Installation count discrepancy

[deathaxe]

The screenshot shows a package which was recently added to Package Control.

The allover install count in the header does not match the one of the graphics. The reason is a difference of the Windows installations.

Which value is valid?

[wbond]

If I recall correctly, the cumulative total counts unique platform/ip combos, whereas the graph shows raw numbers.

[deathaxe]

Means the 7 guys on linux and 30 on OSX installed the package once with luck but one third of the windows guys had to reinstall it one or more times? Seems a bit strange. I wouldn't expect that. It somehow looks like a bug in one of the graphics on a first glance.

[wbond]

Unfortunately I don't have time to dig into this now, but the whole site is open source, so if someone does have the time, here are the places to look:

SQL queries: https://github.com/wbond/packagecontrol.io/blob/master/app/models/package/find.py#L463-L509

Database structure: https://github.com/wbond/packagecontrol.io/blob/master/setup/sql/up.sql#L9-L47

Where installs are recorded: https://github.com/wbond/packagecontrol.io/blob/master/app/models/package/usage.py

[deathaxe]

Another strange example

A package added to repo 5 days ago with only about 8 installs a day gets 6k overall installs.

[wbond]

What is the name? If people had a folder with that name previously, those would be reported as package installs in the past.

[deathaxe]

https://packagecontrol.io/packages/Theme%20-%20Brackets