[FFmpeg-devel] [PATCH 1/5] w32pthreads: always use Vista+ API, drop XP support

[wm4]

On Thu, 21 Dec 2017 22:57:42 -0300
James Almer <jamrial at gmail.com> wrote:

> On 12/21/2017 10:49 PM, wm4 wrote:
> > On Thu, 21 Dec 2017 22:05:11 -0300
> > James Almer <jamrial at gmail.com> wrote:
> >   
> >> On 12/21/2017 9:40 PM, wm4 wrote:  
> >>> On Thu, 21 Dec 2017 21:31:37 -0300
> >>> James Almer <jamrial at gmail.com> wrote:
> >>>     
> >>>> On 12/21/2017 7:22 PM, wm4 wrote:    
> >>>>> This removes the XP compatibility code, and switches entirely to SWR
> >>>>> locks, which are available starting at Windows Vista.
> >>>>>
> >>>>> This removes CRITICAL_SECTION use, which allows us to add
> >>>>> PTHREAD_MUTEX_INITIALIZER, which will be useful later.
> >>>>>
> >>>>> Windows XP is hereby not a supported build target anymore. It was
> >>>>> decided in a project vote that this is OK. (Technically, it could still
> >>>>> be built for Windows XP using an external pthread lib as of this
> >>>>> commit.)
> >>>>>
> >>>>> Windows Vista adds WSAPoll(), and for some reason struct pollfd. Since
> >>>>> we raise the Windows API level globally when enabling w32threads, we
> >>>>> need to move it before configure checks for struct pollfd to avoid that
> >>>>> the compatibility ifdef mess redefines it.
> >>>>> ---
> >>>>> Not sure if there's a better way to do the things configure does.
> >>>>> ---
> >>>>>  Changelog                  |   2 +
> >>>>>  compat/w32pthreads.h       | 269 ++-------------------------------------------
> >>>>>  configure                  |  93 ++++++++--------
> >>>>>  libavcodec/pthread_frame.c |   4 -
> >>>>>  libavcodec/pthread_slice.c |   4 -
> >>>>>  libavfilter/pthread.c      |   4 -
> >>>>>  libavutil/slicethread.c    |   4 -
> >>>>>  7 files changed, 60 insertions(+), 320 deletions(-)      
> >>>>
> >>>> [...]
> >>>>    
> >>>>> --- a/configure
> >>>>> +++ b/configure
> >>>>> @@ -2089,7 +2089,6 @@ TOOLCHAIN_FEATURES="
> >>>>>  "
> >>>>>  
> >>>>>  TYPES_LIST="
> >>>>> -    CONDITION_VARIABLE_Ptr
> >>>>>      kCMVideoCodecType_HEVC
> >>>>>      socklen_t
> >>>>>      struct_addrinfo
> >>>>> @@ -5163,8 +5162,8 @@ probe_libc(){
> >>>>>          add_${pfx}cppflags -U__STRICT_ANSI__ -D__USE_MINGW_ANSI_STDIO=1
> >>>>>          check_${pfx}cpp_condition _mingw.h "__MSVCRT_VERSION__ < 0x0700" &&
> >>>>>              add_${pfx}cppflags -D__MSVCRT_VERSION__=0x0700
> >>>>> -        check_${pfx}cpp_condition windows.h "defined(_WIN32_WINNT) && _WIN32_WINNT < 0x0502" &&
> >>>>> -            add_${pfx}cppflags -D_WIN32_WINNT=0x0502
> >>>>> +        check_${pfx}cpp_condition windows.h "defined(_WIN32_WINNT) && _WIN32_WINNT < 0x0600" &&
> >>>>> +            add_${pfx}cppflags -D_WIN32_WINNT=0x0600      
> >>>>
> >>>> This same check should now also be added to the mingw-w64 part of this
> >>>> function.
> >>>> Until now, this was done this way because mingw32 defaults to Win98,
> >>>> whereas mingw-w64 to WinXP. Now we need Vista in both.    
> >>>
> >>> Oh, I didn't even notice that was mingw32 only.
> >>>     
> >>>> It would be a good idea for that matter to make sure all the required
> >>>> Vista API is actually supported by Mingw32 3.15, or if we need to bump
> >>>> that (or just drop mingw32 support, which wouldn't really hurt).    
> >>>
> >>> The code already checks for CONDITION_VARIABLE. Any other things I
> >>> should check for? Maybe a representative function? (I don't think I'm
> >>> going to add checks for _all_ the symbols.)    
> >>
> >> Didn't realize that was checked, so no, should be good then.
> >>  
> >>>     
> >>>>>          eval test \$${pfx_no_}cc_type = "gcc" &&
> >>>>>              add_${pfx}cppflags -D__printf__=__gnu_printf__
> >>>>>      elif check_${pfx}cpp_condition crtversion.h "defined _VC_CRT_MAJOR_VERSION"; then
> >>>>> @@ -5184,14 +5183,14 @@ probe_libc(){
> >>>>>          # 0x601 by default unless something else is set by the user.
> >>>>>          # This can easily lead to us detecting functions only present
> >>>>>          # in such new versions and producing binaries requiring windows 7.0.
> >>>>> -        # Therefore explicitly set the default to XP unless the user has
> >>>>> +        # Therefore explicitly set the default to Vista unless the user has
> >>>>>          # set something else on the command line.
> >>>>>          # Don't do this if WINAPI_FAMILY is set and is set to a non-desktop
> >>>>>          # family. For these cases, configure is free to use any functions
> >>>>>          # found in the SDK headers by default. (Alternatively, we could force
> >>>>>          # _WIN32_WINNT to 0x0602 in that case.)
> >>>>>          check_${pfx}cpp_condition stdlib.h "defined(_WIN32_WINNT)" ||
> >>>>> -            { check_${pfx}cpp <<EOF && add_${pfx}cppflags -D_WIN32_WINNT=0x0502; }
> >>>>> +            { check_${pfx}cpp <<EOF && add_${pfx}cppflags -D_WIN32_WINNT=0x0600; }
> >>>>>  #ifdef WINAPI_FAMILY
> >>>>>  #include <winapifamily.h>
> >>>>>  #if !WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP)
> >>>>> @@ -5564,6 +5563,51 @@ check_code cc arm_neon.h "int16x8_t test = vdupq_n_s16(0)" && enable intrinsics_
> >>>>>  check_ldflags -Wl,--as-needed
> >>>>>  check_ldflags -Wl,-z,noexecstack
> >>>>>  
> >>>>> +if ! disabled w32threads && ! enabled pthreads; then
> >>>>> +    check_func_headers "windows.h process.h" _beginthreadex &&
> >>>>> +        check_type "windows.h" CONDITION_VARIABLE &&
> >>>>> +        enable w32threads || disable w32threads
> >>>>> +    if ! enabled w32threads && enabled winrt; then
> >>>>> +        check_func_headers "windows.h" CreateThread &&
> >>>>> +            enable w32threads || disable w32threads
> >>>>> +    fi
> >>>>> +fi
> >>>>> +
> >>>>> +if enabled w32threads; then
> >>>>> +    if check_cpp_condition windows.h "!defined(_WIN32_WINNT) || _WIN32_WINNT < 0x0600" ; then
> >>>>> +        add_cppflags -D_WIN32_WINNT=0x0600      
> >>>>
> >>>> Win Vista should be forced regardless of w32threads being enabled or
> >>>> not, so this should be done in probe_libc().    
> >>>
> >>> OK, sounds good. So this means we always set _WIN32_WINNT
> >>> unconditionally (unless it's already set to something higher than
> >>> Vista). This will probably simplify the configure changes.    
> >>
> >> Yes, that's the idea. After this we should also remove the lavf
> >> os_support compat code (things like poll emulation).  
> > 
> > I don't think Windows provides poll(). I don't know why the winsock
> > headers contain a pollfd struct. WSAPoll(), which supposedly works like
> > poll(), is not poll(), and also was rejected by the libcurl devs:
> > 
> > https://daniel.haxx.se/blog/2012/10/10/wsapoll-is-broken/  
> 
> I meant the pollfd struct, sorry. We'd not need to compat definition for
> it anymore.

I'm not really sure. The poll emulation is explicitly written to
support non-Windows OSes. Do we need to support some old UNIXes or
other obscure OSes?