Bump obelisk for GHC 8.6

[3noch]

I merged obelisk/develop into obelisk/cg-fullroute and then bumped the thunk here.

[ali-abrar]

Not so easy to review this - I'll come back after checking out cg-full-route on obelisk independently. Perhaps rhyolite can go back to develop of ob.

[3noch]

Yeah. Note that rhyolite was already on cg-full-route.

[3noch]

I've successfully run the obelisk skeleton with this change.

[3noch]

I have a skeleton app using rhyolite-backend and rhyolite-frontend and everything builds. If an app bumps to this branch and does some testing that would help confirm all these changes are working.

[3noch]

@ali-abrar This is ready for some review.

[ali-abrar]

Could you also update the changelog? Apart from the bump, it might be worth mentioning the improved error messaging.

[3noch]

"Improved error messaging" is one way to say it... Perhaps I should have used fail from MonadFail but that would require pushing MonadFail m constraints everywhere.... also I'm fairly confident that error is equivalent to fail in most of the monads we are using anyway...

[3noch]

Ideally we'd get these two branches merged first:

• Update dependencies for these 1 and base 4.12 vessel#5
• Fixes for new dependencies groundhog#9

I find it annoying to depend on branches of dependencies because it makes bumping much less obvious to the next guy.

[Ericson2314]

Someday it might be nice to use MonadError for those, but at that point we'd want to get postgresql-simple exceptions and stick them in the same error type. Or in the meantime there's some postgresql-simple exception type that can be reused (as something thrown not MonadError) for this.

CC @lpsmith

[benkolera]

Just in case you didn't know, @ali-abrar (just delete this comment if I'm speaking the obvious! 🙂) , that latest bump of vessel you did here brought in obsidiansystems/vessel@fa472dc which seems to be a little tight on it's version bounds. The build for this MR is currently broken:

Configuring vessel-0.1.0.0...
Setup: Encountered missing dependencies:
QuickCheck ==2.13.*, aeson >=1.4.4 && <1.5, these >=1.0.1 && <1.1

Are there actual bugs / new features that vessel is depending on in QuickCheck > 2.12.6.1 , Aeson > 1.4.2.0 or these > 1? Fwiw, I relaxed these bounds on my project with a bunch of unpacked thunks and it seemed to build.

Here is where everything is at and what I changed to get it building.

bkolera at bkolera in ~/src/github/benkolera/cogitate/.obelisk/impl (cg-fullroute●) 
$ git rev-parse HEAD
64f5c928a5e6c39dcc450005e5cffa6d11968a8b

bkolera at bkolera in ~/src/github/benkolera/cogitate/.obelisk/impl/dep/reflex-platform (aa-split-these●) 
$ git rev-parse HEAD
4b4a2c8679d6f3042180d2974b31c43bfc506b91

bkolera at bkolera in ~/src/github/benkolera/cogitate/.obelisk/impl/dep/reflex-platform (aa-split-these●) 
$ git diff | tee 
diff --git a/haskell-overlays/untriaged.nix b/haskell-overlays/untriaged.nix
index eaf987c..b1a3e73 100644
--- a/haskell-overlays/untriaged.nix
+++ b/haskell-overlays/untriaged.nix
@@ -82,7 +82,7 @@ in self: super: {
 
   # Packages not yet in 19.03
   semialign = doJailbreak (self.callHackage "semialign" "1" {});
-  these = self.callHackage "these" "1" {};
+  these = self.callHackage "these" "1.0.1" {};
 
   ########################################################################
   # Packages not in hackage
diff --git a/nixpkgs-overlays/all-cabal-hashes/default.nix b/nixpkgs-overlays/all-cabal-hashes/default.nix
index 1a6686a..dd3d80c 100644
--- a/nixpkgs-overlays/all-cabal-hashes/default.nix
+++ b/nixpkgs-overlays/all-cabal-hashes/default.nix
@@ -1,8 +1,8 @@
 self: _: {
 
   all-cabal-hashes = self.fetchurl {
-    url = "https://github.com/commercialhaskell/all-cabal-hashes/archive/5e782dacf8d2d603f3848cfa37561acba313fbd5.tar.gz";
-    sha256 = "0cms0z38fcmrzplhzl34d3cm7hs8kr4z8p5cnxvhhjm7mf7357z7";
+    url = "https://github.com/commercialhaskell/all-cabal-hashes/archive/17df7a8cee55da9518109cc08f3cbecd3afc2062.tar.gz";
+    sha256 = "0s2g057h7661w4aaaakg8hfmgwaz7ayq8phfnjv9rpjyfndmr3sd";
   };
 
 }

bkolera at bkolera in ~/src/github/benkolera/cogitate/dep/rhyolite (eac@bump-obelisk●●) 
$ git rev-parse HEAD
d162e25ba76aa279f604724c43d680c9646f4e70

bkolera at bkolera in ~/src/github/benkolera/cogitate/dep/rhyolite/dep/vessel (develop●) 
$ git rev-parse HEAD
8469dd1ffbe0e6640c0d1c1f417a49ae8d10b06e

bkolera at bkolera in ~/src/github/benkolera/cogitate/dep/rhyolite/dep/vessel (develop●) 
$ git diff | tee                        
diff --git a/vessel.cabal b/vessel.cabal
index 79866a5..45e5b1f 100644
--- a/vessel.cabal
+++ b/vessel.cabal
@@ -20,7 +20,7 @@ library
   exposed-modules:     Data.Vessel
   other-extensions:    FlexibleInstances, GeneralizedNewtypeDeriving, RankNTypes, ScopedTypeVariables, StandaloneDeriving, TypeFamilies, UndecidableInstances
   build-depends:       base >=4.9 && <4.13
-                     , aeson >= 1.4.4 && < 1.5
+                     , aeson >= 1.4.0 && < 1.5
                      , bifunctors >= 5.5 && < 5.6
                      , containers >= 0.6 && < 0.7
                      , constraints >= 0.10 && < 0.11
@@ -34,7 +34,7 @@ library
                      , reflex >= 0.6.2.4 && < 0.7
                      , semialign >= 1
                      , these >= 1.0.1 && < 1.1
-                     , QuickCheck >= 2.13 && < 2.14
+                     , QuickCheck >= 2.12 && < 2.14
                      , witherable >= 0.2 && <= 0.3.2
   hs-source-dirs:      src
   default-language:    Haskell2010

[ali-abrar]

@benkolera Those version bounds were generated by cabal. They'll need to be extended.

[ali-abrar]

Alright, vessel is good to go now. There's still the groundhog fork to update.

[3noch]

In general I could convert some of these to callHackageDirect. I'm not familiar with how it deals with hackage revisions, though?

[benkolera]

@3noch callHackageDirect just gets the tarball from hackage [1], which is as per the original sdist upload and does not contain metadata revisions[2].

[1] https://github.com/NixOS/nixpkgs/blob/master/pkgs/development/haskell-modules/make-package-set.nix#L192
[2] https://github.com/haskell-infra/hackage-trustees/blob/master/revisions-information.md#what-are-revisions

[3noch]

@benkolera Great. I didn't realize the hackage tarball didn't reflect revisions. That makes me more comfortable using callHackageDirect.

[3noch]

@ali-abrar I've moved the dependencies you mentioned to use callHackageDirect instead of being "thunked".

[3noch]

Actually, if we're going to prefer callHackageDirect over thunks... I think a bunch of these dependencies can be switched over. Perhaps we should have a clear stance as to when we prefer one over the other? Currently I see:

callHackageDirect:

• Uses Hackage
• Shows version number clearly
• Avoids extra files in repo

Thunks:

• Makes it much easier to bump because you don't have to manually update the sha256.
• (Probably could show version number clearly if we modified obelisk thunks to support tags and not just branches)

[ali-abrar]

Please update the changelog.

[3noch]

@ali-abrar Done.