Conversation

Notices

1. Embed this notice
   Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Wednesday, 15-Jan-2025 22:57:11 JST Haelwenn /элвэн/ :triskell:

   open(O_RDONLY | O_TRUNC)[…]
   What these flags specified together do is “undefined” in POSIX, but the Linux manpages ominously state “On many systems the file is actually truncated”, and indeed on Linux, using this open mode actually truncates files most of the time!

   — Haiku Activity & Contract Report, December 2024 (ft. Iceweasel)

   Holy shit, that sounds really bad.

   • Embed this notice
     :umu: :umu: (a1ba@suya.place)'s status on Wednesday, 15-Jan-2025 23:03:02 JST :umu: :umu:

     in reply to
     @lanodan don't rely on undefined behavior, don't get undefined result.
     Haelwenn /элвэн/ :triskell: likes this.
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Wednesday, 15-Jan-2025 23:04:47 JST Haelwenn /элвэн/ :triskell:

     in reply to

     • :umu: :umu:
     @a1ba Yeah, I'd rather have it throw an error, which is what Haiku now does.
   • Embed this notice
     :umu: :umu: (a1ba@suya.place)'s status on Wednesday, 15-Jan-2025 23:04:48 JST :umu: :umu:

     in reply to

     • :umu: :umu:
     @lanodan though truncating file that's opened in readonly mode sounds evil.
     
     Even if it's undefined.
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Wednesday, 15-Jan-2025 23:08:30 JST Haelwenn /элвэн/ :triskell:

     in reply to

     • :umu: :umu:
     @a1ba And I guess if you want to be a bit spicy then an `abort()` could be used as it's a programmer error.
     Truncating with read-only mode… yikes.
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Wednesday, 15-Jan-2025 23:21:13 JST Haelwenn /элвэн/ :triskell:

     in reply to

     • :umu: :umu:
     @a1ba Well… it's one thing to write a bug, it's another to document it into the manpages.
     
     And as Linux isn't MacOS/illumos/OpenBSD, the libc isn't the only gateway into syscalls like open(2).
   • Embed this notice
     :umu: :umu: (a1ba@suya.place)'s status on Wednesday, 15-Jan-2025 23:21:14 JST :umu: :umu:

     in reply to
     @lanodan I wonder if it's just an oversight.
     
     I mean, imagine a VFS developer, who first check O_TRUNC flag, truncate it, and only then set desired mode for the file.
     
     Though, it's not a big deal if kernel is buggy, because libc wrapper could handle this and return error... or abort().
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Wednesday, 15-Jan-2025 23:33:39 JST Haelwenn /элвэн/ :triskell:

     in reply to

     • :umu: :umu:
     @a1ba Well yeah that section of POSIX just defines C interfaces, just happens that unix syscalls (which POSIX doesn't cares about) tends to exactly match some selected C interfaces.
     
     And just seems a lot better if not just correct to me to add a guard in the kernel, which is the one doing the truncation, rather than libc which just wraps and isn't the only wrapper.
   • Embed this notice
     :umu: :umu: (a1ba@suya.place)'s status on Wednesday, 15-Jan-2025 23:33:41 JST :umu: :umu:

     in reply to
     @lanodan it obvious that you can do raw syscalls but also in my opinion it's the libc who shall provide POSIX compatible interface rather than the kernel. Though it's only my opinion.