search

ndmitchell / hoogle

Haskell API search engine
http://hoogle.haskell.org/
Other
738 stars 134 forks source link

Add an upper bound on http-conduit. #235

Closed 23Skidoo closed 6 years ago

23Skidoo commented 6 years ago

Fixes #234. Would be nice if you could also make a Hackage revision.

hvr commented 6 years ago

Would be nice if you could also make a Hackage revision.

Make that about 50 revisions (but iirc only half of them are about http-conduit)...

See https://matrix.hackage.haskell.org/package/hoogle

untitled

23Skidoo commented 6 years ago

Right, though when people do cabal install hoogle, they usually want the latest one.

hvr commented 6 years ago

@23Skidoo ...and cabal install hoogle (or build-depends: hoogle >= 5.0 && < 5.1 for that matter) doesn't guarantee they will get that one if the solver has enough pressure to fallback to a previous one. That's why Trustees can't just fixup the latest version of something but have to fix everything that's accessible to the cabal solver... and I'm already on it... it just takes a bit more time to see what else I'll have to fix while at it... ;-)

ndmitchell commented 6 years ago

Thanks @23Skidoo and @hvr. I've added revisions for hoogle-5.0..5.16, but as @hvr points out, the failures with older Hoogle are a different issue (don't know what, if someone tells me and it impacts them, I'll happily revise another 25 odd versions).

I probably won't ever release this code, but will use it as a stepping stone to then upgrade to the latest http-conduit.

hvr commented 6 years ago

@ndmitchell just give matrix.hho a couple of minutes to catch up, and you'll see the compile failures there... :-)

ndmitchell commented 6 years ago

@hvr - I see it now, bound added to test one of them then will replicate. Although some of my packages fail with a "Fail (1 dep)" and clicking on it I can't see which dep it was. Any way to track that down? If not, might be a nice enhancement.

hvr commented 6 years ago

Although some of my packages fail with a "Fail (1 dep)" and clicking on it I can't see which dep it was. Any way to track that down?

If you're referring to the GHC 7.8 column being dark red: That's due to https://github.com/vincenthz/hs-memory/issues/44 (see also https://github.com/haskell-infra/hackage-trustees/issues/119 for the larger issue)

If not, might be a nice enhancement.

Indeed! This was actually a regression from the big rewrite of the backend for matrix v2 into matrix v3; but that's been handled by last year's GSOC project which brought the webui frontend from v2 to v3 as well... but there's still some loose ends to fixup before it can go live. TLDR: this is going to be addressed in the next major update...

hvr commented 6 years ago

The failures for recent hoogles of the kind

src/General/Conduit.hs:72:35:
    Not in scope: `BS.unsnoc'
    Perhaps you meant `BS.snoc' (imported from Data.ByteString.Char8)
<<ghc: 845086184 bytes, 1541 GCs, 13501702/51114944 avg/max bytes residency (8 samples), 105M in use, 0.00 INIT (0.00 elapsed), 0.48 MUT (0.53 elapsed), 0.49 GC (0.49 elapsed) :ghc>>
cabal: Failed to build hoogle-5.0.14 (which is required by exe:hoogle from hoogle-5.0.14).

are due to a too lax lower bound on bytestring; unsnoc was added in bytestring-0.10.2

ndmitchell commented 6 years ago

I've now removed the upper bounds on http-conduit and fixed the code to work, and also added lower bounds for bytestring and time, and released. With that, all 5.0 series packages have a clean matrix. Thanks for the alert and all the help!