Re: [WIP] mingw64 related changes to Cygwin configure and other assorted files with departed w32api/mingw
On Nov 12 22:31, Christopher Faylor wrote: On Mon, Nov 12, 2012 at 07:02:57PM -0500, Christopher Faylor wrote: On Mon, Nov 12, 2012 at 10:50:23PM +0100, Corinna Vinschen wrote: I'm a bit puzzled about the necessity of some of the changes to source files. Yaakov's Fedora 17 version of the headers is supposedly cut from the mingw64 trunk on 2012-10-16, while JonY's official headers have an upload date of 2012-10-18. They should be practically identical. Why do I not see any problems to build CVS HEAD?!? You can keep asking me this question but I don't really have an answer. Since I don't run Fedora, I'm not going to install it to figure it out. I just checked and there's no difference in the header files at all. Yaakov's version == JonY's version. Actually, an idea came to me in the thinking room that this might be due to the fact that my windows headers may not be considered to be system headers since they aren't in a preinstalled location. I know that gcc can be more lax about redefine symbols in some situations when dealing with system headers. Maybe that's it. Looks like it. The w32api headers are system headers so -isystem rather than -idirafter should show a better result without requiring any of the source file changes. Especially having to define _WIN32 in winlean.h and winsup.h looks pretty wrong. I would also like to keep the ifndef/define brackets in the headers since #ifndef _CYGWIN_IF_H_ #define _CYGWIN_IF_H_ can be tested for in other headers while #pragma once can not. Corinna -- Corinna Vinschen Please, send mails regarding Cygwin to Cygwin Project Co-Leader cygwin AT cygwin DOT com Red Hat
Re: [WIP] mingw64 related changes to Cygwin configure and other assorted files with departed w32api/mingw
On Tue, Nov 13, 2012 at 10:33:01AM +0100, Corinna Vinschen wrote: On Nov 12 22:31, Christopher Faylor wrote: On Mon, Nov 12, 2012 at 07:02:57PM -0500, Christopher Faylor wrote: On Mon, Nov 12, 2012 at 10:50:23PM +0100, Corinna Vinschen wrote: I'm a bit puzzled about the necessity of some of the changes to source files. Yaakov's Fedora 17 version of the headers is supposedly cut from the mingw64 trunk on 2012-10-16, while JonY's official headers have an upload date of 2012-10-18. They should be practically identical. Why do I not see any problems to build CVS HEAD?!? You can keep asking me this question but I don't really have an answer. Since I don't run Fedora, I'm not going to install it to figure it out. I just checked and there's no difference in the header files at all. Yaakov's version == JonY's version. Actually, an idea came to me in the thinking room that this might be due to the fact that my windows headers may not be considered to be system headers since they aren't in a preinstalled location. I know that gcc can be more lax about redefine symbols in some situations when dealing with system headers. Maybe that's it. Looks like it. The w32api headers are system headers so -isystem rather than -idirafter should show a better result without requiring any of the source file changes. Maybe I can use -isystem with ccwraper. I'd previously gotten things working without the wrapper, using idirafter so that's what I stuck with. However, the wrapper may now allow just always including the windows headers last. Especially having to define _WIN32 in winlean.h and winsup.h looks pretty wrong. I would also like to keep the ifndef/define brackets in the headers since #ifndef _CYGWIN_IF_H_ #define _CYGWIN_IF_H_ can be tested for in other headers while #pragma once can not. I think that testing for BLAH_DECLARED for individual definitions is a much better way to see if something is defined than relying on an implementation detail like _CYGWIN_IF_H. Those are not supposed to be a published interface. In theory anyone could change those to something like: #ifndef _CYGWIN_IF_INCLUDED #define _CYGWIN_IF_INCLUDED and they would have been within their rights to do that. I always felt that it was wrong to be testing for those guards, even when I was doing it. However, I won't check those changes in. cgf
Re: [WIP] mingw64 related changes to Cygwin configure and other assorted files with departed w32api/mingw
On Nov 13 12:39, Christopher Faylor wrote: On Tue, Nov 13, 2012 at 10:33:01AM +0100, Corinna Vinschen wrote: I would also like to keep the ifndef/define brackets in the headers since #ifndef _CYGWIN_IF_H_ #define _CYGWIN_IF_H_ can be tested for in other headers while #pragma once can not. I think that testing for BLAH_DECLARED for individual definitions is a much better way to see if something is defined than relying on an implementation detail like _CYGWIN_IF_H. Sure. This might not be of much interest for the headers in the include/cygwin subdir, but there are applications out there which test for such header defines, and there are also applications using system-specific headers liberally. Out of curiosity I had a look and none of the Linux/glibc headers seem to use #pragma once either for some reason. An alternative might be something like #pragma once #define _CYGWIN_IF_H_ It would introduce the new pragma and keep the definition available. Corinna -- Corinna Vinschen Please, send mails regarding Cygwin to Cygwin Project Co-Leader cygwin AT cygwin DOT com Red Hat
Re: [WIP] mingw64 related changes to Cygwin configure and other assorted files with departed w32api/mingw
On Nov 13 13:34, Christopher Faylor wrote: On Tue, Nov 13, 2012 at 12:39:00PM -0500, Christopher Faylor wrote: Maybe I can use -isystem with ccwraper. I'd previously gotten things working without the wrapper, using idirafter so that's what I stuck with. However, the wrapper may now allow just always including the windows headers last. Yep. Adding the windows headers directories dead last as -isystem means that none of my header file changes are needed, *except* for the #define _WIN32. I wonder why you don't need those. My (i.e., Yaakov's) cross compiler doesn't define _WIN32. % /cygwin/bin/i686-cygwin-gcc --version i686-cygwin-gcc (GCC) 4.5.3 20110428 (Fedora Cygwin 4.5.3-4) Copyright (C) 2010 Free Software Foundation, Inc. This is free software; see the source for copying conditions. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. % /cygwin/bin/i686-cygwin-gcc -dD -E -xc /dev/null | grep WIN #define __WINT_TYPE__ unsigned int #define __WINT_MAX__ 4294967295U #define __WINT_MIN__ 0U #define __SIZEOF_WINT_T__ 4 #define __CYGWIN32__ 1 #define __CYGWIN__ 1 So, except for that, mystery solved. My new, smaller diff is attached. Unless you have objections, I'll be checking this in. I have no objections against the patch in general, but I'd rather like to test it first. I'd like to try to figure out what the _WIN32 problem is first, and I'd like to give it a try in the 64 bit scenario. Thanks, Corinna -- Corinna Vinschen Please, send mails regarding Cygwin to Cygwin Project Co-Leader cygwin AT cygwin DOT com Red Hat
Re: [WIP] mingw64 related changes to Cygwin configure and other assorted files with departed w32api/mingw
On Tue, Nov 13, 2012 at 07:47:32PM +0100, Corinna Vinschen wrote: On Nov 13 13:34, Christopher Faylor wrote: On Tue, Nov 13, 2012 at 12:39:00PM -0500, Christopher Faylor wrote: Maybe I can use -isystem with ccwraper. I'd previously gotten things working without the wrapper, using idirafter so that's what I stuck with. However, the wrapper may now allow just always including the windows headers last. Yep. Adding the windows headers directories dead last as -isystem means that none of my header file changes are needed, *except* for the #define _WIN32. I wonder why you don't need those. My (i.e., Yaakov's) cross compiler doesn't define _WIN32. % /cygwin/bin/i686-cygwin-gcc --version i686-cygwin-gcc (GCC) 4.5.3 20110428 (Fedora Cygwin 4.5.3-4) Copyright (C) 2010 Free Software Foundation, Inc. This is free software; see the source for copying conditions. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. % /cygwin/bin/i686-cygwin-gcc -dD -E -xc /dev/null | grep WIN #define __WINT_TYPE__ unsigned int #define __WINT_MAX__ 4294967295U #define __WINT_MIN__ 0U #define __SIZEOF_WINT_T__ 4 #define __CYGWIN32__ 1 #define __CYGWIN__ 1 So, except for that, mystery solved. My new, smaller diff is attached. Unless you have objections, I'll be checking this in. I have no objections against the patch in general, but I'd rather like to test it first. I'd like to try to figure out what the _WIN32 problem is first, and I'd like to give it a try in the 64 bit scenario. Ok, I'm thoroughly confused. I made another few unrelated changes to the wrapper script and now I don't need the _WIN32 setting either. cgf
[WIP] mingw64 related changes to Cygwin configure and other assorted files with departed w32api/mingw
I decided over the weekend to port over configury changes that I made to Cygwin's now-out-of-date GIT repository. These changes basically just cleaned up some of the configure scripts and made it easier to pinpoint where windows headers and libraries come from by adding a --with-windows-headers and --with-windows-libs options. However, some of the assumptions made for the git repository weren't really valid for the CVS repository so there was a fair amount of work involved. I thought that I'd do this so I could easily get up-and-running with the MinGW64 stuff but I ran into some problems building things with gentoo's MinGW64 implementation. So, I switched to using the files from the Cygwin release. As I mentioned in cygwin-developers, getting the most recent version of mingw64 stuff working required making some changes to some Cygwin source files. Most of the changes just involved #undef'ing constants defined in Windows headers. Still, I was surprised that these hadn't already been handled since I thought I was behind the times by still using the Mingw32 stuff. Anyway, is a summary of the changes I've made to files is below. I'll be doing appropriate ChangeLogs too, of course. I've also attached the patch. This is a heads up in case this conflicted materially with any of the w64 development. cgf --- winsup/Makefile.common Gutted. Revamped to just include common definitions and macros. --- winsup/acinclude.m4 Gutted. Now defines: AC_WINDOWS_HEADERS AC_WINDOWS_LIBS AC_CYGWIN_INCLUDES --- winsup/configure.cygwin New file. Does common stuff for cygwin configure files. --- **/autogen.sh New script. Regenerates configure and aclocal.m4. --- winsup/ccwrap New script. Used to run gcc/g++ using defined cygwin/newlib/windows locations. CCWRAP_VERBOSE=1 environment variable shows full expanded path to compiler command line. (I wish gcc had a way to say Don't use this built-in header directory) --- winsup/c++wrap C++ front-end to ccwrap - because CVS doesn't like symlinks. --- **/configure.in Revamped to use new acinclude.m4 directives and to eliminate cruft. --- **/Makefile.in Add .E target to generate macro expanded version of source file. Eliminate VERBOSE= option. Always print everything. Revamp to use new configure framework and c*wrap. Regenerate Makefile if Makefile.in, config.status change. Use WINDOWS_LIBDIR and WINDOWS_HEADERS to find windows stuff. Remove unneeded SHELL:= setting. --- cygwin/Makefile.in Add the very important datarootdir setting! Revamp (slightly) CFLAGS/CXXFLAGS settings. These two things should really be separate but I've gotten into the habit of saying: make CFLAGS=-g to cause things to not be built with optimization. Handle that case but otherwise decouple CFLAGS from CXXFLAGS. Generate tlsoffsets.h in the build directory. --- cygwin/mkvers.sh Fix version.h and winver.rc generation to not assume in-tree windows files. --- cygwin/winlean.h #define _WIN32 cygwin/winsup.h #define _WIN32 --- cygwin/include/netdb.h #undef a bunch of windows-defined constants. --- cygwin/include/cygwin/if.h Use #pragma once --- cygwin/include/cygwin/in.h Use #pragma once Use _IN_PORT_DECLARED to guard against define of in_port_t. --- cygwin/include/cygwin/in6.h Use #pragma once Use _IN_PORT_DECLARED to guard against define of in_port_t. --- cygwin/include/cygwin/socket.h #undef a bunch of windows-defined constants. --- cygwin/include/sys/socket.h #pragma once --- cygwin/include/sys/termios.h #undef FIONBIO before defining it. --- cygwin/minires-os-if.c #define __INSIDE_CYGWIN__ #include sys/time.h Delete some #undef's that are now in header files. --- cygwin/libc/minires.h #pragma once #undef some windows-defined constants. --- utils/Makefile.in Lots of changes. Assume that we always have mingw libraries available for cygcheck. Set CCWRAP variables to control where to find header files for various utilities. --- utils/dump_setup.cc Include zlib.h directly. --- utils/path.cc #define __CRT__NO_INLINE to avoid an error from a windows header. Index: Makefile.common === RCS file: /cvs/uberbaum/winsup/Makefile.common,v retrieving revision 1.61 diff -u -p -r1.61 Makefile.common --- Makefile.common 7 Nov 2012 16:32:07 - 1.61 +++ Makefile.common 12 Nov 2012 19:53:05 - @@ -8,154 +8,23 @@ # Cygwin license. Please consult the file CYGWIN_LICENSE for # details. -# This makefile requires GNU make. +define justdir +$(patsubst %/,%,$(dir $1)) +endef -CFLAGS_COMMON:=-Wall -Wstrict-aliasing -Wwrite-strings -fno-common -pipe -fbuiltin -fmessage-length=0 -D_SDKDDKVER_H -MALLOC_DEBUG:=#-DMALLOC_DEBUG -I/cygnus/src/uberbaum/winsup/cygwin/dlmalloc -MALLOC_OBJ:=#/cygnus/src/uberbaum/winsup/cygwin/dlmalloc/malloc.o - -override srcdir:=${shell cd $(srcdir); pwd} -ifneq (,${filter-out /%,$(srcdir)}) -updir:=$(srcdir)/.. -updir1:=$(updir)/.. -else -
Re: [WIP] mingw64 related changes to Cygwin configure and other assorted files with departed w32api/mingw
On Nov 12 15:02, Christopher Faylor wrote: I decided over the weekend to port over configury changes that I made to Cygwin's now-out-of-date GIT repository. These changes basically just cleaned up some of the configure scripts and made it easier to pinpoint where windows headers and libraries come from by adding a --with-windows-headers and --with-windows-libs options. However, some of the assumptions made for the git repository weren't really valid for the CVS repository so there was a fair amount of work involved. I thought that I'd do this so I could easily get up-and-running with the MinGW64 stuff but I ran into some problems building things with gentoo's MinGW64 implementation. So, I switched to using the files from the Cygwin release. As I mentioned in cygwin-developers, getting the most recent version of mingw64 stuff working required making some changes to some Cygwin source files. Most of the changes just involved #undef'ing constants defined in Windows headers. Still, I was surprised that these hadn't already been handled since I thought I was behind the times by still using the Mingw32 stuff. Anyway, is a summary of the changes I've made to files is below. I'll be doing appropriate ChangeLogs too, of course. I've also attached the patch. This is a heads up in case this conflicted materially with any of the w64 development. Looks good at first sight. I see only one place which won't work for 64 bit, the ccwrap script. It uses i686-pc-cygwin-gcc/g++ hardcoded, but it should use something like ${target_cpu}-pc-cygwin-gcc/g++ to make it platform independent. With a matching change, I can give it a try on 64 bit tomorrow. I'm a bit puzzled about the necessity of some of the changes to source files. Yaakov's Fedora 17 version of the headers is supposedly cut from the mingw64 trunk on 2012-10-16, while JonY's official headers have an upload date of 2012-10-18. They should be practically identical. Why do I not see any problems to build CVS HEAD?!? Corinna -- Corinna Vinschen Please, send mails regarding Cygwin to Cygwin Project Co-Leader cygwin AT cygwin DOT com Red Hat