timbru
commented
10 months ago After discussion - we will close this. It's only useful to do this under very specific circumstances: i.e. there is a 0.14.0/.1/.2 installation that runs into the issue that a tempfile is not used, and a resulting file is half-written due to a hard server crash or disk full situation, and then the user upgrades to this. Otoh it could cause issues in other situations. So, it was decided that it's best not to try any automated recover here. Just let users update to 0.14.3, and talk to use if they get hit by the issue in 0.14.0-0.14.2
This should fix issues. It still needs more manual testing. Unfortunately, it's not easy to test this in automated way currently as our upgrade test code assumes that values can be read before putting them into a memstore (which only accepts valid json) for testing. This needs some thought - or a very thorough review and manual testing.
There is a happy case test, but that one essentially does nothing - it covers the case where an upgrade is done and there were no corrupt files.