If a string is not extractable as utf8, tmxmaker fails and just stops for the locale. That means we generate incomplete cache_xx.php arrays and incomplete xml files for tmx.
I just noticed that with Occitan on the release channel, one of the mail files is not encoded correctly and only 8000 strings are extracted instead of 18000 (and they are not usable because of the broken generated data) .
We need to ignore these errors and just skip the wrongly encoded string to at least not have Transvision breaking on this content.
When using silme for string extraction, don't stop the script on errors as it prevents closing properly the php array and the tmx file, instead just skip the string and keep on indexing the repo.
Note that it also breaks the 'onestring' view on the release channel since it includes broken occitan data:
merged, also manually generated occitan files on the server from the transvision-beta repo as a temporary solution to the onestring view being broken.