I think you misunderstand copyright... you can't violate a license (that's based on copyright) by using ideas you saw in the source code. You really have to actually copy, verbatim, a significant chunk of code to be in violation of a software copyright. Sure, what constitutes a significant chunk is somewhat open to interpretation, but nobody who releases their software under the GPL is likely to ask a court for a ridiculously restrictive interpretation! I'm pretty sure Richard Stallman would be horrified by what you're saying here.
I'm interested in this. From what I have read unless you do a 'clean room' rewrite, that is a team reads the code and describes the algorithm to another team with no view of the code, your work can be classified as a derivative work?
Now there are grey areas for sure, but my problem with the GPL is knowledge should be a commons, not a field day for lawyers and a creator of 'thought crimes'. Maybe I have misunderstood that regarding derivative works but it still seems by enforcing obligations GPL doesn't meet my personal definition of free.
This is really only an issue for project where you're trying re-implement a whole piece of software, and not when you pick up some "good ideas" looking at a piece of code. If you're creating a functional clone of some software and suspect that the owners of that software might take legal action against you, then you'll need to make extra sure that no part of your reimplementation resembles the original source code. Hence the whole "clean room" approach.
But if you're just looking at some code and get some ideas and use them in a totally different kind of program, you're definitely not violating the spirit of copyright, and as far as I know no court has ever decided that you're violating the letter of copyright in such a case either.
You mention "algorithm"... algorithms can't be copyrighted, and although it can be argued that they shouldn't even be patentable (as mathematical equations aren't), in the 1990s courts in the US effectively allowed algorithms to be patented "when part of an implementation". Also in some other countries algorithms were always patentable.
But that's patents, we're talking about copyright here. You absolutely can't copyright an algorithm. And "not looking" doesn't protect you from a patent anyway, so go ahead and look at the code unless you want to write a functional clone and claim that it's in no way a derivative work.
Strictly speaking, as soon as you looked at the GPL code you are tainted. A developer with a good memory will struggly not to pick up code he read through. After all, if you are looking at others people code, you are looking at the interesting parts - and if you see a cleverly written function - so anything which is longer as a single expression - you are in risk violating the copyrights of the creator. As I am respecting the works of others, I try to keep legally and morally clean and not just "being caught". So, being a professional programmer, I usually avoid looking at GPL code myself.
Not for code I just looked at. But if I find the MIT/BSD code useful enough that I want to have such a functionality in my program, I would just use the code and give attribution as the license requires.
In all the cases, where I not intend to include GPL code in my project, I consider it unethical and legally improper to look at it and then write my own version of such a functionality.
