[Nix-dev] I broke the buildfarm again - how to fix for the future?

[Marc Weber]

context:
        buildfarm log:
                getting attributes of path `/tmp/nix-build-vzjf88ha7knkyrcgd5p76ip20sjjnz4w-source-tarball.drv-0/wbcwpmccxjawv2a9fvj1giavxqnvfmmg-svn-export/pkgs/misc/bleeding-edge-fetch-info/pg_hsql.nix': No such file or directory

        @niksnut (SVN log)
        * Comment out ghc68extraLibs (it breaks the Nixpkgs release due to
          missing files and I don't understand the code...)

The reason why I broke the buildfarm is that there was no .nix file in
misc/bleeding-edge-fetch-info/<name>.nix

I've tried rewriting the code of bleeding-edge-repos this way:

  description:
    a ) local tar.gz - not found ? then
    b ) source .nix file containing
            fetchurl { url=<path to my server offering latest .tar.gz versions> .. }
    removed c ) use fetch(svn/darcs/...) removed (doesn't make sense because of hash)

  code:
    sourceByName = name : (
      let localTarGZ = managedRepoDir+"/dist/${name}.tar.gz"; 
          fetchUrlInfoFile = ../../../misc/bleeding-edge-fetch-info + "/${name}.nix"; in
      if pathExists localTarGZ
      then localTarGZ
      else if pathExists fetchUrlInfoFile
           then (import fetchUrlInfoFile) { inherit fetchurl; }
          # if this happens (I) Marc Weber (marco-oweber at gmx.de)
          # have propably forgotten to commit the .nix file containing the fetchurl derivation
          # contact me to fix it and /or install nixRepositoryManager and get a
          # local copy by nix-repository-manager --update name
          # using throw to never make the builfarm fail because of this again
           else throw "no bleeding edge source info (${fetchUrlInfoFile}) found for ${name}");

  trouble:
    I need ../../../misc to get an absolute path (because the function can be called anywhere..)
    however ../../../misc + name.nix fails if it doesn't exist, so my
    throw (which should make the buildfarm skip this expression) is not
    reached

>>> So can the checking of path existance be made lazy ?
    I think this kind of error might happen to me again.
    So using throw seems to be the best solution to me.

Eelco: I'd be glad to introduce you to my code. This part can be
understood in about 2min I think. Just ping me if you're interested.

Sorry for all the inconvinience
  Marc Weber