More than one forum user has somehow managed to update to 1.4 without the db update. Setting version_core to 1.3, they were able to trigger it.
One case that I am thinking is:
But I'm not sure how we could fix this...
and maybe there are other trigger cases?
The build 1.4 (builds.piwik.org) seems to be different as the latest.zip file.
Using the lastest.zip file I'm not able to update Piwik
const VERSION = '1.2.1'; in core/Version.php
Using the build it's fine...
Hope it helps
Lolo Irie: I just downloaded latest.zip, and Version.php says "1.4". It's possible you have a proxy in your network that's caching an older version of latest.zip. Within Piwik, we add a cache buster to the URL, so it'll always get the latest version, which helps should one happen to install an older version.
matt: I think this would be prevented by CoreUpdater_Controller's check for updateCorePlugins=1. This is short-circuited when run from the CLI ... maybe archive.sh triggers the update? (Fix: CLI would also require the updateCorePlugins=1 parameter.)
Oh I see... r3743 bypasses the updateCorePlugins=1 check. So yes, it's possible the Live plugin could have triggered the update before the new scripts in core/Updates/ were uploaded.
In that case, other triggers would include API calls and widgets. I suppose even an anonymous user navigating to the Login form would also trigger the update?
Also, this implies we have a race condition during the one-click update (while files are being copied).
Maybe something like this? (Pseudo-code:)
if (there are updates)
if (user is superuser)
if ((module == 'CoreHome' || module == 'MultiSites')
and (action is empty || action == 'index'))
redirect to CoreUpdater
// else pass through the request
else if (module != 'Login' && module != 'Proxy')
Piwik_ExitWithMessage('You must login as superuser to complete the update.');
r3743 indeed looks pretty bad now that you highlight it. Maybe we just revert and always show the "Update page" which should really improve the situation?
I agree that asking for super user login might be a good idea, but at the same time it also adds complexity where maybe reverting part of r3743 would fix most issues?
(In ) fixes #2403
(In ) refs #2403 - comment:3
(In ) fixes #2403 - typo