It seems to be still correct, and it is no longer complaining about syntax
errors when seeing commented-out HTML (esp. with TMPL_* directives).
Don't try to translate stuff between <title>...</title> too, the stuff in
the middle is supposed to be PCDATA.
(Actually, xgettext.pl did not know how to handle them in the files listed
in the list of files.)
If the po file is empty (corrupted), $href->{'""'} will be undefined.
We just blindly dereferenced this null value without checking.
Minor comment update
Removed forced backups and the comment about interrupting xgettext.pl
corrupting the po file, now that we seem to be detecting the situation.
when minor whitespace changes occur in the original templates; it also
makes the strings much easier to read (e.g., instead of "foo\n\n\t\t bar",
xgettext.pl will now always generate "foo bar" and tmpl_process3.pl will
understand it to be the same as the original string).
Early termination of analysis if we encounter some strings, such as </h1>
or | or ||, in order to avoid extracting strings that are unnecessarily
long and which doesn't add any meaningful context.
to be combined together into one string. This seems to have the desired
effect (that "<b>foo</b> bar" type strings are now recognized in one piece).
However, "<h1>foo</h1>\nexplanation"-type things may now also be (arguably
wrongly) recognized as one piece.
either iso8859-1 or utf8, msgmerge(1) won't crap out. The code is ugly;
the conversion table is hard-coded, and in some place not very appropriate.
However, this does fix the case where a few strings containing French
characters can't be translated. As a side effect, tmpl_process3 can now
also be used for French or other languages using iso8859-1.
gives better strings. (Always allowing combinations gives havoc, we
currently avoid this by allowing combination only if the first and last
tokens are both TEXT.)
word order is too different than the word order of the target language to
yield meaningful translations.
The new scripts use a different translation file format (namely standard
gettext-style PO files).
This seems to reasonably work (e.g., producing an empty en_GB translation
then installing seems to not corrupt the "translated" files), but it likely
will still contain some bugs. There is also little documentation, but try
to run perldoc on the .p[lm] files to see what's there. There are also some
spurious warnings (both from bugs in the new scripts and from buggy third-
party Locale::PO module).
