I checked this in r7005 (which supports roundrect) and if I add a Stack with In1.Cu, kiCad crashes.
Further tests show Inner pours, pulls back only from the Size provided.
It does not use the safer, more sensible, larger of Drill+Size.
Given kiCad seems to clone/copy a size onto inner layers, it really should allow specific define on inner layers too.
Allowing In*.Cu would seem sensible ?
Until that is fixed, I see more issues with translating designs from other CAD packages.
OK, I read through the blog post and I appreciate the development team for continuing to put out a LOT of effort since the 4.0.0 release.
Is there anything in the Release 4.0.3 that is NOT in the Nightly Build r7005 (and, presumably, subsequent nightly builds)? I have been running nightly builds since around the first of the year without any noticeable problems. This is on three different machines, though all are Win7/64-bit OS.
Keep up the great work!
Dale
Some things will be the same, some will be drastically different. Maybe some interface in 4.0.x has been changed and no longer in the nightly. There are absolutely features in the nightly that are not in 4.0.3, but I know that’s not what you asked.
No, anything in the minor releases (4.0.X) will be in the nightlies as well.
The minor releases are only for bug fixes! This should probably be made clearer in the announcement since so many people get confused by it. 
However, the test I did above, of adding a PAD Stack on an In1.Cu layer, simply crashed r7005, but the 4.0.3 release notes suggest that is not so brutal on 4.0.3 ?
I could not easily check 4.0.3 as by that stage, I had uninstalled it.
Does 4.0.3 still crash when loading a file with a Stack cloned onto In1.Cu ?