Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

TLA+ Action Properties

hillelwayne.com

Post reply on HN
Print view View original

31–38 of 38 posts

Re: TLA+ Action Properties

#31

Post by lakecresva »

pron wrote: post #19
jacques_chester wrote: post #18

Earlier quoted context omitted.

> much of which is immediately familiar to anyone who reads maths Most programmers don't read math. A tool pitched to programmers gains no advantage from being readable primarily by mathematicians.

But that tool is mathematics. As far as I know -- at least, that's what things were like in my time in school -- when programmers learn mathematics they use standard notation, not their own. When they use the tool called natural language to write documentation, they also use standard notation, not their own, and when they study diagrams they (at least I) prefer looking at a pictures with lines rather than SVG source.…

I think your judgment is clouded by this insistence that you're doing math rather than something which is squarely in the middle of the cs/math venn diagram. Relational algebra has an accepted notation for left outer-join (⟕). Do you think `I><|` is preferable to `LEFT JOIN` because mathematics™? If I were to sample 100 TLA+ specifications, do you think I would find more people specifying things like paxos and concurrency models, or would I find people doing algebraic geometry and number theory?

Re: TLA+ Action Properties

#32

Post by pron »

lakecresva wrote: post #31
pron wrote: post #19

Earlier quoted context omitted.

But that tool is mathematics. As far as I know -- at least, that's what things were like in my time in school -- when programmers learn mathematics they use standard notation, not their own. When they use the tool called natural language to write documentation, they also use standard notation, not their own, and when they study diagrams they (at least I) prefer looking at a pictures with lines rather than SVG source.…

I think your judgment is clouded by this insistence that you're doing math rather than something which is squarely in the middle of the cs/math venn diagram. Relational algebra has an accepted notation for left outer-join (⟕). Do you think `I><|` is preferable to `LEFT JOIN` because mathematics™? If I were to sample 100 TLA+ specifications, do you think I would find more people specifying things like paxos and concur…

If you were to sample 100 TLA+ specifications, you would find 100 descriptions of software systems using mathematical formulas. Physicists don't use different mathematical notation when they write an ODE for a spring or electrical current. Also, I don't understand what notation would be appropriate. Functions in Haskell are much more similar to functions in Java than they are to either operators or functions in TLA+. You could use Java or Haskell notation, but what you'd be using it for would still behave differently from either one, which would only be more confusing. Even when it comes to simple things like conjunction and disjunction, over the past 50 years, programming languages have used at least three different popular representations, while over the last 100 years, mathematics have used one. Why is picking one of the three superior for writing mathematics in TLA+ than picking the one?

Also, you should know that specification languages based on programming languages were the norm before TLA+. TLA+ intentionally broke away from that to offer something different, because Lamport believed it is simpler. I agree, but if you don't, you can use the specification languages that preceded TLA+, like Isabelle and Coq. So you're going to the one language that tried to make things easier by changing things up and asking, why did you change things up? Despite all that, after a three-day workshop, programmers use TLA+ to specify large and complex software systems, something that those who learn the programming-language-based specification languages do not, so perhaps Lamport was right.

Re: TLA+ Action Properties

#33

Post by pron »

jacques_chester wrote: post #29
pron wrote: post #19

Earlier quoted context omitted.

But that tool is mathematics. As far as I know -- at least, that's what things were like in my time in school -- when programmers learn mathematics they use standard notation, not their own. When they use the tool called natural language to write documentation, they also use standard notation, not their own, and when they study diagrams they (at least I) prefer looking at a pictures with lines rather than SVG source.…

So is it for programmers or is it for mathematicians?

It is mathematics for anyone who wants to describe discrete and hybrid systems, including programmers. But I believe mathematics is used by mathematicians, physicists, engineers of many kinds, and biologists for many different things -- that's what mathematics is for -- and they all use pretty much the same notation. Also, and this is important, there is no standard programming notation for mathematics. TLA+ operators and functions simply don't behave like subroutines in any programming language. I think using the same symbols, but with a different meaning, would be more rather than less confusing.

For example, when in TLA+ you specify the function, f, from the integers to the integers such that f(x) = -f(x), you've obviously defined the zero function; I'm not aware of any programming language that works like that. While TLA+ doesn't have macros, the value of x = 3 ∧ Foo(x) = 5 might not be equal to x = 3 ∧ Foo(3) = 5, and I'm not aware of any non-logic programming language that works like that, either, at least not without some hidden complexity (simple mathematics also doesn't work like that, but this is because TLA+ is a modal logic), and logic programming languages have syntax that isn't any more familiar to non-logic programming language programmers than TLA+'s.

Re: TLA+ Action Properties

#34

Post by pron »

hwayne wrote: post #23
pron wrote: post #6

Earlier quoted context omitted.

First, what you posted (and the article uses), is not really TLA+ syntax, but its ASCII source; it's like publishing the LaTeX source for your typeset maths. TLA+ syntax looks like what you see here: https://pron.github.io/posts/tlaplus_part2 Some TLA+ bloggers publish the typesetting source rather than actual pretty-printed TLA+; I'm not thrilled with that, but there's a good, mundane, reason for doing it: there is…

It's still a barrier for beginners. I teach both TLA+ and Alloy and people struggle with remembering TLA+ tokens much more than they do with Alloy, which has synonyms for common operators. You can write `implies` instead of `=>` if you want, which makes the teaching experience much smoother. If I could have students use `&&`, `||`, and `!` instead of ⋀ ⋁ ¬ that'd cut out a lot of early friction. This is also why I "p…

I agree it's a barrier for beginners, but I think this is a good barrier, which makes the next, bigger ones, easier to overcome. A step is also a barrier, but I'd rather face more, lower steps than fewer taller ones. By the time you realise neither operators nor functions work at all like functions from programming, the syntax has helped to put your mind in a "this is not programming" mode. Also, || and && are almost universally used in programming languages today, but they weren't thirty years ago, and might not be thirty years from now (Zig doesn't use them and neither does Nim), and even in C, C++, and Rust they have similar but subtly different meaning than | and & involving evaluation, a concept that TLA+ doesn't have -- now there's a more interesting and more important unfamiliar TLA+ thing -- while ∨ and ∧ have been used much longer than that and aren't going anywhere. So every person might prefer something that's already familiar to them, but there is no alternative that's really universal. If you go to Wikipedia -- say, https://en.wikipedia.org/wiki/Logical_connective -- they use "TLA+ notation," not C notation.

Anyway, there are other, and more important, things in TLA+ that are much less familiar from programming (e.g. there's no evaluation) than how to write logical or, and still TLA+ has been more successful at teaching programmers to specify more complex systems more quickly than any other specification language I'm aware of.

Re: TLA+ Action Properties

#35

Post by sfdjrmw3 »

lakecresva wrote: post #16
pron wrote: post #6

Earlier quoted context omitted.

First, what you posted (and the article uses), is not really TLA+ syntax, but its ASCII source; it's like publishing the LaTeX source for your typeset maths. TLA+ syntax looks like what you see here: https://pron.github.io/posts/tlaplus_part2 Some TLA+ bloggers publish the typesetting source rather than actual pretty-printed TLA+; I'm not thrilled with that, but there's a good, mundane, reason for doing it: there is…

I think you'll be more successful as an advocate for TLA+ if you just say "yeah, the syntax isn't for everyone", or "the tooling has some issues" instead of writing long missives about how syntax like `[][x' /= x => y' = x]_ >` is actually motivated by some deep mathematical insight, or suggesting that the desire for an editing environment that doesn't feel 20 years old stems from a failure to understand the differen…

Do you know TLA+? You sound like you've totally missed the point.

Re: TLA+ Action Properties

#36

Post by sriku »

I find it easier to chunk and read "always eventually" - i.e. the combination "[]" - as "progresses" and "eventually always" - i.e. the combination "[]" - as "stabilizes".

So []A can be read as "A progresses" and []A as "A stabilizes".

Re: TLA+ Action Properties

#37

Post by neolog »

barakm wrote: post #26
gambler wrote: post #2

[][x' > x]_x TLA+ seems like an incredibly useful tool, but damn it could do with a better syntax. Math notation is designed to be written on paper and to handle extreme repetitiveness when you do proofs or calculations. Both of those premises are largely irrelevant in the domain TLA+ is intended for.

I've started working on doing TLA+-style models but using Python as the language for doing so (disassembly and all) with an easier to deploy model checker. Super super early and I'm not quite yet ready to announce it, so don't expect miracles (or post it everywhere) but if you're reading this comment and want to see hacks in this space... https://github.com/timewinder-dev/timewinder

How does the performance compare to TLA+? Especially curious if you've run it under PyPy.

Re: TLA+ Action Properties

#38

Post by pron »

michael_j_ward wrote: post #22
hwayne wrote: post #10

Earlier quoted context omitted.

We're seeing more people create DSLs that compile to TLA+. For example, salt lets you write TLA+ specs in Clojure: https://github.com/Viasat/salt

wow, that feels extremely more approachable. Are there downsides to using that?

I haven't used salt, but the thing to understand is that even when the syntax used is Clojure, the semantics is TLA+'s. So, for example, if you use a Clojure function syntax to define a TLA+ operator, that will not behave like a Clojure function, but like a TLA+ operator, which behaves very differently from Clojure functions. So the symbols on the screen might be more familiar, but their meaning will be different from their meaning in Clojure. I think that is more confusing, which would be a downside, but if you find that more approachable -- great, use that.
Post reply on HN