Am Sat, 09 Jul 2016 11:00:41 -0400 schrieb "Edward Z. Yang" <ezyang@cs.stanford.edu>: > OK, well, you can still get the ABI of a package in all cases > (it's just not in the ID anymore.) Something like: > > ezyang@sabre:~$ ghc-pkg field base-4.9.0.0 abi > abi: 5e731f0a06895b09ada57a6853a117b3 I pushed new commits to the short-ids branch, which should use this mechanism and not depend on parsing the ids at all. I didn't test these changes as much as I had liked, but I won't to it today and I think I won't be able to do it over the weekend, as I will be travelling back from DebConf during that time, so if someone else wants to do it, feel free. If not, I will do it when I'm home again. The branch also contains a split of haskell-devscripts into haskell-devscripts and haskell-devscripts-nocolour, which makes bootstrapping new architectures easier. Packages which are build-dependencies of hscolour can now depend on haskell-devscripts-nocolour without introducing a dependency cycle. This means ghc can use haskell-devscripts instead of duplicating its functionality and hscolour itself can be changed to depend on -nocolour, so it can be easier bootstrapped. Regards Sven > > Excerpts from Joachim Breitner's message of 2016-07-09 09:35:46 -0400: > > Hi, > > > > Am Freitag, den 08.07.2016, 17:06 -0400 schrieb Edward Z. Yang: > > > The "official" reason they don't have hashes is because, > > > as bundled libraries, they have special status and are thus > > > deified to not have hashes. (If these IDs are opaque it > > > shouldn't matter, right?!) > > > > well, they might be opaque, but it would be really good if they > > would differ if the ABI differs. So if we change the source code > > (bugfix, different flags etc.), we really really want to see a > > different hash, in order to detect what other packages need to be > > rebuilt. Basically our whole infrastructure relies on these hashes > > to identify ABIs. > > > > Greetings, > > Joachim > > >
Attachment:
pgpz0dNvnyPQu.pgp
Description: Digitale Signatur von OpenPGP