5 ms·

I am curious how it round-trips. For the output languages that are also supported input languages, can you transpile back and get something close to or identica

by drewm1980 4y ago

I am curious how it round-trips. For the output languages that are also supported input languages, can you transpile back and get something close to or identical to your input?

082349872349872 4y ago
Even if it doesn't round trip, hopefully it stabilises, so that A->B; B->A; A->B is equivalent to A->B?
- Syzygies 4y ago
  > Galois connections may be used to describe many forms of abstraction in the theory of abstract interpretation of programming languages. https://en.wikipedia.org/wiki/Galois_connection https://en.wikipedia.org/wiki/Galois_connection
  - 082349872349872 4y ago
    Since you've brought out the power tools*, may I ask you to briefly (2-5 min?) consider the last paragraph of https://news.ycombinator.com/item?id=32617344 https://news.ycombinator.com/item?id=32617344 : "if we think geometrically" et seq.? advThanksance * Galois connexions subsume, among other things, Dijkstra's weakest preconditions/strongest postconditions.
    - Syzygies 4y ago
      I believe that these "groupoid" ideas lead into higher category theory, and the supporting mathematics for theorem provers such as Lean.
      - peter_d_sherman 4y ago
        I think you may be right! Also, I think that some of the next breakthroughs for AI might be as follows: 1) Given any Turing-complete computer programming language A, construct a translator to a higher-order intermediate representation, then from that higher-order intermediate representation, construct a translator to programming language B... 2) Do this for all known Turing-complete computer programming languages; 3) Do this for all known Mathematical categories, groups, lemmas, theories, etc., and other Mathematical "objects" (for lack of a better term!) -- in other words, create an intermediate data model "map" (analagous to an intermediate higher-order language used by a transpiler) -- between all of these things in all of Mathematics... Category Theory -- seems like it would be "The next level up"... (Note that I am speculating about most of this(!) -- I claim to be an idiot in [some/many/all] of these areas! <g>)
peter_d_sherman 4y ago
That is truly a great question(!) -- because for rigorous testing/verification purposes, it is not merely enough to code what is necessary to support a target language -- round -trips (transpilation + reverse transpilation) must match the original identically (formatting aside -- and/or sent through a deterministic code formatter on both sides, pre and post...). Otherwise there is something wrong (or potentially wrong!) -- in the transpiler and/or target language and/or the data model of the target language... and the tiniest error could potentially turn into something way bigger later if not addressed at that stage... I am curious how it round-trips, too! Anyway, excellent question!

I am curious how it round-trips. For the output languages that are also supported input languages, can you transpile back and get something close to or identical to your input?

082349872349872 4y ago

Even if it doesn't round trip, hopefully it stabilises, so that A->B; B->A; A->B is equivalent to A->B?

Syzygies 4y ago
> Galois connections may be used to describe many forms of abstraction in the theory of abstract interpretation of programming languages. https://en.wikipedia.org/wiki/Galois_connection https://en.wikipedia.org/wiki/Galois_connection
- 082349872349872 4y ago
  Since you've brought out the power tools*, may I ask you to briefly (2-5 min?) consider the last paragraph of https://news.ycombinator.com/item?id=32617344 https://news.ycombinator.com/item?id=32617344 : "if we think geometrically" et seq.? advThanksance * Galois connexions subsume, among other things, Dijkstra's weakest preconditions/strongest postconditions.
  - Syzygies 4y ago
    I believe that these "groupoid" ideas lead into higher category theory, and the supporting mathematics for theorem provers such as Lean.
    - peter_d_sherman 4y ago
      I think you may be right! Also, I think that some of the next breakthroughs for AI might be as follows: 1) Given any Turing-complete computer programming language A, construct a translator to a higher-order intermediate representation, then from that higher-order intermediate representation, construct a translator to programming language B... 2) Do this for all known Turing-complete computer programming languages; 3) Do this for all known Mathematical categories, groups, lemmas, theories, etc., and other Mathematical "objects" (for lack of a better term!) -- in other words, create an intermediate data model "map" (analagous to an intermediate higher-order language used by a transpiler) -- between all of these things in all of Mathematics... Category Theory -- seems like it would be "The next level up"... (Note that I am speculating about most of this(!) -- I claim to be an idiot in [some/many/all] of these areas! <g>)

peter_d_sherman 4y ago

That is truly a great question(!) -- because for rigorous testing/verification purposes, it is not merely enough to code what is necessary to support a target language -- round -trips (transpilation + reverse transpilation) must match the original identically (formatting aside -- and/or sent through a deterministic code formatter on both sides, pre and post...). Otherwise there is something wrong (or potentially wrong!) -- in the transpiler and/or target language and/or the data model of the target language... and the tiniest error could potentially turn into something way bigger later if not addressed at that stage... I am curious how it round-trips, too! Anyway, excellent question!