Probably not many people care about i686-linux any more, but building
all these images is fairly expensive (e.g. in the worst case, every
Nixpkgs commit would trigger a few gigabytes of uploads to S3).
Previously this was done in three derivations (one to build the raw
disk image, one to convert to OVA, one to add a hydra-build-products
file). Now it's done in one step to reduce the amount of copying
to/from S3. In particular, not uploading the raw disk image prevents
us from hitting hydra-queue-runner's size limit of 2 GiB.
This update was generated by hackage2nix v20160308 using the following inputs:
- Nixpkgs: 47c2c01ea7
- Hackage: 435f8156b7
- LTS Haskell: 6bcb171af7
- Stackage Nightly: a161c91389
Also, update the list of broken builds in configuration-hackage2nix.yaml.
SQLAlchemy-0.9.9 was broken recently, probably because of an sqlite
update. So we just use the latest version.
The build expression used to take the entire package set as an argument,
which is poor style because it hides the true dependencies from the
interface. I've change the code to expect the appropriate individual
packages instead.
It seems Activemq was "split" in to 5.8.0 and 5.12.1 only to keep from
trampling someone on 5.8.0. This is to the point of not upgrading the
default, but making a new activemq attribute specific for 512.
As far as I can tell, there is no good reason to stay providing 5.8.0,
so this is the proposal to delete it.
Note: There are over 1,000 fixes between 5.9.0 to 5.13.2.
The idea is to bundle ruby, bundler and bundix together. I was
having issues where bundler was installed with ruby 2.3.0 and I wanted to use
ruby 2.0.0.
With this change all the developer has to do is install `ruby_2_0_0.dev`
either in his environment or in a nix-shell.
Having a separate rubygems package can lead to split-brain scenarios.
Since rubygems is designed to replace himself on a ruby installation,
let's do that.
Just call `fetchurl` on the ‘official’ download link.
`fetchFromGitHub` was introduced in 54916ba, and stayed around when
we moved back to a regular release. Which worked fine, but required
keeping `autoreconfHook` and some `postPatch` hackery which can now
both be dropped.
Thanks to @kragniz for drawing my attention to this.
