MinGW release of gcc-3.3.1
with the change you are reporting reverted before the release of
R-1.8.1 (of course one should document not to use MinGW 3.1.0-1 as is).

Uwe Ligges <ligges@statistik.uni-dortmund.de> writes:

Simon Urbanek wrote:

There are two possible fixes for this:
1) make sock.h use long instead of int.
2) use something like:
#if defined Win32 && !defined _SSIZE_T_

Fixes that are NOPs on non-broken setups should be allowable...
Version 2) looks most correct to me: We want to use the system
definition wherever it is available, and it would obviously only
change anything on Win32.

It compiles fine in 3.3.1 with change 2).  I'll commit the change.  I
don't have 3.2.3 installed to test there, so could you try a build
soon Simon?