The second iteration of Rebis is out!

[mthom]

.. on the rebis-dev branch! I'm sorry it took so long. I made quite a few changes.

If you want to add new instructions, you should modify the ClauseType structs in crates/instruction-template/src/lib.rs. instructions.rs is now generated programmatically using the syn crate. Also, the atom! macro is now used to render the string representations of atoms to their new WAM representation as relative offsets into the atom table. I should write all this down in the Wiki eventually. It could be the subject of our first meetup, maybe? Among other changes of course.

[triska]

Wow, thank you a lot Mark!

[mthom]

You're welcome!

[pmoura]

Great news. Is the branch in sync with master? Should we favor the rebis-dev branch for testing going forward?

[triska]

I think this is due to a simple typo in:

scryer-prolog/crates/instructions-template/src/lib.rs

Line 265 in 83378e4

#[strum_discriminants(strum(props(Arity = "2", Name = "working_directory")))]

This should read "$working_directory", i.e., with $.

[pmoura]

There's some other bug. After fixing the typo you mentioned, this works:

?- working_directory(Chars, Chars), atom_chars(Directory, Chars).
   Chars = "/Users/pmoura/scrye ...", Directory = '/Users/pmoura/scryer-prolog'.

But with a bug.pl file with the contents:

:- use_module(library(files), [working_directory/2]).

d(Directory) :-
	working_directory(Chars, Chars),
	atom_chars(Directory, Chars).

We get:

?- [bug].
   true.
?- d(D).
false.

[triska]

I can reproduce this! I have filed #1224 for the issue about $working_directory, could you please file this additional problem as an issue so that we can track it? Thank you a lot!

[triska]

I think Mark has resolved this already, I can no longer reproduce this issue with the latest rebis-dev version!

[pmoura]

Thanks for the fixes. Next bug (preventing loading Logtalk) is #1231.

[dcnorris]

Looking forward to running some benchmarks! Having these build errors, tho:

david@riesenheim:~/scryer-prolog$ git show origin
commit d0b74a95f48eb2ffdc88978a117acd422b2753ff (origin/master, origin/HEAD, master)
Merge: c60ada8 9d06229
Author: Mark Thom <markjordanthom@gmail.com>
Date:   Thu Jan 6 20:46:59 2022 -0700

    Merge pull request #1156 from mthom/proper_var_list_order_issue_1121
    
    Correct order of variables in read_term/3 equation and variable lists

david@riesenheim:~/scryer-prolog$ git branch
  master
* rebis-dev
david@riesenheim:~/scryer-prolog$ cargo run --release
   Compiling to-syn-value_derive v0.1.0 (/home/david/scryer-prolog/crates/to-syn-value_derive)
   Compiling openssl v0.10.38
   Compiling sodiumoxide v0.2.7
   Compiling git-version v0.3.5
   Compiling markup5ever v0.8.1
   Compiling html5ever v0.23.0
   Compiling phf_generator v0.9.1
   Compiling string_cache v0.7.5
error: expected expression, found `.`
  --> crates/to-syn-value_derive/src/lib.rs:19:6
   |
19 |     }.into()
   |      ^ expected expression

   Compiling phf_macros v0.9.0
error[E0308]: mismatched types
  --> crates/to-syn-value_derive/src/lib.rs:9:5
   |
9  | /     quote! {
10 | |         use to_syn_value::*;
11 | |
12 | |         impl ToDeriveInput for #ty_name {
...  |
18 | |         }
19 | |     }.into()
   | |_____^ expected `()`, found struct `TokenStream2`
   |
   = note: this error originates in the macro `quote` (in Nightly builds, run with -Z macro-backtrace for more info)

For more information about this error, try `rustc --explain E0308`.
error: could not compile `to-syn-value_derive` due to 2 previous errors
warning: build failed, waiting for other jobs to finish...
error: build failed

[josd]

I can report already 3.6–3.7x speedups of some calculations. Thank you, Mark! Will post a link to the code once benchmarks done and annotated.

This is confirmed while running the deep taxonomy benchmark from the late Harold Boley.
For the master branch we get

$ scryer-prolog -v
"v0.8.123-750-gd0b74a9"

$ time scryer-prolog -g run,halt retina.pl dt.pl
[] :retina """n1(z)""".
[] :retina """n10(z)""".
[] :retina """n100(z)""".
[] :retina """n1000(z)""".
[] :retina """n10000(z)""".

real    0m23.151s
user    0m22.166s
sys     0m0.980s

and for the rebis-dev branch we get

$ scryer-prolog -v
"v0.8.123-793-g693b6d2"

$ time scryer-prolog -g run,halt retina.pl dt.pl
[] :retina """n1(z)""".
[] :retina """n10(z)""".
[] :retina """n100(z)""".
[] :retina """n1000(z)""".
[] :retina """n10000(z)""".

real    0m5.969s
user    0m5.422s
sys     0m0.539s

That's really great progress!

[mthom]

I am slightly disappointed it's not greater still. But there's a lot more that I can do yet.

[dcnorris]

"I'll wait."
— Arnold Schoenberg, upon hearing from Jascha Heifetz that his Violin Concerto would require a sixth finger

[dcnorris]

Using v0.8.123-860-ge6c4ecf built just now, I can report a further ~1.4x speedup since my last post 16 days ago (net 5x!):
dcnorris/precautionary@2f92abe

[dcnorris]

Here's a little bit of https://en.wikipedia.org/wiki/Gustafson's_law for you:

%% Enumeration of all trial paths:
%?- J+\time((D = 6, length(Ds, D), maplist(=(0/0), Ds), setof(Path, phrase(path([]-Ds), Path), Paths), length(Paths, J))).
%@    % CPU time: 304.874s % D=6 (rebis-dev)
%@    J = 2890.
%@    % CPU time: 120.060s % D=5 (rebis-dev)
%@    J = 1162.
%@    % CPU time: 43.237s  % D=4 (rebis-dev)
%@    J = 442.
%@    % CPU time: 222.425s % D=4 (master)
%@    J = 442.

Reading from the bottom up, you see problem size increasing substantially (J is the number of paths in a traditional 3+3 dose-escalation trial run on D prespecified dose levels) with time held (roughly) constant.

[aarroyoc]

Fantastic news Mark!! About the meetup, I think with these changes is the perfect time to do it. About the date and time of that meetup, I think you're the one to choose one that you're comfortable with. Then we can discuss the agenda in #1089