Talk:Pootle
Preliminary
The Goal is to manage SME translations with Pootle hosted on contribs.org.
Once we have everything worked out how to interact with pootle and the formmagick stuff then I'll get something up on contribs.org that everyone can use. It would be really nice if we could automate the extraction/import of files that need to be translated into pootle but first things first.
Slords 18:23, 24 January 2008 (MST)
Pootle Usage
Test pootle site: http://www.unixlan.com.ar:8888
untranslated words
click "Show Editing Functions" and finally at "Quick Translate" for each file or whole language.
Also you can see the suggestions clicking at "Review Suggestions".
checks
click "show checks" to see a list of syntax errors
acronyms 6 strings (2%) failed brackets 17 strings (6%) failed doublequoting 4 strings (1%) failed doublespacing 1 string (0%) failed endpunc 11 strings (3%) failed endwhitespace 4 strings (1%) failed numbers 2 strings (0%) failed puncspacing 2 strings (0%) failed sentencecount 2 strings (0%) failed simplecaps 16 strings (5%) failed startcaps 6 strings (2%) failed unchanged 7 strings (2%) failed untranslated 53 strings (19%) failed
click on one of the links offered and fix them
See http://translate.sourceforge.net/wiki/guide/pofilter_examples and http://translate.sourceforge.net/wiki/guide/translation/commonerrors
Merging new strings
Add, remove or modify strings in the template/*.pot file then click 'update from template'.
- New strings are added
- modified strings use existing data and made fuzzy (i think)
- deleted strings are moved to the bottom of the file and commented out with #~
Formagick
Normando suggested a few tools, I'm using XML2PO, see the others in the history http://wiki.contribs.org/index.php?title=Talk:Pootle&oldid=7649#I_need_your_help
XML2PO
( http://linux.die.net/man/1/xml2po )
I have packaged for a better installation. You can download from
http://mirror.contribs.org/smeserver/contribs/nhall/sme7/contribs/pootle/rpm/gnome-doc-utils-0.12.0-1.noarch.rpm
Before try, you must edit a few lines.
/usr/share/xml2po/empty.py
Line 27 from "return ['base', 'A', 'a', 'i', 'I', 'B', 'b' ,'P' ,'p' ,'h2' ,'H2', 'div', 'DIV', 'font'] Line 31 from "return []" to "return ['trans']" Line 35 leave "return []" Line 39 from "return []" to "return ['trans']" // or ['lang'] testing
Create .po and export xml
To test the lexicons
xml2po -m empty -e -o backup.po backup
View the new bakup.po file in the new PO format. Excellent. Now you can translate PO with pootle, and return again to formmagick panel with this command:
xml2po -p backup.po backup > backup.new
As you can see, if you not translate backup.po, new_backup file is equal to original backup file, BUT with one difference, backup.new file has added a line at the header: We remove this line with newxml
<?xml version="1.0" encoding="utf-8"?>
shell scripts
xml2po ignores tags such as CDATA, CDATA does not appear at the PO file.
The workaround is to find and replace the problem code
Create a clean .po file with newpo, either edit your lexicon or add s///g commands to workaround new problems
After you have a clean english .po you could use that as a template and copy and paste the translation in. A better idea is to write a merge tool see below
newpo
#!/bin/bash
#
#SME Server Create lexocon .po
if [ -f $1 ]
then
#convert to UTF-8 while working on pootle
mv $1 $1.bak
/usr/bin/iconv -f ISO-8859-1 -t UTF-8 $1.bak > $1
#echo "Remove CDATA and reformat problem codes in $1"
perl -pi -e 's/<!\[CDATA\[/STARTCDATA/g' $1
perl -pi -e 's/\]\]>/ENDCDATA/g' $1
perl -pi -e 's/\&/AMP/g' $1
perl -pi -e 's/P\>/p\>/g' $1
perl -pi -e 's/A\>/a\>/g' $1
perl -pi -e 's/\<(br|BR)\>/breeak/g' $1
perl -pi -e 's/\<\/(font|FONT)\>//g' $1
#echo "Create $1.po"
xml2po -m empty -e -o $1.po $1
#echo "Replacing CDATA in $1"
perl -pi -e 's/STARTCDATA/<!\[CDATA\[/g' $1.po
perl -pi -e 's/ENDCDATA/\]\]>/g' $1.po
perl -pi -e 's/AMP/\&/g' $1.po
perl -pi -e 's/breeak/\<bXr\>/g' $1.po #wiki display error, fixme
#basic testing
A=`cat $1.bak |grep '<entry>' |wc -l`
B=`cat $1.po |grep msgid |wc -l`
C=`expr $B - 1`
echo "entries $A, msgid $C"
if [ $A -ne $C ]
then
echo "Errors in formatting"
tail $1.po
fi
#restore original
mv $1.bak $1
if [ ${#2} -gt 0 ]
then
echo "<base> entries"
cat $1 |grep base |sort
fi
else
#print usage informamtion
echo "Usage: $0 LexiconFilename"
echo "Usage: $0 LexiconFilename check"
fi
- I am wondering why we are using perl to replace as we can also use sed and if readability is not an issue we can do the whole CDATA tag conversin at once
sed -e 's/<!\[CDATA\[/STARTCDATA/g' -e 's/\]\]>/ENDCDATA/g' -e 's/\&>/AMP/g' < original.file > conversion.file
- And back
sed -e 's/STARTCDATA/<!\[CDATA\[/g' -e 's/ENDCDATA/\]\]>/g' -e 's/AMP/\&/g' < conversion.file > new.file
- yes but there are more conversions, so it was easier to operate on the one file, efficiency isn't a concern
newpomerge
The aim of this is to take a blank en .po and merge in the values for a translation .po
en.po has
msgid "Yes" msgstr ""
fr.po has
msgid "Oui" msgstr ""
we want fr.po to look like
msgid "Yes" msgstr "Oui"
We can't assume the translation will be in the same order, so it involves looking back at the lexicon base field to do the match
or rewriting the fr lexicon first so it is in the same order, with empty fields if necessary, then you can just go through in order picking out records.
Sample .po output attached, it will always be in this format (but translation msgid's may be missing or out of order)
msgid "" msgstr "" "Project-Id-Version: PACKAGE VERSION\n" "POT-Creation-Date: 2008-02-04 02:31+1100\n" "PO-Revision-Date: YEAR-MO-DA HO:MI+ZONE\n" "Last-Translator: FULL NAME <EMAIL@ADDRESS>\n" "Language-Team: LANGUAGE <LL@li.org>\n" "MIME-Version: 1.0\n" "Content-Type: text/plain; charset=UTF-8\n" "Content-Transfer-Encoding: 8bit\n" #: functions/useraccounts:6(trans) msgid "Create, modify, or remove user accounts" msgstr "" #: functions/useraccounts:33(trans) msgid "Create or modify" msgstr "" #: functions/useraccounts:66(trans) msgid "Modify the admin account" msgstr ""
newxml
#!/bin/bash
#
function usage {
echo ""
echo "Create .xml from .po"
echo ""
echo "Not enough parameters provided."
echo "Usage: $0 filename (dont add .po)"
echo ""
echo "Optional: compare against original"
echo "Usage: $0 filename check"
echo ""
}
#check for required parameters
if [ ${#1} -gt 0 ]
then
#echo "Remove CDATA in $1.po"
cp $1.po $1.bak
perl -pi -e 's/<!\[CDATA\[/STARTCDATA/g' $1.bak
perl -pi -e 's/\]\]>/ENDCDATA/g' $1.bak
perl -pi -e 's/\&/AMP/g' $1.bak
#echo "Create xml"
xml2po -p $1.bak $1 > $1.xml
#echo "Replacing CDATA"
perl -pi -e 's/STARTCDATA/<!\[CDATA\[/g' $1.xml
perl -pi -e 's/ENDCDATA/\]\]>/g' $1.xml
perl -pi -e 's/AMP/\&/g' $1.xml
#this is added at line 367 xml2po, it needs to be removed or better not added
#perl -pi -e 's/\<\?xml version="1.0" encoding="utf-8"\?\>//' $1.xml
#remove first line
perl -i.old -ne 'print unless 1 .. 1' $1.xml
rm $1.bak $1.xml.old
if [ ${#2} -gt 0 ]
then
#echo "#diff -n $1 $1.xml"
diff -n $1 $1.xml
fi
else
#print usage informamtion
usage
fi
yum
- newxml needs a parameter to set: lang="fr"
- why doesn't this work ?
- xml2po -l fr -p yum.po yum > yum.xml
- -l --language=LANG Set language of the translation to LANG
- you can't have a base equal to a trans, needs a new bug (if pootle goes ahead)
- most en base lexicons do this,
- easy fix is to tweak the trans for Form_Title, capitalise or punctuate
- -<trans>Software installer</trans>
- +<trans>Software Installer</trans>
<lexicon lang="en-us">
<entry>
<base>FORM_TITLE</base>
<trans>Software installer</trans>
</entry>
<entry>
<base>Configuration</base>
<trans>Configuration</trans>
</entry>
<entry>
<base>Software installer</base>
<trans>Software installer</trans>
</entry>
<lexicon lang="en-us">
<entry>
<base>FORM_TITLE</base>
<trans>Mise à jour logicielle</trans>
</entry>
<entry>
<base>Configuration</base>
<trans>Configuration</trans>
</entry>
<entry>
<base>Mise à jour logicielle</base>
<trans>Mise à jour logicielle</trans>
</entry>
Console
update all .po files
- Templates have been implemented with up to date strings,
- now it's over to users to translate them
.po file names are inconsistent, SV & FR add .tmpl.po
- .tmpl.po is correct, other languages have been updated
Use UTF-8 for console .po files
- Pootle has problems with the current charset=iso-8859-1
- we now use UTF-8
revision control using CVS/SVN
Today I had a quick look at the pootle pages and found a wiki as well, which has some valuable information like for instance revision control using CVS/SVN: http://translate.sourceforge.net/wiki/pootle/version_control - Cactus 03:26, 25 January 2008 (MST)
progress
we'll look at this later, perhaps starting with contribs
- A few links for future implementation:
- http://subversion.tigris.org/tools_contrib.html#po_update_sh
- http://subversion.tigris.org/tools_contrib.html#verify_po_py
- http://subversion.tigris.org/tools_contrib.html#svnmerge_py
- http://subversion.tigris.org/tools_contrib.html#svnmerge_sh
Remarks
Translation Workflow
Draft suggestions ...
The original files are in cvs.
.po files
- gettext strings are in various files in cvs, currently we think we have them all in pootle as .pot template files
- we need a script to search a maintained list of files to extract current gettext strings
- this may find missing (or modified) strings or remove old ones saving on pointless translations
- existing translations have been placed in pootle and merged with the templates
- this shows missing translations very nicely and is ready for user testing
- all .po files have been converted to UTF-8, they can be converted back easily if necessary
- changes to original files, ie changed gettext strings, have to be tracked as usual in the bug tracker
- these are imported into the templates, other languages are updated with a click
- at some point we diff against cvs po files, the patch is checked and applied.
FormMagick files
- these are XML files with a <base> and <trans> pair of strings, these can be converted to .po files with some work
- some rough scripts are being worked on above, there may be others on the net waiting to be found...
- we need to clean up some inconsistencies in the english lexicons and they will be our .pot templates
- we need to script the conversion of existing translations into .po files, discussed above
- all .po files have been converted to UTF-8, they can be converted back easily if necessary
- changes to original files, have to be tracked as usual in the bug tracker
- the following is speculative ...
- these are imported into the .pot templates, other languages are updated with a click
- I think any new <base> strings have to be added to the cvs xml file for each language
- at some point we run xml2po to apply the new translation back to the xml file,
- then diff against cvs, make patch, check and apply
Pootle issues
UTF-8
Bug created, 'Use UTF-8 for console .po files' http://bugs.contribs.org/show_bug.cgi?id=3858
- SME Translation files currently use iso-8859-1
- we currently convert any file in pootle to UTF-8
- we can convert back when returning to SME if necessary, by adding one line in a script.
Testing Pootle I have found a few issues. I will try to describe.
Pootle (and Translation Toolkit used by Pootle) use as default a charset UTF-8. Everytime I have create a new language translation, Pootle parse from templates encoding as UTF-8. Everytime I have "Update from templates" the target PO files are updated as UTF-8, and if target has set charset=iso-8859-1, it is then corrupted. I can't found a method to set Pootle to use ISO-8859-1 as default. Because this, and to avoid these issues I have decide to use UTF-8 as charset in Pootle. Please see http://translate.sourceforge.net/wiki/guide/locales/glibc?s=charset#editing
Not only Pootle. Every desktop translation tool set UTF-8 as default charset.
If you want to merged or override a PO file, convert to UTF-8 (if not yet). Then you can upload without corrupt the target file.
One note for admins when Pootle run at contribs.org: I suggest to use a tool for convert from UTF-8 to ISO-8859-1 before make a new language rpm update package (if gettext can't support utf-8 encoding).
I have use this great script to convert between any charset files in a directory:
#!/bin/bash
#./dir_iconv.sh dir cp1251 utf8 - converts all files from directory dir .. cp1251 (windows-1251) to utf8.
ICONVBIN='/usr/local/bin/iconv' # path to iconv binary
if [ $# -lt 3 ]
then
echo "$0 dir from_charset to_charset"
exit
fi
for f in $1/*
do
if test -f $f
then
echo -e "\nConverting $f"
/bin/mv $f $f.old
$ICONVBIN -f $2 -t $3 $f.old > $f
else
echo -e "\nSkipping $f - not a regular file";
fi
done
This apply also to panels (if formmagick can't support UTF-8 encoding)
There is an open posibility to modify Pootle or translation toolkit to make this automatically, but I don't know how to do. --Normando Hall 12:45, 2 February 2008 (MST)
- . Formmagick supports UTF-8: http://search.cpan.org/~mitel/CGI-FormMagick-0.89/lib/CGI/FormMagick.pm
- . GetText does support UTF-8 as well: http://www.gnu.org/software/gettext/manual/gettext.html#Charset-conversion
- It is quite common for translations/lexicons to be in UTF8 as this is a well known locale independent character set, probably also the reason why Pootle is sticking to UTF-8. - Cactus 15:01, 2 February 2008 (MST)
