417192 - Add-on version confusion

Status: RESOLVED WORKSFORME

(addons.mozilla.org Graveyard :: Statistics, defect)

Description

[Dwayne Bailey]

My addons for dictionaries make use of date tags eg YYYYMMDD at https://preview.addons.mozilla.org/en-US/firefox/statistics/addon/5127 it looks like its picking these up correctly until you see versions like 20071033 which should never exist.  If you click on the dropdown to change the version for a specific colour plot you see many versions for a product that has only had at most 4 released versions.

Comment 1

[Justin Scott [:fligtar]]

I'm not sure what to tell you - people can and do change the default extensions.update.url string to be different which can cause abnormalities, but I don't know how something like this could happen unless those versions exist.

https://preview.addons.mozilla.org/en-US/firefox/statistics/csv/5127/version shows that plenty of people seem to be using those versions.

https://preview.addons.mozilla.org/en-US/firefox/statistics/json/5127/summary might be easier to read for the total counts of each version.

Are there any releases of this dictionary not hosted on AMO that are distributed elsewhere with different versions that don't have an updateURL and would still ping AMO?

Comment 2

[Dwayne Bailey]

We used to host versions at translate.org.za but now redirect to the AMO version.  But those had no updateURL defined.  

This still doesn't explain the completely bizaar version numbers as I can understand things being out in the field but not version numbers that are all over the place and which we've never defined for the dictionary.

I had considered that these might have been testing versions but then we would never have created a date like "20070669".  One thing I have wondered about is that the year and month seem never to be messed up, yet the day is.  Is there a limit to the value in the version?  Could this be some sort of overflow?

I'm also rather at a loss on this one.

Comment 3

[Dwayne Bailey]

I've looked through the data again and its just so consistently spread that it can't be someone hacking the URL or even old releases.   We have only really had massive uptake after hosting at AMO so we'd expect a valid release like 20071116 to rise to to the top, but it has the same average pings as everyone else.

Another idea.  Could the fact that the version is a valid date be confusing a SQL query somewhere?  Either when the data is captured or when it is read from the DB.

Comment 4

[Dwayne Bailey]

No change in this.  The reports still contain a predominance of faulty version numbers that aren't in any released version of the plugin

Comment 5

[Dwayne Bailey]

This is still happening in current data.  With no other versions now out there this looks seriously like an overflow error on the version number.

Comment 6

[Dwayne Bailey]

Whatever was changed on 2009-02-01 sorted the data out. We are now only getting pings on the two version of the spell checker that are really in the wild.

I'm marking as fixed.  I'll thank the magic SQL fairies :)