Geth rakudo: ugexe++ created pull request #6618:
RakuAST: only decide a compile-time dispatch for a Routine
00:19
rakudo/main: 75b29a1de7 | (Nick Logan)++ | 2 files
Assert lexical lowering by the sentinel value in the QAST shape tests

Previously the lex2local shape tests told a lowered declaration from a kept one by its decl alone: static meant lowered away, contvar meant kept. A static declaration also carries a real container when a frame declares its containers without cloning, so the decl alone cannot say which of the two it is. This makes the helpers look at the declared value, and a lowered declaration is the one whose value is the Rakudo::Internals::LoweredAwayLexical sentinel.
00:25
rakudo/main: 00ccf31f60 | (Nick Logan)++ | 7 files
Share unit lexicals between BEGIN code and the mainline

Previously a closure created by code run at BEGIN time, such as the phaser a variable trait adds through Variable.willdo, resolved a unit lexical to the serialized master container, while the unit frame cloned that master on first read. What the mainline saw depended on who touched the container first: a read in the mainline before the traited routine ... (20 more lines)
rakudo/main: 99be3abe19 | (Nick Logan)++ (committed using GitHub Web editor) | 9 files
Merge pull request #6616 from ugexe/ugexe/unit-lexicals-static

Share unit lexicals between BEGIN code and the mainline
rakudo/main: 4 commits pushed by (Nick Logan)++ 00:34
rakudo/main: aaa7d4c83c | (Nick Logan)++ (committed using GitHub Web editor) | 2 files
RakuAST: only decide a compile-time dispatch for a Routine (#6618)

Previously the compile-time dispatch analysis accepted any Code object bound once to the callee's lexical. The QAST stage then read the inline info attribute, which only a Routine carries, so calling a constant bound to a pointy block died with a missing attribute error. This gates the analysis on Routine, so a block callee compiles as an ordinary call.
01:26
04:52 kjp left 04:56 kjp joined 12:16 Pixi` joined 12:19 Pixi left
Geth rakudo/lizmat-35: 1bb6f23e75 | (Elizabeth Mattijsen)++ | src/core.c/Process.rakumod
Simplify initialization of $*REPO

Did not find a reason why CALLERS:: should be inspected. Commit 081e4ba0666a7900b44276f2 could not tell me either.
So Remove that and prepare the code to get rid of the legacy backend handling.
14:12
rakudo: lizmat++ created pull request #6619:
Simplify initialization of $*REPO
14:30 timo left 15:20 timo joined
Geth nqp: ugexe++ created pull request #863:
Give a block reference the frame of its own compilation
20:15
rakudo: ugexe++ created pull request #6620:
RakuAST: bind a BEGIN-called routine to the running compilation
20:17
rakudo: ugexe++ created pull request #6621:
RakuAST: gate flatten candidates on the runtime binder flag
20:43
21:17 finanalyst joined 22:20 coleman left, summerisle left, ugexe left 22:24 coleman joined, summerisle joined, ugexe joined
Geth rakudo: ugexe++ created pull request #6622:
RakuAST: Compile the block without modifier via the without op
22:27
nqp/main: 64400a5c44 | (Nick Logan)++ (committed using GitHub Web editor) | 2 files
Give a block reference the frame of its own compilation (#863)

Previously a QAST::BVal took whatever frame the shared frame table held for the block's cuid. A compilation nested in another writes its frames into that table too, so a reference compiled ahead of the block in the outer unit resolved to the nested compilation's frame, whose index names an unrelated frame of the outer unit. A reference ahead of a ... (7 more lines)
22:28
rakudo: ugexe++ created pull request #6623:
Bump NQP to give a block reference the frame of its own compilation
22:30
23:18 sivoais left 23:19 sivoais joined 23:42 finanalyst left
Geth rakudo/main: 5354259569 | (Nick Logan)++ (committed using GitHub Web editor) | src/Raku/ast/var-lowering.rakumod
RakuAST: gate flatten candidates on the runtime binder flag (#6621)

Previously the shape gate for a flatten candidate did not consider whether the body's code needs the runtime binder. That was safe only because every parameter shape that sets custom-args is also rejected by IMPL-FLATTEN-ARG-PARAMETER. This adds the flag to the shape gate so such a body takes the general walk instead, where the existing rule already poisons any custom-args frame, rather than depending on that overlap.
23:50