Conversation

Notices

1. Embed this notice
   Sergey Bugaev (bugaevc@floss.social)'s status on Friday, 14-Jul-2023 23:50:30 JST Sergey Bugaev

   #Meson needs to learn that Linux is not an operating system unto itself.

   No really, like 95% checks for host_machine.system() == 'linux' are wrong, unless you really care about *kernel* specifics like Linux kernel modules. Typically what you want to check is:
   - is my compiler GCC / Clang? (use the Meson compiler object for that!)
   - is my libc (a recent enough) glibc?
   - does this system use a "freedesktop" GUI stack?
   - etc etc

   • Haelwenn /элвэн/ :triskell: likes this.
   • Embed this notice
     Sergey Bugaev (bugaevc@floss.social)'s status on Friday, 14-Jul-2023 23:51:23 JST Sergey Bugaev

     in reply to

     • Abbie Normal

     @abbienormal but then they return 'android' when Android is detected, even though it too is running on the Linux kernel! You'd then think 'linux' means GNU/Linux then, but no, it still returns 'linux' on non-GNU non-Android Linux systems too.

     Haelwenn /элвэн/ :triskell: likes this.
   • Embed this notice
     Abbie Normal (abbienormal@floss.social)'s status on Friday, 14-Jul-2023 23:51:24 JST Abbie Normal

     in reply to

     @bugaevc

     I suppose this is the confusion about the fact that "Linux" is not a platform

     It's a kernel

     It seems we haven't got a shared definition of what "Linux" encompasses

     Oh well...

   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Friday, 14-Jul-2023 23:52:45 JST Haelwenn /элвэн/ :triskell:

     in reply to

     • Abbie Normal
     @bugaevc @abbienormal Reminds me of the non-Apple Unix stuff in other buildsystems (where it typically makes a bit more sense).
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Friday, 14-Jul-2023 23:55:32 JST Haelwenn /элвэн/ :triskell:

     in reply to
     @bugaevc More like meson users, nah?
     Which is the rather typical issue of most developers being awful at system integration.
   • Embed this notice
     翠星石 (suiseiseki@freesoftwareextremist.com)'s status on Saturday, 15-Jul-2023 00:05:35 JST 翠星石

     in reply to
     @bugaevc It's not too hard to patch the uname SYSCALL to return the correct OS name (GNU/Linux) and it's very nice to have applications display the correct OS name.
     
     It's trivial to patch uname(1) to handle the uname SYSCALL returning the correct string as well, which means that build systems that use that work fine, the only issue then is build systems that rely on incorrect uname(2) behavior.
     
     
     It seems the only easy fix to the problem is to switch to GNU/Hurd.
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Saturday, 15-Jul-2023 02:15:29 JST Haelwenn /элвэн/ :triskell:

     in reply to

     • Abbie Normal

     @bugaevc @abbienormal Like CMake scripts with if(APPLE AND NOT UNIX) or C preprocessors macros with !defined(__APPLE__) && defined(__unix__)

   • Embed this notice
     Sergey Bugaev (bugaevc@floss.social)'s status on Saturday, 15-Jul-2023 02:15:31 JST Sergey Bugaev

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Abbie Normal

     @lanodan @abbienormal hm, what do you mean, which non-Apple Unix stuff?

   • Embed this notice
     Sergey Bugaev (bugaevc@floss.social)'s status on Saturday, 15-Jul-2023 02:19:06 JST Sergey Bugaev

     in reply to

     • Haelwenn /элвэн/ :triskell:

     @lanodan Meson users too should be aware of this and use it properly, but Meson shouldn't have made it that easy to misuse this.

     Like ugh, why didn't they make machine.kernel() and machine.userland() (would return "linux"/"gnu", "linux"/"android", "freebsd"/"freebsd", "freebsd"/"gnu", "xnu"/"darwin")

     Haelwenn /элвэн/ :triskell: likes this.
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Saturday, 15-Jul-2023 02:23:56 JST Haelwenn /элвэн/ :triskell:

     in reply to
     @bugaevc Userland wouldn't really be the right one, at least coming from unixes with a mix of base system + GNU software on the side for compatibility.
     
     I think they should have exposed something like the target triple and dedicated functions for each part of it, plus matching on the compiler/linker/… being used (you can mix-match there).
   • Embed this notice
     Sergey Bugaev (bugaevc@floss.social)'s status on Saturday, 15-Jul-2023 02:24:51 JST Sergey Bugaev

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Abbie Normal

     @lanodan @abbienormal if (UNIX AND NOT APPLE) you mean, but yeah, I see.

     I would argue that that is an OK pattern if they really want to single out Darwin (maybe for Mach-O or something), but not quite right if what they actually care about is freedesktop-ish vs Cocoa-ish. You can certainly have GNUstep on non-Apple systems, and run a regular X11 desktop on Darwin.

     Haelwenn /элвэн/ :triskell: likes this.
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Saturday, 15-Jul-2023 02:38:50 JST Haelwenn /элвэн/ :triskell:

     in reply to
     @bugaevc That's because you cannot assume anything about the OS installation, the userland can be a complete patchwork, in fact I'd say it always is.
     
     You should always do clean atomic detections based on exactly what you want to use.
     Just like how Unix doesn't means X11 or that the entirety of freedesktop stuff is present.
     
     Heck, Android could gain wayland support at some point.
   • Embed this notice
     Sergey Bugaev (bugaevc@floss.social)'s status on Saturday, 15-Jul-2023 02:38:51 JST Sergey Bugaev

     in reply to

     • Haelwenn /элвэн/ :triskell:

     @lanodan yes, target triple + some sane (i.e. not just a regex) API to match against it would also be good. And yes "GNU" is problematic for that reason, and also I have no idea what to call the various non-GNU userlands seen on Linux in the wild (as seen in: Alpine, Chimera Linux, OpenWrt...) — just calling each one by its own name is not going to be any useful.

     Maybe we should file an issue and write a proposal over at Meson's issue tracker.