13 Comments

I found your tangent discussing Milman Parry and his lineage at 32:27 interesting. You have touched on myth and narrative here and there in your writings. Did William Alfred have any direct influence on how you think about narrative in general today? Enjoyable episode of the podcast, I added Albert Lord's The Singer of Tales to my Amazon wishlist after listening to this.

Expand full comment

Glad you liked that!

> Did William Alfred have any direct influence on how you think about narrative in general today?

I don't know. It seems likely that some things he said sunk in, but I can't remember specifics. It was a very long time ago!

I think I was mainly struggling to learn the language. Old English is one ancestral source for modern English, but it's much closer to modern Scandinavian languages, especially Icelandic. (When I still remembered it, I could basically read Icelandic newspapers.)

Beowulf begins:

Hwæt! We Gardena in geardagum,

þeodcyninga, þrym gefrunon,

hu ða æþelingas ellen fremedon.

Oft Scyld Scefing sceaþena þreatum,

monegum mægþum, meodosetla ofteah,

egsode eorlas. Syððan ærest wearð

feasceaft funden, he þæs frofre gebad,

weox under wolcnum, weorðmyndum þah,

oðþæt him æghwylc þara ymbsittendra

Expand full comment

People's reactions to the CHERI project are very revealing about their lineage. Like, Howie Shrobe, who at one point our DARPA Programme Manager, was like: you've written a C compiler for a LISP machine, this is outrageous. I guess if you're from MIT CSAIL, CHERI does look like a funny kind of LISP machine that doesn't even run LISP and for some inexplicable reason has a C compiler instead. Or ,it's designers come from a different lineage that includes CAP and Burroughs B5000, which had Algol compilers....

And then other people will be like, "Wait, CHERI CCall is a Smalltalk message pass with hardware support." Which gives away their lineage.

Rorshach Test for which Programming Language or Operating System lineage you're from.

Expand full comment

Oh, that’s funny! Howie Shrobe was my first research mentor, when I did a project in the Programmer’s Apprentice group.

I’m not sure what the objection is, though. You could easily write a lisp compiler, or one for any language, for CHERI, unless I’m missing something important.

The value is that the hardware does the security critical checks, so they’re efficient, and also there’s no way (ideally) for malware to get around them, or for programmers to say “well this program really has to run fast, so I’ll turn it off.”

Expand full comment

Yes, clearly you could write a LISP compiler for CHERI, if you wanted one.

The joke was more, if you're from MIT CSAIL, everything looks like a LISP machine.

Expand full comment

I guess, being from the former MIT AI lab, nothing looks to me like a lisp machine, which is a great disappointment!

Expand full comment

> There’s a pretty-good AI-generated transcript available via a button, if you view this in the Substack app or on the web

I cannot find this in the mobile app or the web view

Expand full comment

On the web, it's in the row of buttons under the author names, at the right end.

I can't find it in the app either, sorry! It's supposed to be there. Substack features are rather annoyingly haphazard.

Expand full comment

No worries

My comment was more of a “in case author view is different from reader view”

Expand full comment

Oh, and when I said "doesn't even run LISP" -> no=one was interested in writing a LISP compiler for CHERI, not that it couldn't be done.

On the other hand, when we need formally verified code to run on CHERI, turns out the provably correct compiler we have is for ML, so those functional programmers will get us in the end.

Expand full comment

At some point, while resolving some continuous integration error in the CHERI build, I am thinking to myself, "If I didn't know better, I;'d swear those guys are playing some kind of game where they get as many programming languages into the build process as possible."

C naturally. C++ too, naturally, Your test scripts are in Python, for testing framework reasons. A bunch of the actual tests have to be in assembly language,. BlueSpec, used as a hardware description language, is kind of Haskell. The magic for doing tandem execution of the hardware against its formal specification relies on compiling the formal spec into executable ML. Proof automation in HOL4 is in ML naturally... The fuzzer that generates random test sequences for tandem execution is in ML....

Expand full comment

Many years ago, while working on reference implementations of the OSI communications protocols, particularly X.509, I somehow ended up writing in LISP a compiler for ASN.1 that could handle the wonderous ASN.1 MACRO notation. This ,,, got me a hard stare from the dev lead.

Expand full comment

"People will say, oh, it's just like lex and yacc and then five minutes later, realise they don't know how lex and yacc work."==Steve Kille, on writing parsers for ASN.1.

Expand full comment