LLVM exists because its original developers wanted to do research, not because of GPL and at the time there wasn't even a C compiler for it, not to mention that GCC wasn't designed at all to do what LLVM did.

The license angle with Apple did happen because of GCC's switch to GPLv3 but that came years later - and GCC still wasn't (and AFAIK still isn't) designed to be used like LLVM or (later) Clang, like using it as a backend for IDEs, analyzers, language servers, etc. The two projects aren't interchangeable with only the license being the thing that differentiates them.

They could have very well used lots or reusable parts of GCC. It was possible, and were many projects trying to change GCC.

gcc was specifically designed to not have many reusable parts to avoid people ripping them out. modularity only happened after LLVM and clang specifically started eating its lunch.

> LLVM and clang specifically started eating its lunch

I don't know how you can eat GCC's lunch. I feel like in this analogy GCC eats from a communal trough and is very happy to share. A company that sells licenses can lose customers, the GNU project just wants the world to benefit. GCC is still ubiquitous and has plenty of developers and mindshare, what more could they want?

Basically most of the embedded folks that used GCC forks are now on clang, and if in the old days they were breaching GPL, nowadays they are fully compilant, so not upstreaming is not even a legal issue any longer.

Fair enough. I guess you're talking about microcontrollers rather than lightweight CPUs, at least in my experience it's still standard to use GCC on the sort of dinky ARM-based devices my company produces.

I'll note that GPL never says anything about upstreaming. It can certainly make life easier if you want to keep your patches up to date and hand off maintenance responsibility, but it suffices to make a written offer of source code under the same license (on a conventional medium) to your customers to comply with any GPL version. You're never obliged to upstream or even publish your code.

Sure, but they didn't even do that most of the time, with MIT that isn't even part of the conversation.

Most certainly not. The modularity was not great, but BIG chunks of code could be copied. I did that myself.

I don't even understand how you can point to LLVM as an example of this failing. GCC is still alive and well, and perfectly suited for the task it was always intended to fill. People that wanted a commercial C compiler always had the option to purchase one at-cost, GCC was intended as an alternative to that process, not the alternative. It may be hard to remember, but less than 50 years ago squabbles over compiler licensing was what most developers had to deal with.

Respectively, LLVM fills it's own gap nicely. It has the classical BSD-like issue where hundreds of unique and disperate forks exist with no roadmap to bring them together, but that's also technically a "feature" of each license and not a bug. If GCC developers felt their job was obsoleted by Lattner, there wouldn't be a reason to update and maintain it. The end result has been healthy competition, accountability for either team, and better optimizations for everyone using either compiler.

Also after LLVM came to be, GCC gained on its official collection Go, D, and Modula-2 frontends.

And, hopefully, Rust soon as well[1][2][3].

[1] https://rust-gcc.github.io/

[2] https://github.com/Rust-GCC/gccrs

[3] https://blog.rust-lang.org/2024/11/07/gccrs-an-alternative-c...