|
00:00
Sario left,
Sario joined
|
|||
| timo | with mi-malloc version 3.5 we can, as the author also suggested, use the free function where you can sometimes statically but at least at run time give it a hint whether the allocation you are freeing was a "big" one or a "small" one for what they claim is a good performance boost | 00:25 | |
|
01:42
disbot left,
disbot joined
06:06
rba left
06:07
rba joined
06:28
[Coke] left,
[Coke] joined
07:40
ShimmerFairy left,
ShimmerFairy joined
09:27
gfldex left,
gfldex joined
11:20
lizmat left,
lizmat_ joined
11:30
lizmat_ left,
lizmat joined
|
|||
| lizmat | And yet another Rakudo Weekly News hits the Net: rakudoweekly.blog/2026/08/25/2026-...r-by-coke/ | 11:31 | |
|
12:01
lizmat_ joined,
lizmat left
12:08
lizmat_ left,
lizmat joined
|
|||
| patrickb | Building moar on an ancient ubuntu (16.04, probably gcc 5.3.1) fails with paste.sr.ht/~patrickb/d00f1d426bab...a21480e47f | 14:01 | |
| [Coke] | GCC 5.3December 4, 2015 | 14:04 | |
| patrickb | I think it's gcc being pedantic about constant expressions. What's the crowds gut feeling? Should we try to adapt zmij or our build pipeline, or adapt the release pipeline? | 14:05 | |
| Yes, it's dead old. That's intentional. The distro I use to build the thing marks the lower bound where the build will run. | 14:06 | ||
| [Coke] | We have other cases where we keep a copy of the 3rd party with patches and use that as our immediate upstream, aye? | 14:08 | |
| patrickb | We do. But the more we do that the more work every change gets. So I hope to get around that... | 14:09 | |
| timo | can you try if replacing `static const uint64_t zeros = 0x0101010101010101u * '0';` with `#define zeros ((uint64_t)(0x0101010101010101u * '0')` makes it work? | 14:15 | |
| otherwise, it should also work to make it `0x3030303030303030u` because '0' is 0x30 | 14:18 | ||
| [Coke] | patrickb: get around it how? if we don't update the source and we must support that compiler... | 14:19 | |
| you mentioned pipeline - if we use the pipeline to make the source compilabale, that's just pushing the patch from a git copy to the pipeline. | |||
| patrickb | I more thought about adding some GCC option via an env var or some such... | 14:21 | |
| timo: I'll try | |||
| [Coke] | oh, ok. sure. | 14:23 | |
| patrickb | paste.sr.ht/~patrickb/d21e768a8ccf...572460d770 | 14:33 | |
| That's the new output with the zero substitution | |||
| fgcc 6.3.0 (debian 9) fails the same | 14:34 | ||
| *gcc | |||
| timo | quite possibly a C feature / syntax that this uses that's not available in those versions yet | ||
| patrickb | Yeah. That's my guess. | ||
| timo | oh | 14:35 | |
| I gave you a ) too few | |||
| in the define | |||
| patrickb | looks better | 14:36 | |
| the zero error is gone | |||
| paste.sr.ht/~patrickb/92c36bce5a47...0dd8cbded2 | 14:37 | ||
| timo | I guess we'd have to replace all these consts with a define then | 14:39 | |
| patrickb | I'm already trying... | ||
| timo | div_100_sig, div_10k_sig, neg10k seem to be the only ones? | 14:40 | |
| patrickb | paste.sr.ht/~patrickb/6b22610ea1c2...8bd3b53d87 | 14:41 | |
| doesn't like my macros yet | |||
| timo | the first error is because there's a variable defined with the same name as the macro | 14:43 | |
| so in the place where the variable name was meant to be, the macro got expanded | |||
| patrickb | neg10k? | 14:44 | |
| timo | 3rdparty/zmij/zmij.c:1881:72: note: in expansion of macro 'neg10k' | ||
| const __m128i neg10k_v = _mm_load_si128((const __m128i*)&static_data.neg10k); | |||
| same here where neg10k is a field in a struct | |||
| ah, it's not a variable name it's the name of this struct field | 14:45 | ||
| patrickb | oh | ||
| arg! | |||
| Renaming the define makes it work | 14:46 | ||
| paste.sr.ht/~patrickb/cecdda9f7f5b...07366dd231 | 14:48 | ||
| That's the patch | |||
| lizmat | PRit MERGEoit PROFit | 14:51 | |
| patrickb | github.com/vitaut/zmij/pull/148 | 15:02 | |
| timo | cool | 15:04 | |
|
19:14
RakuIRCLogger left
21:06
tbrowder left,
tbrowder joined
|
|||
| Geth | MoarVM/coke/readme: 756843921b | (Will Coleda)++ | README.md Remove reference to other backends. |
21:23 | |
| MoarVM: coke++ created pull request #2036: Remove reference to other backends. |
21:24 | ||