Reini, it seems like you think everyone who disagrees with you is stupid. This isn't an explanation for anything.
We were boxed in for two really good not-primarily-technical reasons:
* we had users we didn't want to abandon or cause churn * we had architectural decisions we had to improve
Now I know I'm not as smart or as experienced or as knowledgeable as you are, but that doesn't mean I'm a drooling buffoon, and it certainly doesn't mean I don't have good reasons. Sometimes it means I make mistakes. Sometimes they're not even mistakes; sometimes they're just differences of opinion.
It is the simpliest explanation. In fact it was stupid and arrogant leadership mostly.
There was no disagreement, as we didn't say a beep then. They made those stupid decisions all by themselves, and didn't listen to any advice. The better people just silently left, shaking their heads.
It's the laziest explanation.
Did you ever ask yourself why, for example, all of the calling convention code paths were consolidated into a single code path? It wasn't to make a single, inefficient code path as you claim.
It was to provide a gradual transition for clients to a better designed calling convention system which could then be optimized for actual client uses.
The goal was never to create the fastest possible VM at every single point in time. That's where I think you never understood the goal of the project, and that's where I think you've never understood the goal of p5p.
The goal was to make something that works, continues to work, and can be improved while continuing to work. That's why you're no longer welcome in p5p -- because your goals and your actions are incompatible with that.
I wish you understood this. You're very smart and very talented and you have a lot to offer, if you can get out of your own way and accept that people who don't share your exact goals in the exact same way aren't irredeemably stupid.
How odd then that ~90% of the discussions about ditching Parrot were primarily complaining about the deprecation policy and first-class Rakudo support, not speed.
I suppose you would know better though, being objectively smarter than everyone else who ever made a decision about the project. How unfortunate that we can't take your word for it; we can only read all of the public discussion about it.
It was a CPS compiler, but after they drove the first creators away, a bunch of people who had no idea about compilers nor CPS started destroying it. First the destroyed the jit. The technical justification was bullshit. Then they destroyed the run loops, and finally the CPS (2007 YAPC workshop). This caused every function call to allocate 2 objects, continuation contexts were not reused anymore, but created on the fly for every call. After this desaster they gave up (their new idea was to create a 2nd LLVM, called something with m), and I came back to cleanup the mess. I readded the compiler optimizations, added proper lock-less threading (their current VM is lockful) and started fixing the CPS, the run loops, the GC and the Jit. But before that they decided to kill it off, and rewrote everything by themselves. But not as committee anymore and without the previous leadership, which was a very good decision. It's still a very naive architecture still, that's why it's even slower than python, the slowest scripting language on earth.
Realizing that this is futile, I started improving perl5 towards perl6 instead. Just without breaking backcompat. perl11 (5+6)