Stephen Kell @[email protected]

@jonathoda @ilyasergey @chrisshank23 @__protected @icfp_conference @splashcon @pvh @qualmist @geoffreylitt I think from Onward!'s point of view, the main issue is that we are not on the same days as OOPSLA which we normally would be. Whether we need sessions over two or three days is a level of detail down (and programme-dependent). Let's talk about both of these by e-mail (I'll send)

@ilyasergey @chrisshank23 @__protected @icfp_conference @splashcon As chair of the LIVE Steering Committee I can confirm that we want to participate at SPLASH'25. cc @pvh @qualmist @geoffreylitt As a member of the Onward! Steering Committee I'd like to request that we get a full 3 days on the schedule as usual. cc @stephenrkell Thanks

Next year @icfp_conference will be colocated with @splashcon !

@JAldrichPL @cellllla @davidthewid I agree with the general point, but it's slippery, e.g. I see not just ethical but political issues in compilers. Huge power gap between user and compiler author; the latter form a bloc (read: big tech) skewing tools/languages to suit certain projects/parties; "whose efficiency?"

@cellllla @davidthewid Example: part of my research is compilers. There are some ethical considerations in compilers, but they are mostly of the kind you get in general engineering: build your compiler to be secure so the output code isn't exploitable, make it efficient so it doesn't waste resources.

let me get this straight… a general chair of the world's biggest conference on Artificial Intelligence and Ethics thinks it is both possible and desirable to "not mix science with politics".

@TitzerBL @samth @matt_dz I agree that moving in C++ is problematic because it is so error-prone. It is best seen as an optimisation for rare/careful/expert use. But for copy constructors and copy assignment, isn't the copying explicit? I don't see it as behind-the-back.

@samth @stephenrkell @matt_dz Moving is also problematic because it can leave dangling references and confuse the identity of mutable storage.

This is an interesting essay about mutable state but I particularly appreciate the critique of the absolute insanity that is "value semantics" (that is, implicitly copying mutable data behind your back).

@matt_dz @TitzerBL @samth I so do not understand this discussion... which is my problem. But if you feel like helping me out: when we talk about C++ in this context, what are the C++ features/idioms that are being critiqued? Isn't copying pretty explicit in C++?

@TitzerBL @samth To be fair, I'd see implicit sharing of mutable state (as in using reference semantics by default) as more problematic for reasoning about correctness (fully connected object graph != good default baseline) *or* performance (e.g., heterogeneous memory hierarchy, false sharing).

@mauricioaniche @JAldrichPL I mostly agree that it's possible to do both. "Aren't willing" might mean "have unreasonably high other workload pressures". (Not always!) Also the "teach foundations" aspect is (here in the UK) under constant attack by "teach vocational skills". So people naturally dig in....

@JAldrichPL Unfortunately, many academics hide behind the “I teach the foundations” excuse. It *is* possible to teach the foundations together with practical engineering skills. The teachers just have to work hard on creating such a course but many aren’t willing to do so

While it's true that a computer science degree should be based on principles that have enduring value, they should indeed prepare students for the software engineering jobs that most will take. That includes knowing how to start a project and what an environment variable is.

@JAldrichPL @terrible_archer It is increasingly possible to get a CS degree without an OS course. :-( I teach a (short, compulsory) OS course and it just about touches on environment vars. Students find the raw volume of "stuff" very challenging, and the practical work is what they complain loudest about...

@terrible_archer Maybe they haven't taken an OS course--or at least haven't taken a good one!

@laurencetratt @abhi9u @lemire (The first one is really an executable specification of linking, but the work never quite got that far... it was more successful as an arguably comprehensible linker written in an ML-like language.)

@laurencetratt @abhi9u @lemire I co-wrote a paper about a "very simple" (relative) linker #oopsla16b" target="_blank" rel="nofollow noopener">humprog.org/~stephen/#oops… and a blog post about a very simple dynamic linker (loader) that omits much of the important stuff (but is still fun). #elf-chain-loading" target="_blank" rel="nofollow noopener">humprog.org/%7Estephen/blo…

Linking your compiled binaries can take a lot of time.

@filpizlo @stbrunthaler @laurencetratt @RanjitJhala OK, but now imagine that the change the hacker wants to commit is 's/gcc/hcc/' in the Makefile, where 'hcc' is the one that does the checks. Make only succeeds when 'make check' tests pass. Surely this sets the bar equally high? "Converting files" isn't the framing.

@stephenrkell @stbrunthaler @laurencetratt @RanjitJhala Fil-C gives you no such joy. If you convert just one file, then your code still traps on start-up, so you only get to declare victory once you convert 1000/1000.

In my spare time, I made a memory-safe C that I call Fil-C. github.com/pizlonator/llv…

@filpizlo @stbrunthaler @laurencetratt @RanjitJhala Maybe! I suppose I don't follow why you think anyone would "declare victory" after only turning on checking for one file in 1000. What's the rationale?

@stephenrkell @stbrunthaler @laurencetratt @RanjitJhala I don't follow. Can you elaborate?

@stbrunthaler @filpizlo @laurencetratt @RanjitJhala Interesting q about how much checking would a checking-aware programmer actually check... Rust people grapple with this in a different way. It's partly cultural. But if you *can* turn it on for ~all C code, you'd have to justify *not* doing that. Work needed ofc.

@stbrunthaler @filpizlo @laurencetratt @RanjitJhala Checked C should really be called "C unioned with another language specified to require checking". Not necessarily a bad approach but you can also implement plain old C with checking... that is mine.

@laurencetratt @filpizlo @stbrunthaler @RanjitJhala Thanks for the plug. :-) Though no need for the past tense! I am only a few yak-shaves away from being right back on this.... (And "fewer changes than CHERI" is definitely a goal.)

@filpizlo @stbrunthaler @RanjitJhala @stephenrkell I think Stephen was trying to balance backwards compatibility concerns in some (but not all) of those cases, so it might partly be subtly different goals? [The largest compatibility comparison is CHERI: quite a lot of software does need a minor tweak or two before it works.]

To: the crazy ones, the misfits, the rebels, the troublemakers, the ones who see things differently Subject: you should write an Onward! paper alarmingdevelopment.org/?p=1708

@PrincetonUPress Please can you stop spamming me? You're spamming my address firstname.lastname at kcl.ac.uk. I refuse to "unsubscribe" when I never subscribed... I've already contacted you by e-mail but you ignored it. Reputable companies don't spam people.

@Zephraph @geoffreylitt @Unison I totally applaud the emphasis on systems not just languages per se... and agree it is a pity that many takes (especially academic ones) are blind to this. But going further: what's a reasonable "radius of co-design"? How can we program with tools drawn from beyond that radius?

@geoffreylitt @Unison I absolutely did. Yeah, it's been formative in my recent thoughts about software and computing environments. I've been harping a lot on programming systems over programming languages.

I feel like language design is stuck in the mindset of just considering syntax and semantics. I believe the future is seeing the problem of software (creating, editing, distributing, versioning, etc) as a wholistic problem. In that way, @unison is the future.

@AstroBio_Ben It's a great list... though I've found it impossible to stick to these sorts of practices with even a moderate departmental workload (teaching, admin, whatever). E.g. the problem is that eventually (not just in Soviet Russia) meetings schedule you.

In academia, there is always work to do. So, I've developed some rules that help me maintain my peace in this busy career. 1/4 1. No meetings on Mondays. -Mondays for me are for catching up on work and e-mails. This helps me feel productive at the beginning of every week.

I'm going 'on tour' soon... among others, going to be in the Bay Area on Tuesday 7th and Wednesday 8th November. Happy to chat with anyone who's around! Also will have some time in New Zealand and Australia the fortnight after that...

@mgrnbrg Congrats! (BTW I will be around the generalised New York area on 30th October if a flying visit / coffee might appeal... just let me know.)

😊 stevens.edu/news/building-…

@rsnous Might I ask the source? One thing printf gets right is: people prefer typing fewer character. Who wants to write "setw(2) << setfill('0') << " when you can write "%02"? Don't answer that. Of course I think it only works well for notations that are fairly commonly used.

"Effectively the advantage of the printf program is that it's easier for _other programs_ to manipulate. The idea that backfires is that program structure may be encoded in abitrarily complex ways and the only one who ever has to worry about it is the compiler writer."

"Clearly the printf interface gets something right that iostream doesn't, since it seems to save us lots of trouble. What is it that printf gets right?"