ESR is making an early invalid assumption - that "fast transparent garbage collection will happen".
Sorry, no. The smartest people in the CS world - possibly the
smartest in the world, period (specifically those at MIT AI Lab,
Xerox PARC, BBN, TJ Watson, and Stanford) worked the GC problem
for literally 20 years, throwing hardware at it, software, tagged
architectures, secondary processors, all that.
They never cracked it. GCing at realtime speed is just a tough problem.
Unless ESR can show me code that can GC in faster than O(n) time
AND not have to freeze the allocator process for O(n) time, he's just
pitiably wrong.
(and no, I don't count flip and sweep GC as workable in this, as it
means that a buffer that DMA hardware is writing to will move without
warning. Nor is "generational" GCing, all that does is to stave off the
inevitable full-out GC for a few minutes to hours, which is fine for a
hacker sitting at a terminal but no good at all for a self-driving car or
SaaS server).
Now, I could be wrong; if he *has* a realtime garbage collection algorithm
then he deserves the Turing award.
But I'm betting "not".