Welcome to the main channel on the development of MoarVM, a virtual machine for NQP and Rakudo (moarvm.org). This channel is being logged for historical purposes.
Set by lizmat on 24 May 2021.
Nicholas Bah humbug, #moarvm :-) 09:38
nine Ah, I know what's wrong with the Curlie example 10:34
For RW int args I have spesh emit appropriate assign_i instructions after the native call: github.com/MoarVM/MoarVM/blob/mast...sp.c#L1249 10:43
What causes this to break is that the register holding the variable we try to assign to is already released at this point. So in the Curlie example the rw variable register is also used for the native functions return value. 10:44
MasterDuke nine: then that must be an additional bug, because the shorter golf still sigaborts even with spesh disabled 11:11
nine one bug at a time... 11:14
Looks like I have a fix 11:18
MasterDuke nine++'s christmas present to the raku community 11:33
Geth MoarVM/fix_rw_native_calls: 364d0a378b | (Stefan Seifert)++ | src/spesh/disp.c
Fix speshed native calls with rw args assigning into wrong target

For rw args of native calls we emit appropriate assign_i instructions after the call to write back the native values from the argument registers to the NativeRef. However the register holding the NativeRef will already have been released at the callsite (as for non-speshed calls writing back is handled by MVM_nativecall_dispatch). If one of the ... (9 more lines)
MoarVM: niner++ created pull request #1627:
Fix speshed native calls with rw args assigning into wrong target		 11:34
lizmat given this code: sub prefix:<√>($value) is looser(&[+]) { sqrt $value } 13:09
does it makes sense that the "is looser" trait will cause the prefix:<√> to become uninlineable ? 13:10
target has a :noinline instruction - ins: takeclosure
without the "is looser" trait, prefix:<√> inlines almost immediately 13:11
japhb That seems like a very odd entanglement to me .... 19:47
timo i won't say anything without a bytecode and/or spesh dump of the affected frames :) 21:56
japhb Well, it can still be very odd while still matching the code as written ... 22:10
timo certainly 22:19

Powered by Cro and the Raku® Programming Language.

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

Thank you!