Andrew Hundt

> @ahundt In general do you think the sort of things you're debugging would this require an Xcode project for each package or a unified Xcode workspace (.xcworkspace) that combines...

It looks like cocoapods does exactly this, so the best reference is probably the code that is run upon the `cocoapods install` command. Here is the [most relevant looking workspace...

Looks like there is a project to do just this called [cocoapods/xcodeproj](https://github.com/CocoaPods/Xcodeproj). pro: already implemented and can be used as a reference. con: it is in ruby

@wjwwood that looks like it would be an even better option. @jbohren presumably that should be integrated, correct? Perhaps I'm misunderstanding what is meant by subprocess call?

This really needs to be documented and parameterized. It is very hard to make sense of this, and a bunch of things seem to be hard coded. Can we also...

@j-varun Hey we are in the middle of porting things to pytorch, can the keras specific parts be in a separate file? GitHub.com/ahundt/costar_dataset @RexxarCHL we will need to bring these...

Hey any updates on the classifier component?

80 is annoyingly narrow, even something simple like the following goes too long in two places. ```python # example URL for wen downloading files is already way over the limit...

halfway at 120 perhaps? :-)

Also I try not to use 160 but since the linter makes it impossible to go over when there is a fair reason like a URL, so that's worth considering....