On 02/24/2015 04:44 PM, Vladimír Čunát wrote:
On 02/24/2015 04:33 PM, Luca Bruno wrote:Afaik it's not only a python "disabled" problem. It also happens for "meta.broken" that doesn't get propagated.Yes, one great advantage of "throw" is how it propagates nicely. Is there a reason not to use that, except for errors in Hydra evals? We might just tweak/customize error-logging from the current "throw" instead of doing anything complex.
Oh, I've let you mislead me (and my memory just returned): meta.broken is implemented via throw run as an assertion in stdenv.mkDerivation evaluation.
See <nixpkgs>/pkgs/stdenv/generic/default.nix for the implementation.Therefore, it certainly should get propagated whenever anything from the derivation is used (even just name or meta stuff).
If we need another semantically distinct option, like meta.disabled, it will be easy to implement in the same way. However, for any throw-based solutions, Hydra just outputs the error thrown, so we that's a point where it should get more clever. Perhaps we could be having some tags at the start of the string, such as "Broken:" or "Disabled:", so Hydra or anyone could easily filter the throws and keep only the ones interesting in the particular use case.
Vladimir
smime.p7s
Description: S/MIME Cryptographic Signature
_______________________________________________ nix-dev mailing list nix-dev@lists.science.uu.nl http://lists.science.uu.nl/mailman/listinfo/nix-dev
