Mirror of forum.nim-lang.org

9137 :: CLOEXEC on old Linux kernel.

• AlexKotik (orginal) [2022-05-03T12:26:57+02:00] view original

  Hello. I'm trying to make async sockets work on old embedded system with an old Linux kernel (2.4 brunch) and SOCK_CLOEXEC/O_CLOEXEC appeared in 2.6 brunch according to manuals. Nativesockets in Nim are using CLOEXEC internally when creating a socket. I'd like to use asyncnet for async sockets and not rewriting it myself using posix module. Is there any way I could make Nim's builtin async sockets work without CLOEXEC? Or rewriting it myself using the posix APIs is the only option?

---

• shirleyquirk (orginal) [2022-05-05T10:48:57+02:00] view original

  It looks like asyncdispatch checks to see if accept4 is declared, and only then uses SOCK_CLOEXEC.

  I'd try editing posix.nim here and adding another guard, like

   (defined(linux) and not defined(ancient_linux)) or defined(bsd):
  Run

  then pass in --define:ancient_linux when you compile

  Similar changes might need to be made in nativesockets.nim

---

• cblake (orginal) [2022-05-05T11:30:45+02:00] view original

  BTW, if you are hacking/replacing posix.nim then you may also want to know about the patchFile construct for myModule.nims.

---

• AlexKotik (orginal) [2022-05-10T09:00:36+02:00] view original

  After reading some Nim code from stdlib, I think I can get away by defining SOCK_CLOEXEC to 0. What do you think? I could pass -DSOCK_CLOEXEC=0 to a C compiler, but I was thinking if I could do it using Nim. I've created fixups.nim file with:

  {. emit("""

  #ifndef SOCK_CLOEXEC

  #define SOCK_CLOEXEC 0

  #endif """) .}

  And then added --include:fixups.nim to the generation command, but for some reason the emmitted code isn't added to the to top of C files generated. Maybe I'm doing something wrong here? I thought that --include parameter includes code to all modules, including stdlib, isn't that so?

---

• shirleyquirk (orginal) [2022-05-10T09:15:49+02:00] view original

  If whatever is opening the socket is setting, CLOEXEC it doesn't expect to have to close the socket (see here` so I don't think setting it to 0 will be enough

---