|
|
@ -862,31 +862,16 @@ chmod -x $RPM_BUILD_ROOT/%{basisinstdir}/program/hid.lst
|
|
|
|
#remove spurious exec bits
|
|
|
|
#remove spurious exec bits
|
|
|
|
chmod -x $RPM_BUILD_ROOT/%{basisinstdir}/program/gengalrc
|
|
|
|
chmod -x $RPM_BUILD_ROOT/%{basisinstdir}/program/gengalrc
|
|
|
|
|
|
|
|
|
|
|
|
#We don't need to carry around all the letter templates for all the languages
|
|
|
|
#ensure a template dir for each lang
|
|
|
|
#in each langpack! In addition, all the bitmaps are the same!
|
|
|
|
|
|
|
|
pushd $RPM_BUILD_ROOT/%{basisinstdir}/share/template
|
|
|
|
pushd $RPM_BUILD_ROOT/%{basisinstdir}/share/template
|
|
|
|
mkdir -p wizard
|
|
|
|
|
|
|
|
for I in %{langpack_langs}; do
|
|
|
|
for I in %{langpack_langs}; do
|
|
|
|
if [ -d $I/wizard/bitmap ]; then
|
|
|
|
mkdir -p I
|
|
|
|
cp -afl $I/wizard/bitmap wizard/
|
|
|
|
|
|
|
|
rm -rf $I/wizard/bitmap
|
|
|
|
|
|
|
|
ln -sf ../../wizard/bitmap $I/wizard/bitmap
|
|
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
if [ -d $I/wizard/letter/$I ]; then
|
|
|
|
|
|
|
|
mv -f $I/wizard/letter/$I ${I}_wizard_letter_${I}
|
|
|
|
|
|
|
|
rm -rf $I/wizard/letter/*
|
|
|
|
|
|
|
|
mv -f ${I}_wizard_letter_${I} $I/wizard/letter/$I
|
|
|
|
|
|
|
|
else
|
|
|
|
|
|
|
|
rm -rf $I/wizard/letter/*
|
|
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
done
|
|
|
|
done
|
|
|
|
popd
|
|
|
|
popd
|
|
|
|
|
|
|
|
|
|
|
|
#Set some aliases to canonical autocorrect language files for locales with matching languages
|
|
|
|
#Set some aliases to canonical autocorrect language files for locales with matching languages
|
|
|
|
pushd $RPM_BUILD_ROOT/%{basisinstdir}/share/autocorr
|
|
|
|
pushd $RPM_BUILD_ROOT/%{basisinstdir}/share/autocorr
|
|
|
|
|
|
|
|
|
|
|
|
#en-ZA exists and has a good autocorrect file with two or three extras that make sense for South Africa
|
|
|
|
|
|
|
|
en_GB_aliases="en-AG en-AU en-BS en-BW en-BZ en-CA en-DK en-GH en-HK en-IE en-IN en-JM en-NG en-NZ en-SG en-TT"
|
|
|
|
en_GB_aliases="en-AG en-AU en-BS en-BW en-BZ en-CA en-DK en-GH en-HK en-IE en-IN en-JM en-NG en-NZ en-SG en-TT"
|
|
|
|
for lang in $en_GB_aliases; do
|
|
|
|
for lang in $en_GB_aliases; do
|
|
|
|
ln -sf acor_en-GB.dat acor_$lang.dat
|
|
|
|
ln -sf acor_en-GB.dat acor_$lang.dat
|
|
|
@ -895,6 +880,8 @@ en_US_aliases="en-PH"
|
|
|
|
for lang in $en_US_aliases; do
|
|
|
|
for lang in $en_US_aliases; do
|
|
|
|
ln -sf acor_en-US.dat acor_$lang.dat
|
|
|
|
ln -sf acor_en-US.dat acor_$lang.dat
|
|
|
|
done
|
|
|
|
done
|
|
|
|
|
|
|
|
#en-ZA exists and has a good autocorrect file with two or three extras that make sense for
|
|
|
|
|
|
|
|
#neighbouring english speaking territories
|
|
|
|
en_ZA_aliases="en-NA en-ZW"
|
|
|
|
en_ZA_aliases="en-NA en-ZW"
|
|
|
|
for lang in $en_ZA_aliases; do
|
|
|
|
for lang in $en_ZA_aliases; do
|
|
|
|
ln -sf acor_en-ZA.dat acor_$lang.dat
|
|
|
|
ln -sf acor_en-ZA.dat acor_$lang.dat
|
|
|
|