Stop. I just understand the logic. it is not the "title" field but "qid" of subquestion.
So, in v7, how does the answers restore works?

I've just done a test. if I delete a subquestion ad I recreate with same code (title), it doesn't restore data from oldresponses$sid_$timestamp. In my opinion it's a bit problematic.

I've just done a test. if I delete a subquestion ad I recreate with same code (title), it doesn't restore data from oldresponses$sid_$timestamp. In my opinion it's a bit problematic.

If you do same in 6.X and it work: yes, it's problematic. Must be reported separately.

(I think we try to use Expression Code for column names in 7.X or 8.X, I don't really understand the interest here.)

in v6 it works, because code is part of the field name....
in screenshot you can see fields for subquestion of question 2453. If I delete and recreate a subquestion with code "13" I obtain the same field name and I can restore field values.

in v6 it works, because code is part of the field name....

Then it's clearly an issue : no question here !

The premise is correct, the new fieldname looks like Q<qid>_S<sqid>, where qid is the id of the main question and sqid is the id of the subquestion. There are many other forms, depending on the type, however, there is a misunderstanding that we need to clarify first.

Whilst the title of the subquestion or the code of the answer was an identifier in LS6, this is no longer the case. So the 2454 in your Q2453_S2454 is no 2453 + 1, but a qid of 2454, which happened to be created as the next question after the one with the qid of 2453 was created and since qid is incremental, its value happened to be 2454, but there is no causal relation between it being the code of 1 and the main question being 2453 and the result of 2454. If you remove a subquestion and recreate a new one with the same code, it will have a different qid. Your 2436 in Q2435_S2436 happened to be the next question to be created after 2435, however, your 2443 in Q2435_S2443 was created later, there were a few questions created in-between.

Since the title/code is no longer identifying a fieldname and since we use the actual identifier to identify it, if you had, say Q2435_S2436 and it had a code of 1, you deactivated the survey, removed this subquestion and created a new one with a code of 1, then that new subquestion will have a different qid from 2436 and hence upon reactivation its values will not hydrate your newly created field.

Indeed, that worked in LS6, but it's no longer the case in LS7 because the fieldnames are no longer reliant on code/title and therefore this is why your values were not brought back. We could enhance the reactivation screen with a mapping between what seem to be valid connections between old and new fields which have the same code and main question id so they could be brought back, but we'll need to discuss this idea with decision-makers before we can promise anything.

that worked in LS6, but it's no longer the case in LS7

Then it's not an improvement! It's a new issue.

I always use VV export/import because it's better. But clearly, it's a new issue here.

Thank you all for the answers.
In this situation, for being sure data are recovered, I need to export/import in VV, which caption in row 2 is qcode_scode.
I agree the new field names schema, that wipes away useless group id and finally starts with char instead of digit, but it creates this problem in rehydrating table.

Fix committed to master branch: http://bugs.limesurvey.org/plugin.php?page=Source/view&id=41219