I believe our editor already asked you to stop with the personal insults...

Posted Sep 23, 2020 21:13 UTC (Wed) by foom (subscriber, #14868)
In reply to: I believe our editor already asked you to stop with the personal insults... by farnz
Parent article: Lua in the kernel?

> Except that BPF has no such requirement

Oh come on, this is a useless nitpick.

There is effectively no use for eBPF, except to feed it to the Linux kernel. And for that purpose it must be verifiable. So, yes, fine -- when I said "BPF requires" I really meant "BPF-with-verifier-as-implemented-by-the-linux-kernel-for-use-from-user-space requires". _That_ is the target we're talking about in this whole thread, not the practically-useless "BPF-without-verifier" target.

I believe our editor already asked you to stop with the personal insults...

Posted Sep 24, 2020 10:23 UTC (Thu) by farnz (subscriber, #17727) [Link]

But that limit can be raised, too - it's an artificial limit the kernel imposes so that it can safely accept user programs that get to hold the CPU. We can change that limit if it's a problem - we can (for example) allow programs that can't be proven to terminate a fixed number of executed instructions after which a default result is produced.

Which puts us right back at the original problem that HelloWorld is continuing to not answer; what practical programs would you like to express in C that cannot be compiled down to eBPF right now, and can we fix that up?