data-default-class -> data-default

[9999years]

data-default 0.8 deprecates data-default-class by moving the Default class from Data.Default.Class to Data.Default.

See: kazu-yamamoto/crypton-certificate#11
See: haskell-tls/hs-tls#486
See: commercialhaskell/stackage#7545

[snoyberg]

Thanks

[Bodigrim]

Something is off, but I cannot figure out whose fault it is (crypton-connection?):

http-client-tls-0.3.6.4$ cabal build -c 'data-default>=0.8'
Resolving dependencies...
Build profile: -w ghc-9.6.6 -O1
In order, the following will be built (use -v for more details):
 - http-client-tls-0.3.6.4 (lib) (configuration changed)
Configuring library for http-client-tls-0.3.6.4...
Preprocessing library for http-client-tls-0.3.6.4...
Building library for http-client-tls-0.3.6.4...
[1 of 1] Compiling Network.HTTP.Client.TLS [Data.Default package changed]

Network/HTTP/Client/TLS.hs:118:40: error: [GHC-39999]
    • No instance for ‘Default NC.TLSSettings’
        arising from a use of ‘def’
    • In the first argument of ‘mkManagerSettings’, namely ‘def’
      In the expression: mkManagerSettings def Nothing
      In an equation for ‘tlsManagerSettings’:
          tlsManagerSettings = mkManagerSettings def Nothing
    |
118 | tlsManagerSettings = mkManagerSettings def Nothing
    |                                        ^^^

Network/HTTP/Client/TLS.hs:192:100: error: [GHC-39999]
    • Could not deduce ‘Default NC.TLSSettings’
        arising from a use of ‘def’
      from the context: MonadIO m
        bound by the type signature for:
                   newTlsManager :: forall (m :: * -> *). MonadIO m => m Manager
        at Network/HTTP/Client/TLS.hs:186:1-39
    • In the third argument of ‘mkManagerSettingsContext'’, namely
        ‘def’
      In the expression:
        mkManagerSettingsContext'
          defaultManagerSettings (Just globalConnectionContext) def
          msocksHTTP msocksHTTPS
      In an equation for ‘settings’:
          settings
            = mkManagerSettingsContext'
                defaultManagerSettings (Just globalConnectionContext) def
                msocksHTTP msocksHTTPS
    |
192 |         settings = mkManagerSettingsContext' defaultManagerSettings (Just globalConnectionContext) def msocksHTTP msocksHTTPS
    |                                                                                                    ^^^

Network/HTTP/Client/TLS.hs:208:81: error: [GHC-39999]
    • Could not deduce ‘Default NC.TLSSettings’
        arising from a use of ‘def’
      from the context: MonadIO m
        bound by the type signature for:
                   newTlsManagerWith :: forall (m :: * -> *).
                                        MonadIO m =>
                                        ManagerSettings -> m Manager
        at Network/HTTP/Client/TLS.hs:202:1-62
    • In the third argument of ‘mkManagerSettingsContext'’, namely
        ‘def’
      In the expression:
        mkManagerSettingsContext'
          set (Just globalConnectionContext) def msocksHTTP msocksHTTPS
      In an equation for ‘settings’:
          settings
            = mkManagerSettingsContext'
                set (Just globalConnectionContext) def msocksHTTP msocksHTTPS
    |
208 |         settings = mkManagerSettingsContext' set (Just globalConnectionContext) def msocksHTTP msocksHTTPS
    |                                                                                 ^^^
Error: [Cabal-7125]
Failed to build http-client-tls-0.3.6.4.

[juhp]

Yes, also reported in #547

[juhp]

Also note that data-default-class 0.2.0 was since released for compatibility.
Not sure if it helps though - I think it just re-exports from data-default iiuc.

[juhp]

Dunno if it is possible to do a revision for this?

[snoyberg]

I won't be spending more time on this issue. it's not an ecosystem issue I'm willing to spend my time debugging.

[9999years]

@Bodigrim Unfortunately these PRs are all pretty interdependent -- e.g. kazu-yamamoto/crypton-certificate#11 and kazu-yamamoto/crypton-connection#6 are also needed to get the data-default changes fixed for this group of packages.

[Bodigrim]

I think the solution is to revise the latest release of http-client-tls with data-default < 0.8 (I can do it myself as a Hackage trustee). When related PRs are merged, we can revise it once again with data-default >= 0.8, crypton-certificate >= X.Y.Z, crypton-connection >= U.V.W.

[hamishmack]

@Bodigrim it looks like the issue we have been seeing is that hoogle has a constraint in hackage that prevents it from using the new crypton-connection package.

See ndmitchell/hoogle#435

[Bodigrim]

@hamishmack you can build hoogle with cabal build -c 'data-default-class>=0.2'. Cabal does not know that it has to avoid build plans using both data-default-class < 0.2 and data-default >= 0.8.

It's tempting to revise crypton-x509-validation (which is at the root of all these build plans) with data-default-class >= 0.2. This will be an annoyance for Stackage though, because it would not be able to pull this revision without updating LTS...

[Bodigrim]

As of now cabal install hoogle just works, so hopefully the migration to the new data-default is complete.