The old license only required providing source code if you redistributed it. The new license requires it even if you never distribute anything, e.g., you use it in the backend of a SaaS app.
That more or less just brings the license up to date with the original intent. You used to typically have to distribute apps if you wanted end-users to run them. Now you can half-distribute the app by only distributing the client half to browsers, and claim that that doesn't constitute redistribution. The AGPL, on the other hand, defines that kind of half-distribution as distribution.
> you can half-distribute the app by only distributing the client half to browsers
Calling the client portion of the code of an SaaS application "the client half" is a stretch. 1% is a better estimate, and "1%-distribution" is a better word than "half-distribution". The client portion of code has its source distributed at the same time as when it runs on the browser.
> That just brings the license up to date with the original intent
The only clue to intent users of software have is the specific wording of the license. Changing the license cripples the software. How do we know the original intent wasn't to bait users in with an wrongly used license, then change it later on?
Yes, that's what I said, it's the definition of copyleft.
For most companies the open source version was not acceptable anyway. E.g. most customers with sensitive information (such as government organizations) require on-site firewalled installations of your software anyway. So, you either pony up the relatively high per-CPU cost, or use a less restrictively licensed alternative such as leveldb.
This change will mostly make a difference for web applications. But my point was that Berkeley DB was already licensed quite restrictively before, contrary to some people's beliefs.