LLVM is truly a wonder of modern software. By making the development of new programming languages easier, it's accelerating the development of new tools and techniques.
I really don't understand this point of view. LLVM is really cool, but its not revolutionary. It, like most (all?) other modern compilers uses an IR. GCC - the obvious comparison - has gone through at least two over the years. The original 1987 release had one. Promoting IR as an interop layer isn't new either (1992: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.41.1...). The JVM+CLR were both around by the time LLVM came to be too; CLR was supposed to be multi-language from the start. Modern GCC's IR layers were around not much later.
At the time apple bet on LLVM, it wasn't nearly as polished as it is now: they didn't pick it because it was great, but because of the liberal licensing (just like khtml, for that matter).
That's not to denigrate LLVM - clean, modern code that's easy to use is wonderful - but that doesn't make it a wonder. If I had to guess, I bet it's like 99% of other projects: lots of work, copying good ideas liberally from several sources, and solving the right problem at the right time.
The major difference with GCC's and other IRs is that LLVM and clang made it easy to use and embed the APIs in other software (due to the license as well as exporting a library). This made it much more convenient to make compilers, as well as code inspection and analysis tools (and even graphics drivers).
Sure, everything was possible before, but much harder for the average programmer interested in this stuff. With just Python and LLVM-python you can do very interesting things, quickly, without delving into GCC sources.
It may be possible that GCC cached up in that aspect, but back in the day I heard from language researchers, the GCC low and high level IR were a hell to introspect and required delving into the GCC source at a deep level.
The difficulty of accessing GCC's IR is partially deliberate - Stallman didn't want any convenient way for non-free software to take advantage of GCCs internals.
It would seriously surprise me if Stallman had much of anything to do with the current GCC IR, which is from 2003. But maybe I'm mistaken.
Do you have any reference for this intentional obfuscation in any version? It'd be disappointing even if it were the case in earlier versions, but conceivable, I guess...
