Builds and passes tests with GHC 7.10.1-rc1.

A few dependent packages need minor fixes for GHC 7.10 so it
won't cabal install out of the box - I'll try to get those fixes
upstream.

I've also checked it builds with GHC 7.4,7.6,7.8.

1 patch for repository darcs-unstable@darcs.net:screened:

Wed Dec 31 17:08:41 GMT 2014  Ganesh Sittampalam <ganesh@earth.li>
  * support GHC 7.10/base 4.8

Attachments

• patch-preview.txt
• support-ghc-7_10_base-4_8.dpatch
• unnamed

What is the status of this? Have the upstream fixes been released to
hackage by now?

At least one has been released. Unfortunately I forgot about sending 
all the fixes upstream.

I'll check the situation again with GHC 7.10RC2, but given that the 
Prelude changes for 7.10 are now being reconsidered (see 
https://www.haskell.org/pipermail/libraries/2015-
January/024777.html), it may be best to hold off a bit for now 
anyway. Probably simplest to leave this under review for now given 
it's already in screened.

Rather than try to update a package that hasn't been uploaded since 2011 and has 
a dead github repository URL, I've submitted patch1299 to drop the dependency on 
datetime.

That leaves shellish and cmdlib, for which I sent patches to the maintainer 
yesterday.

Guillaume, I think this should go into 2.10 as we always find ourselves wanting 
to support recent GHCs on the stable branch so we might as well do this now.

Agreed. The patch is in branch 2.10 now.

Just a general update on the GHC 7.10 support:

GHC 7.10rc3 also requires a patch to ansi-wl-pprint which others are 
already chasing: https://github.com/batterseapower/ansi-wl-
pprint/pull/11

Still not heard anything about shellish/cmdlib.

So it's probably a good idea to wait for the final release of GHC 7.10
before releasing darcs 2.10.

Would it make sense to get rid of the shellish and cmdlib dependencies?
They are used in the test harness only.

shellish could be replaced by either shell-conduit or turtle; cmdlib by
cmdargs, or the standard flag parsing modules.

Might make sense. There's also shelly which was forked from shellish.

Here's a patch that switches the test harness from cmdlib/shellish
to cmdargs/shelly.
The code is full of Shellish.FilePath <-> Text <-> String
conversions and could be enhanced in some places. But everything
seems to work (so far only tested with ghc 7.6 under Linux).

hashed-storage's test script only depends on test-framework so
it does not require any change.

1 patch for repository http://darcs.net:

patch a916f4d2e44706ac6252d6d1f01b23ba588d8057
Author: Guillaume Hoffmann <guillaumh@gmail.com>
Date:   Tue Mar 31 01:25:34 ART 2015
  * port test harness to cmdargs and shelly

Attachments

• patch-preview.txt
• port-test-harness-to-cmdargs-and-shelly.dpatch
• unnamed

With the cmdargs/shelly patch, the shell tests don't seem to print out 
the output of failing tests any more. I haven't investigated in any 
more detail yet.

I've tested it also with GHC 7.10 and failing tests are properly reported.

I've looked into the problem some more now.

In general failing tests are properly reported, but two tests in particular start failing silently 
(whereas they were passing before the switch). The two tests are emailformat.sh and latin9-input.sh 
and it turns out that the test harness itself is throwing something like "fd:25: hGetLine: invalid 
argument (invalid byte sequence)", which explains the lack of output.

These two tests have funnily encoded characters in them so I assume that some handle somewhere isn't 
in the right mode. I'll do some more digging.

OK, I understand the details better now: 
https://github.com/yesodweb/Shelly.hs/issues/96

Hopefully it shouldn't be too hard to fix or workaround.

I see. Thanks for looking into this bug! Doing this we also contribute
to enhancing Shelly.

I'm going to self-accept the shelly/cmdargs patch into branch 2.10 to
prepare the release, and open a separate bug for shelly.