40146[BUGFIX] Do not write useless registry entries for extensions
Commit message
Review
-
Code review CR 0
-
Verified V 0
Files 1
Review log 6
-
Markus Klein uploaded patch set 1 06/10/15, 09:24
-
Markus Klein updated the commit message in patch set 2 06/10/15, 09:27
-
Xavier Perseguers commented on patch set 2 CR -1 06/10/15, 09:31
no ticket number -
Christian Kuhn commented on patch set 2 06/10/15, 19:43
i'm a bit biased here, since this changes the current hidden magic behavior in another way: if an extension is installed that has no import file, it is currently marked as "import is done". So, if later the extension is upgraded and it then comes with import files, those will *not* be installed. If we change this now, it may happen that if an extension is upgraded, then unloaded and loaded again, it would suddenly import stuff. Depending on how intrusive this import is, it would be an unexpected thing that may even commit hazard ... So, I'm a bit unsure here and maybe we need some gui / hinting in this area? -
Markus Klein commented on patch set 2 06/10/15, 21:01
You are right of course Christian, but on the other hand, how likely is it that an extension "all of a sudden" ships some imports which would cause massive trouble? I guess that would be a real edge-case. Basically this change was mostly intended to show some - to my opinion - misconception in Core in that area as I stumbled over it by reproducing some bug report. As I said on Slack, I opt for actually dropping the adt.sql feature completely as it has close to no value for extensions. It would be much cleverer to write a custom ext_update script (like static_info_tables) or leverage some EM signal to deal with data changes necessary for an extension. IMHO the Core has far to few information on what those data mean or are intended for, so the true decision on when data needs to be imported/migrated/deleted can only be taken by the extension itself. Therefore the extension should consequently provide this functionality on its own. The table definitions of course would still belong to the ext_tables.sql file, in order to allow the Core to clean up after the ext has been removed. -
Markus Klein abandoned the change 06/28/15, 12:10
Ignore this for the moment. ;-)