Re: [lmi] wx ./configure options

[Greg Chicares]

On 2017-05-27 11:38, Vadim Zeitlin wrote:
> On Sat, 27 May 2017 10:44:32 +0000 Greg Chicares <address@hidden> wrote:
[...]
> GC> [BTW, is there any way I can easily see which options were ignored?
> GC> On my first attempt, I mistakenly used this:
> GC>   --enable-enable-no_deps
> GC> and I didn't find an error message diagnosing my mistake.]
> 
>  For some reasonable but still annoying reasons, configure doesn't check
> --enable-xxx and --with-xxx options validity by default. This, however,
> can be explicitly enabled with --enable-option-checking=fatal and I think
> it would be a good idea to do it by default (you would still be able to use
> --disable-option-checking if really needed, e.g. if you wanted to pass some
> options not recognized by the main configure script to tiff or expat
> configure which is run by it if the corresponding system library couldn't
> be found).

I'll omit the "=fatal" because using it has this result:

=== configuring in src/expat 
(/opt/lmi/wx-scratch/wxWidgets-3.1.0-p1/gcc-491-89775012799fe5c9bd8ea61e5fa43da7/src/expat)
configure: running /bin/bash ../../../src/expat/configure 
--disable-option-checking '--prefix=/opt/lmi/local'  
'--exec-prefix=/opt/lmi/local' '--build=x86_64-unknown-linux-gnu' 
'--host=i686-w64-mingw32' '--disable-apple_ieee'
[...big snip...]
 '--without-subdirs' 'CFLAGS=-fno-omit-frame-pointer' 
'CXXFLAGS=-fno-omit-frame-pointer -std=c++11' 
'build_alias=x86_64-unknown-linux-gnu' 'host_alias=i686-w64-mingw32' 
--cache-file=/dev/null --srcdir=../../../src/expat
configure: error: unrecognized options: --disable-apple_ieee, --disable-aui, 
--disable-compat30, --disable-dependency-tracking, --disable-fswatcher, 
--disable-gif, --disable-mediactrl, --disable-precomp-headers, 
--disable-propgrid, --disable-ribbon, --disable-richtext, --disable-stc, 
--disable-webview, --enable-monolithic, --enable-std_iostreams, --enable-stl, 
--enable-vendor, --with-cxx, --without-libjpeg, --without-libtiff, 
--without-opengl, --without-subdirs
configure: error: ../../../src/expat/configure failed for src/expat
/opt/lmi/src/lmi/install_wx.make:202: recipe for target 'wx' failed

which prevents wx from building. That means I'll have to grep the
output for /WARNING/ occasionally, but that's okay.

Aside from that, I've made all the changes you suggested, and they
have a remarkable effect on the time it takes to build wx:

* i686 before changes

time make $coefficiency --output-sync=recurse -f install_wx.make  >../log 2>&1  
                      
make $coefficiency --output-sync=recurse -f install_wx.make > ../log 2>&1  
2588.31s user 158.13s system 1416% cpu 3:13.86 total

* i686 after changes

time make $coefficiency --output-sync=recurse -f install_wx.make  >../log 2>&1
make $coefficiency --output-sync=recurse -f install_wx.make > ../log 2>&1  
2911.87s user 146.39s system 2145% cpu 2:22.53 total

* x86_64 before changes

/opt/lmi/src/lmi[0]$time make $coefficiency 
exec_prefix=/opt/lmi/x86_64-w64-mingw32/local host_type=x86_64-w64-mingw32 
--output-sync=recurse -f install_wx.make >../log 2>&1
make $coefficiency exec_prefix=/opt/lmi/x86_64-w64-mingw32/local   -f  >  2>&  
2636.68s user 163.24s system 924% cpu 5:02.79 total

* x86_64 after changes

/opt/lmi/src/lmi[0]$time make $coefficiency 
exec_prefix=/opt/lmi/x86_64-w64-mingw32/local host_type=x86_64-w64-mingw32 
--output-sync=recurse -f install_wx.make
>../log 2>&1
make $coefficiency exec_prefix=/opt/lmi/x86_64-w64-mingw32/local   -f  >  2>&  
2899.20s user 147.71s system 1807% cpu 2:48.55 total

Summarizing the "total" values, which are clock times:

           before    after    ------time ratio------  --speedup--
  i686    3:13.86  2:22.53    142.53 / 193.86 = .735     36%
  x86_64  5:02.79  2:48.55    168.55 / 302.79 = .557     80%

I don't quite know how to interpret that in detail. If PCH and .deps
are arbitrarily assumed to take about the same amount of time for
either architecture, then I'd expect a greater speedup in the i686
build, because it takes less total time. Maybe one of those newly
suppressed options is just more inefficient for x86_64. But at any
rate the speedup is most welcome.