This is the mail archive of the cygwin mailing list for the Cygwin project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: converting from -mno-cygwin


On 4/27/2016 08:32, Lee wrote:
> On 4/26/16, JonY  wrote:
>> On 4/27/2016 05:08, Lee wrote:
>>> Questions:
>>>
>>> How to tell if I should be using libwinpthread or pthread?  I had no
>>> idea so installed both:
>>> /usr/i686-w64-mingw32/sys-root/mingw/bin
>>> $ ls -l *hread*
>>> -rwxr-xr-x 1 root None 47635 Apr  7 08:54 libwinpthread-1.dll
>>> -rwxr-xr-x 1 root None 65024 Jul  6  2013 pthreadGC2.dll
>>>
>>>
>>
>> pthreadGC2 is a compatibility left over.
> 
> So I should uninstall it?  I have both installed now:
> mingw64-i686-pthreads                 20100619-5                    OK
> mingw64-i686-winpthreads              4.0.6-1                       OK
> 

Sure, its there because pthreadGC was around longer than winpthreads.
Notice there is no import library for pthreadGC, it is only provided for
compatibility with any app that still use it.

> and maybe it's a problem.  I haven't tracked it down yet, but GNUmakefile.in has
> # PThreads library, if needed.
> PTHREAD_LIB  = @PTHREAD_ONLY@@PTHREAD_LIB@
> 
> which, after running autoheader & autoconf, creates a GNUmakefile with
> # PThreads library, if needed.
> PTHREAD_LIB  = -lpthreadGC2
> 
> which doesn't work :(
> 

Fix your configure.ac assumptions.

>>> Is there a standard way to figure out if the compiler is gcc-v3 with
>>> the -mno-cygwin flag set?
>>
>> No, don't do this, it'd turn into a giant hairball fast.
> 
> Not really.. it's just a few places that I've had to change the source
> code to deal with the differences between cygwin 1.5 + gcc v3 &
> whatever the current cygwin is + i686-w64-mingw32-gcc
> 

You'd get into a bigger hairball when you involve 64bit Cygwin where
long is 64bit unlike the rest of the Windows world.

>>> I had to make a few changes to the code to get this far & I'd prefer
>>> to have the changes wrapped inside an #IFDEF or something.  For
>>> example, I just commented out the include since it conflicts with
>>> something
>>>
>>> #ifdef __MINGW32__
>>> /* -LR- #include "cygwin.h" */
>>> /* -LR- const char cygwin_h_rcs[] = CYGWIN_H_VERSION; */
>>> #endif
>>>
>>> Under cygwin 1.5, gcc -mno-cygwin  requires cygwin.h to be included.
>>> Using i686-w64-mingw32-gcc if cygwin.h is inculded gcc barfs with a
>>> conflicting definition of [i don't remember].
>>> It'd be nice if I could build using the old or new method without
>>> having to change the source code, so I'm guessing I want some kind of
>>> ifdef wrapper for the include??
>>>
>>
>> What are you even trying to do? You shouldn't mix different runtimes.
> 
> I'm not mixing runtimes.  I'm trying to keep it so that the program
> continues to build under the old cygwin 1.5/gcc -mno-cygwin as well as
> building under the current cygwin/i686-w64-mingw32-gcc cross-compiler.
> Since I have to do different things depending on if I'm using the
> cross compiler or the old gcc -mno-cygwin I'm hoping there's a flag or
> something I can use so the exact same source code works with either
> build system.
> 

Using Cygwin headers in purely win32 program is wrong. I understand you
are simply trying to get Cygwin version strings, it'll just end up as a
compile error if the toolchain isn't Cygwin hosted. You can try using
autoconf to run "uname -a" instead and grab that instead.



Attachment: signature.asc
Description: OpenPGP digital signature


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]