ugexe i guess thats what github.com/rakudo/rakudo/blob/b830...=1#L54-L59 is talking about 02:39
02:47 lucs joined
lizmat ok, will refrain from adding ad-hoc typechecks... 10:32
in the hope of some future automatic type checking logic 10:33
meanwhile, I'm trying to trace why Identity::Utils.bytecode-io is not pointing to existing bytecode files
and it looks (at least to me) that it sometimes gets the prefix wrong, even though it is using the repo provided by the compunit 10:34
this problem then breaks Code::Coverable, and that in turn breaks Code::Coverage, which breaks Test::Coverage 10:37
theory: could it be that if a module is installed, the bytecode winds up in the "site" repo 10:42
which is typically install/share/perl6/site/precomp/ 10:43
however, when rakudo gets compiled again, and the module is loaded again, that the bytecode file winds up in ~/.raku/precomp ? 10:44
m: use NativeCall; my size_t $a .= new 11:56
camelia ( no output )
lizmat m: use NativeCall; my size_t $a .= new; dd $a
camelia 0
lizmat m: Q|use NativeCall; my size_t $a .= new; dd $a|.AST.EVAL
camelia Can only compile an assign initializer on a native
in any IMPL-TO-QAST at src/Raku/ast/variable-declaration.rakumod line 1853
in any IMPL-TO-QAST at src/Raku/ast/statements.rakumod line 1209
in any IMPL-TO-QAST at src/Raku/ast/statements.r…
lizmat ugexe: ^^ this breaks LibGit2 11:57
m: use NativeCall; my size_t $a = size_t.new 11:58
camelia Cannot instantiate a native type
in block <unit> at <tmp> line 1
lizmat so it looks like legacy is silently ignoring an error there 11:59
12:17 kjp left 12:18 kjp joined
[Coke] did another run through b83057b10c, only 3 failures, Polyglot::Regexen, GLib, JSON::GLib::Node 13:31
(so, no change, really) 13:32
updated github.com/coke/raku-blin-release-...ailures.md 13:33
(note that the table at the bottom of the redo runs are not complete, it's only those items tested specifically during the redo, most of which are the moar-induced zef failures.
(and also why the run only took 18 min) 13:34
I think we need to get an nqp+rakudo with the moarvm revert that liz did (and hopefully that clears the additional failures we're seeing in the first-run blin runs)
ugexe so many incorrect core ast class method types... 14:28
some of them aren't even actual types/classes 14:29
[Coke] can you give an example?
ugexe Raku::ArgList, Raku::Regex, RakuAST::Atom, RakuAST::Args, and like 7 others 14:36
i'll hopefully have a pr with the type checking and corrections to all these types today or tomorrow
[Coke] oh - looks like liz already pulled that moarvm change into nqp?
so, ugexe - that moarvm change didn't impact the blin run - still many of those zef can't find the module errors that go away when not under huge load. 14:39
lizmat yeah, it's in Rakudo also already 15:17
lizmat gives up on libgit2 module for now (it needs unbitrotting even without RakuAST) 16:17
hmmm judging from docs.raku.org/type/IO/Handle#Buffe..._terminals we don't have a way in Raku to unbuffer a terminal *unlike Python's -u argument?) 17:53
Geth rakudo/main: df6b1f8400 | (Elizabeth Mattijsen)++ | src/Raku/Grammar.nqp
RakuAST: typed-panic is spelled with a hyphen

And not an underscore. Probably a copy-pasto from the legacy grammar.		 20:44
lizmat spotted in 6.c-errata 20:45
bisectable6: dd my num $a 20:49
bisectable6 lizmat, Will bisect the whole range automagically because no endpoints were provided, hang tight
lizmat, Output on all releases: gist.github.com/f0e8cae5a01675fdce...51d6cdf6d8 20:50
lizmat, Bisecting by output (old=2018.10 new=2018.11) because on both starting points the exit code is 0
lizmat, bisect log: gist.github.com/38fa7a86f51fa161f7...0e58d7a2e5
lizmat, (2018-10-27) github.com/rakudo/rakudo/commit/46...96034408c8
lizmat, Output on all releases and bisected commits: gist.github.com/5ec1e9913c9439b295...ed98653ffb 20:51
lizmat ok, so 6.d made it default to 0e0 instead of NaN 22:22
m: use v6.c; dd my num $a # indeed
camelia NaN
lizmat m: Q|use v6.c; dd my num $a|.AST.EVAL 22:23
camelia 0e0
lizmat aha!
Geth nqp: ugexe++ created pull request #877:
Check the object types declared on parameters		 22:49
rakudo/main: 7a3abc5aba | (Elizabeth Mattijsen)++ | src/Raku/ast/variable-declaration.rakumod
RakuAST: make my num $a initialize to NaN in 6.c

In 6.c the default state of a native num was NaN, as opposed to 0e0 in 6.d and later. The RakuAST bytecode genning did not take this into account: it now does.
Also skip setting of init values for int, uint and str, as they were never actually used in the previous version of this code, and keeping them in there with this change, would generate unneccesary initializing bytecode in those cases.		 23:18
lizmat
.oO( that's two fails in 6.c-errata fixed )
sleep&
ugexe im kind of surprised that doesnt break anything 23:20
optimizations would look at things like NaN vs 0e0
[Coke] about to try a blin run as soon as 7a3abc5aba is baked. 23:27
linkable6 (2026-09-18) github.com/rakudo/rakudo/commit/7a3abc5aba RakuAST: make my num $a initialize to NaN in 6.c
[Coke] We'll see.
Geth nqp/main: 63a83ef00a | (Nick Logan)++ (committed using GitHub Web editor) | 5 files
Check the object types declared on parameters (#877)

Previously a parameter's object type, such as QAST::Var $var, was recorded on the routine for multi dispatch and otherwise ignored, so a plain sub or method took anything. Only native types were enforced, by the unbox.
... (16 more lines)		 23:38
rakudo: ugexe++ created pull request #6693:
Bump NQP to check the object types declared on parameters		 23:45
ugexe after that one passes CI and gets merged, there will be a really big PR that corrects all the typing of Raku/ast/*.rakumod code 23:46
also re: the optimizer comment, it would be for rakuast specifically since the optimizer was developed without that change existing unlike legacy 23:49

Powered by Cro and the Raku® Programming Language.

Please report any issues / comments / feature requests as an issue on App::Raku::Log.

Thank you!