#2 When updating file in Zanata, translation are removed

Probably related: https://zanata.atlassian.net/browse/ZNTA-1134

But the ticket is old and still not solved, and options exists to deals with copy/compare translations 🤔

Also it talks about the key system, perhaps the actual compilation does not use them properly. Specifically I need to check the different commentary format for gettext, I should have missed something to allow Zanata recognize identical translations

I have created a debug version in Zanata to made some tests on it.

https://translate.zanata.org/iteration/view/thea2/debug-version?dswid=-8148

The pool of options to check existing translation and decide to copy or not copy them when uploading files seems a good start :-)

Okay in fact Zanata keep the translated info in its translation memory, BUT does not copy directly if the english text has too massively changed.

In the attached screenshot, English change from The target has to be an empty card slot in either ally melee or ally ranged row. to Targets an empty card slot in either ally melee or ally ranged row. so Zanata chooses not to copy the translation, as it matches "only" 88%...

The good new is the translation is not lost, we could copy it from the memory by using the "Copy" button.

But it's still manual action, clearly not perfect...

There is a tool in Zanata which do exactly what we need, the merge translation from translation memory

Basically it search all files for string matching translation memory and copy it as fuzzy for review.

From test, it found a lot of untranslated strings, and even fill files we have still not translated :-)

Copied as Translated
100% Match
28,441 word(s) — 152,887 character(s) — 1,483 message(s)

Copied as Fuzzy
100% Match
26,735 word(s) — 142,783 character(s) — 1,800 message(s)

90% to 99% Match
4,040 word(s) — 21,915 character(s) — 154 message(s)

80% to 89% Match
2,638 word(s) — 14,396 character(s) — 140 message(s)

0% to 79% Match
1,858 word(s) — 10,012 character(s) — 101 message(s)

There are somes pains however:

• It's a manual option, so I need to launch it each time I update documents
• It takes ages! Around 2h30 / 3h with all the text we have...
• It takes at maximum 75% similarity. Bad luck, lots of DES translation are 71% similarity, and thus not copied...

I am looking forward for how we can increase this process!

Okay, new test for the 370 version:

• Create a new translation from the last one (v0.1-ea here) (get translated AND reviewed text)
• Upload new files (lost everything that is not 100% similar)
• Run a Merge Tranlation with:
  • locale fr
  • 75% similarity
  • Use precedent versions with priority to the last one
  • Copy everything as translated (from other doc, from other context)

With this, we get all reviewed text, translated text still similar, and the majority of texts that has changed.

Everything in fuzzy needs a review.

Tested with the version 382, it does the job.

Can be closed, details are in other tickets.