[Nix-dev] Toward a Purer stdenv

Marc Weber marco-oweber at gmx.de
Wed Aug 17 22:06:02 CEST 2011


Excerpts from Shea Levy's message of Wed Aug 17 21:37:32 +0200 2011:
> I've been thinking about creating a nixpkgs branch to find and eliminate 
> some of the remaining impurities in the Linux stdenv.

"linux stdenv" ? What are you referring to exactly?

Impurities:

- cosmic ray (they may change bits), so build results do not only depend
  on your computer, but also where it is located

- amount of RAM

- filesystem being used (btrfs caused different behaviour in previous
  older kernels)

- kernel (you talked about it)
   * version
   * settings (over commit or such)
   * ..

- hardware:
 * disk (if you think about database apps they might choose to use on
   ore the other implementation depending on how fast your harddisk is
   don't think this happens in practise)

 * CPU (features ?, eg is MMX available or not)

 * graphic card 

 * .. you can think about more issues such as USB being plugged in (does
   not make sense, but you could write such a prog which behaves
   differently if some kind of printer is plugged in)

- time / timings
  * clock time
  * speed of CPU
  * ...

- number of multiple cores when using parallel builds (eg kernel
  derivation)

- running virtualized or not. Eg if you're running in a virtualized env
  you can't run qemu tests (eg nixos installer) because you can't nest
  vitrualized systems due to hardware restrictions (correct?)..

- /dev/random (thus your keyboard input)
  This includes builds using rand() function etc.
  The fix is to fix builds if this happens.
  Older ghc versions may have had this issues according to a commit log.

- network access

- of course the user name building a package

- the store location (for completeness)

The impurity having had greatest impact on me is kernel version when
trying 3.0. In fact newest kernel made btrfs worse so my fix was
downgrade to 2.6.38.8. You don't know when the next big kernel numbering
scheme is going to happen so its hard to say big the impact on us is at
all.

I'm pretty sure that I've still missed some impurities.

Marc Weber



More information about the nix-dev mailing list