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.
nine coleman: why Meson btw? Why not e.g. cmake? 06:07
Nicholas I think he mentioned i in the PR, but my experience of cmake is that it's impossible to bootstrap it on less common OSes. It's an albatross 06:46
whereas Python is actually portable 06:47
if cmake isn't already installed or packaged, 50/50 you're stuffed
Woodi dropping Perl family for build system will be huge loss for the future... looks cmake with ninja is very fast but cmake part is a mess 06:55
and builds are not so complicated overall, even Makefile dependency graph is overkill, why not have something working hierarchical/declarative with caches ? partial rebuilds part is what need to be fast ? 06:58
also tool discovering what to rebuild can be augmented by telling it what to rebuild 07:00
nine I'm not a big fan of cmake either. Have never tried Meson though. Just curious really. 07:02
Woodi they say cmake is standard :)
and Bazel is Java - everyone have his own build system :)
maybe it is fast total rebuild what devs would like ? ccache is there for C/C++. also git have all info what files changed 07:11
Nicholas I think it was ccache that I couldn't build the current version of. Because it needed cmake. And cmake wouldn't bootstrap on whatever-it-was. So I used the previous version of ccache 07:12
Woodi ccache is older then cmake by decades...
Nicholas whatever-it-is, I'd be happy if Configure.pl and all our special snowflake could be replaced by something someone else maintains, that's no less portable than what we have
IIRC version 4 changed its build system 07:13
it was something like that. The previous version of ccache was perfectly good
Woodi "CMake 3.15 or newer."... 07:14
Nicholas "perfectly good for what I needed" is what I meant. I have no knowledge of how much of a rats nest the ccache build is
Woodi I have impression that ccache is very standard Unix tool that just blends into background and is invisible... and it is so old that corner cases are well documented: ccache.dev/manual/4.9.1.html#_caveats 07:29
alternative is github.com/mozilla/sccache 07:33
but actually would be nice to have list of wishes for build system first and then brainstorm solutions :) 07:35
0. remove system detection from build script... why not have script that detects is that CC working on HalloWorlds, etc and _do not run_ this script on every rebuild ? :) 07:38
Nicholas good news: I have improved my Wifi 07:40
bad news: It's still a long work video call meeting 07:41
Woodi 1. git submodules are meh and security violation sometimes but conceptually clean, complete and usefull... do separate repos would be nicer ? submodules are usually forks with patches... 08:17
2. do big projects have "expectations" ? eg. do they automatically know how to plug-in dependency using cmake/meson ? maybe that can be added on top of existing/improved build ? 08:23
xiaomiao cmake only exists because people insist on building stuff on windows 08:24
it's best treated as a leftover from the old times and not encouraged
Woodi xiaomiao: /me feels old now :) 08:25
xiaomiao Woodi: I am often surprised about the unironic use of such legacy systems
Woodi xiaomiao: it's what old people know :) I still didn't finished reading "UNIX administration" by AEleen Fresh ;) 08:27
but if somethink brings value it should be promoted, few times even... 08:28
xiaomiao Woodi: nah "old" people would use proper tools like ... autotools
or just a plain Makefile, because most of the time you don't care what cc is set to 08:29
Woodi yep, plain Makefiles pls ;) but how much ninja could replace it ? 08:30
xiaomiao no ;) 08:31
ninja is just half of the problem space, handwriting ninja files is unpleasant
so you need something like meson or a makefile to generate that
Woodi also how much of "build problems" are influenced by: parallel builds, kubernets like infra, TI/CI and other new toys ? what is a must have now ?
xiaomiao and what problem does it solve?
the modern thing is having a makefile that curlbashes into docker 08:32
Woodi but we have grammars ;)
xiaomiao that way everything is ubuntu 22 ;)\
Woodi xiaomiao: let's just filter some sudocurl|bash and have better future :) 08:33
xiaomiao you mean curl | sudo tee bash
Woodi can ninja files be templated for usual case ? how often that files change ? they describe whole project, right ? and are human-readable, simpler then Makefile's, "always parallel". I do not understand what else valuable outer builder like cmake adds... 08:42
xiaomiao cmake adds ... uh, lots of pain? :) 08:48
the ninja format is not designed for 'manual' use, I wouldn't recommend trying to use it directly 08:49
08:51 nine left, nine joined
Voldenet theoretically build systems separate deps detection from deps usage in the project, I have no idea how it practically is, because I usually hardcode paths 08:56
simply using /lib/somelib.so worked for me in so many cases that it never occured to me to complicate this 08:58
xiaomiao you shouldn't try to do that by path directly 08:59
it usually works by accident, some of the time
Voldenet It usually works a lot of times, but when it doesn't, I can fill correct paths myself 09:00
ymmv though, maybe I should use weirder unix distros
Woodi xiaomiao: but ninja is described as simpler then make and Makefiles are supoused to be write manually :) also it is designed to be "human-readable" which is also human-writable :) 09:01
xiaomiao Woodi: so is maven ;) 09:02
just because someone writes a sequence of words doesn't make them correct or useful
Voldenet: well ... you assume dynamic linking 09:03
Voldenet I'm fully aware that it's wrong
but I do adore the simplicity ;> 09:04
xiaomiao Woodi: ninja mostly solves the problem that "make is slow" and that starts becoming human-scale measurable on something the size of the linux kernel, maybe
Voldenet theoretically cmake isn't horrible either and it's described as "simplifies stuff" 09:05
xiaomiao yeah but it's fractally wrong
most checks are pattern matching, so they fail randomly when a new compiler outputs a line extra and such unexpected things 09:06
Woodi but ninja recomended usage looks unnecessary self-conflicting and looks like cargo culting
xiaomiao and debugging it is something I'd prefer not to do so often, it's exhausting
Voldenet tbh. I've had bad experience simply building cmake projects
otoh perl script is a breeze to debug 09:07
xiaomiao yeah perl is opinionated in just the right ways
cmake syntax alone is enough to frustrate me 09:08
Woodi Configure.pl was before automake/autoconf ? 09:10
Voldenet …have you ever tried building c projects with ant? Now this is a painful syntax 09:11
Woodi both are all-in-one tools for software distribution ?
xiaomiao Woodi: configure.pl avoids pulling in a whole "foreign" system like automake 09:12
because that means bootstrapping is now also the problem of somehow bootstrapping automake, which then requires random stuff ...
09:13 sena_kun joined
Voldenet I never thought configure.pl was supposed to replace autoconf, it was written for codegen iirc 09:13
Woodi but you need to "bootstrap" plain make/Makefile too and that is solved by C compiler presence. so it's not like toolchain bootstrapping 09:15
Voldenet nevermind, I checked again and it does exactly what autoconf 09:16
xiaomiao Woodi: for make files you need some flavour of make - and that roughly depends on a C compiler 09:18
Woodi Voldenet: I think auto* were mimicking Configure.pl becouse it was earlier. and they was meant for distribution and today we need better tools for rebuilding during development. but can be mistaken on this
> also how much of "build problems" are influenced by parallel builds, kubernets like infra, TI/CI and other new toys ? what is a must have now ? what you are missing in curent Raku/Perl development process ? 09:23
09:56 sena_kun left 16:18 vrurg_ joined 16:21 vrurg left 18:20 sena_kun joined, vrurg_ is now known as vrurg
coleman nine: regarding "why not cmake"... I am literally incapable of understanding cmake. :-/ 21:31
Meson generates out-of-tree ninja files, in another directory, conventionally called "builddir". I will continue to try to get things building as an experiment that lives side-by-side with the perl 21:32
The perl has never failed me, fwiw 21:33
I think having a more widely-used option will help us keep the perl up to date, too 21:46
when one of them breaks we have a reference to another
Geth MoarVM: AntonOks++ created pull request #1796:
build_release.yml: Add initial build workflow		 22:02
23:24 sena_kun left

Powered by Cro and the Raku® Programming Language.

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

Thank you!