github.com/moarvm/moarvm | IRC logs at colabti.org/irclogger/irclogger_logs/moarvm
Set by AlexDaniel on 12 June 2018.
samcv AlexDaniel, wikipedia updated 08:23
nine Our Makefile copies src/gen/config.h to /usr/include/moar in addition to /usr/include/moar/gen. Is this actually useful? moar.h #includes "gen/config.h" and that's all an embedder would include. 08:34
Guest38485 nwc10: I actually had my RPi 4 standing vertically. It did help a little. Anyway I have since bought a flirc.tv/more/raspberry-pi-4-case 12:07
brrt \o 12:37
nine Damn it, I screwed things up. 12:59
The whole support for serializing NativeCall sites that underlies the NativeCall precompilation improvements has a fatal flaw: it breaks packaging 13:00
That's because the path to a %?RESOURCES library will change between precompilation and runtime. But now the old path is baked into the precomp file :(
I fixed that issue in the HLL by having %?RESOURCES return not strings but objects that resolve paths relative to the current path of the repository and NativeCall knows how to deal with those objects. But the same won't work for MoarVM 13:02
lizmat interesting NativeCall issue: stackoverflow.com/questions/591067...rent-order 14:58
nine lizmat: would be most helpful to know which exact rakudo version 15:05
lizmat will ask
Geth MoarVM: 8a9444b115 | (Stefan Seifert)++ | 3 files
A couple more minor signedness fixes
 16:07
MoarVM: ef7d0ecf4d | (Stefan Seifert)++ | 16 files
Fix numerous signedness inconsistencies in the JIT compiler

Try to stick to using unsigned numbers for counting things and for indexing into arrays. Fixes compiler warnings.
Geth MoarVM: a335add83b | (Stefan Seifert)++ | src/debug/debugserver.c
Fix potential NULL pointer dereference in debugserver

find_handle_target can return NULL		 16:28
nine There is a -Wnull-dereference since GCC 6 that can be very handy 16:29
jnthn nine++
nine Btw. there are only 2 warnings left. One in complicated string code that I'm not sure about and one in our imported memmem implementation 16:30
That's why I'm experimenting with additional warnings not even covered by -Wall and -Wextra 16:31
Kaiepi do builds work on windows? my pr is failing the appveyor build because of a syntax error in src/core/ops.c, even though i haven't changed it ci.appveyor.com/project/samcv/moar...le2c5tcado 16:34
nine src/spesh/optimize.c:2208:52: warning: potential null pointer dereference 16:36
That warning seems legit 16:37
jnthn nine: Will take a quick look 16:40
Hm, it thinks logged_ann could be null?
Hm, I think I agree 16:41
nine jnthn: yes and the author of the code thought so, too, as there's an if(logged_ann) above 16:42
jnthn Wonder who wrote this... :)
Maybe `else if (logged_ann)` is sensible 16:43
Though really this should probably never happen
Geth MoarVM: 14f581683d | (Stefan Seifert)++ | src/spesh/graph.c
Fix misplaced closing curly brace

Probably a merge error. It's harmless but certainly confusing and may cause issues when code is added at some point.		 16:45
nine That ^^^ one was was cought by -Wjump-misses-init 16:46
nine Oooh....we actually do not even enable -Wall! 16:59
lizmat nine: I seem to recall that jnthn didn't like that 17:00
jnthn The only thing I can remember not liking was the warning about not inserting extra parens in case folks don't know that & is tighter than | :P 17:01
nine We probably also don't want -Wunused-parameter as it flags too many places where the parameter is just part of the interface and not used by all implementations 17:03
But we can disable those separately
nine There are some warnings about undefined behaviour that are certainly worth a look 17:24
Geth MoarVM: 20804df5bb | (Stefan Seifert)++ | src/profiler/configuration.c
Remove return type of function that doesn't actually return anything.
 17:28
Geth MoarVM: f59f8f5de6 | (Stefan Seifert)++ | 2 files
Fix strict aliasing violations when accessing floating point number's memory

C does not allow for two different objects of different types to use the same memory location. While the compiler accepts casting the address of a local variable to a different pointer type and to dereference the result, this actually yields undefined behaviour. The compiler may e.g. reorder reads and writes, so the result of the operation may be some unrelated leftover from memory.
Fix by assigning to and reading from a union.		 20:08
Geth MoarVM: 13b618c25d | (Stefan Seifert)++ | 2 files
Fix strict aliasing violation in MVMHash_gc_mark

C does not allow for two different objects of different types to use the same memory location. While the compiler accepts casting the address of a local variable to a different pointer type and to dereference the result, this actually yields undefined behaviour. The compiler may e.g. reorder reads and writes, so the result of the operation may be some unrelated leftover from memory.
Fix by adding MVM_gc_worklist_add_object_no_include_gen2_nocheck for MVMObject*		 20:29
nine This is undefined behaviour: MVM_VECTOR_ENSURE_SPACE(compiler->spills, idx = compiler->spills_num++); 20:57
And frankly, I wouldn't really know what exactly it's supposed to do.
It does after all expand to: MVM_VECTOR_ENSURE_SIZE(compiler->spills, (compiler->spills_num) + (idx = compiler->spills_num++))
So should it end up as compiler->spills_num * 2 or as compiler->spills_num * 2 + 1? 20:58
Since this is about growing a vector, I guess just x2 21:00
Geth MoarVM: 8ac4ac2cf4 | (Stefan Seifert)++ | src/spesh/arg_guard.c
Fix warnings about unhandled enumeration values

Simply adding a default to the switch statement makes it clear that the other values don't need special treatment. Makes the compiler specific pragmas obsolete.		 21:09
MoarVM: de1bebc72a | (Stefan Seifert)++ | 6 files
Remove a bunch of variables that were written to but never read
MoarVM: d91f86ba81 | (Stefan Seifert)++ | src/jit/compile.c
Fix undefined behaviour in MVM_jit_spill_memory_select

The MVM_VECTOR_ENSURE_SPACE macro expanded to MVM_VECTOR_ENSURE_SIZE( compiler->spills, (compiler->spills_num) + (idx = compiler->spills_num++)) So it was unclear whether the size should be compiler->spills_num * 2 or compiler->spills_num * 2 + 1. I'm pretty sure the former was meant, so let's make that clear. Fixes a compiler warning.
MoarVM: 4ce5570e24 | (Stefan Seifert)++ | 2 files
Fix warnings about control reaching end of function without return value
 21:33
MoarVM: cac9b9a9d7 | (Stefan Seifert)++ | src/6model/sc.h
Remove useless assertion - unsigned value will always be >= 0
 21:39
MoarVM: 91e3530acf | (Stefan Seifert)++ | src/6model/sc.h
Fix signedness mismatch in comparisons in assertions

Powered by Cro and the Raku® Programming Language.

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

Thank you!