System.Cmd.Utils.pipeBoth subject to pipeline stalls with ghc 7

[nomeata]

Hi John,

this is to track the bug reported by Joey at http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=624389 here, to ensure it is not forgotten. For your convenience, here is the original report:

For quite a while I have been using missingh's pipeBoth with success; but as
soon as my program was rebuilt with ghc 7, it started stalling when large
quantities of data needed to be passed through the pipe.

Here is a simple test case. It needs to run in a git repository.

import System.Cmd.Utils

main = do
        as <- checkAttr "blah" $ map show [1..100000]
        sequence $ map (putStrLn . show) as

checkAttr attr files = do
        (_, s) <- pipeBoth "git" params $ unlines files
        return $ lines s
        where
                params = ["check-attr", attr, "--stdin"]

It queries git for attribute values for 100000 files. With ghc 6, it
should run to completion. With ghc 7, it stalls, deadlocked, after
a varying number of files, under 1000:

select(2, [], [1], NULL, {0, 0})        = 1 (out [1], left {0, 0})
write(1, "\"701: blah: unspecified\"\n", 25"701: blah: unspecified") = 25
select(2, [], [1], NULL, {0, 0})        = 1 (out [1], left {0, 0})
write(1, "\"702: blah: unspecified\"\n", 25"702: blah: unspecified") = 25
select(2, [], [1], NULL, {0, 0})        = 1 (out [1], left {0, 0})
write(1, "\"703: blah: unspecified\"\n", 25"703: blah: unspecified") = 25
select(2, [], [1], NULL, {0, 0})        = 1 (out [1], left {0, 0})
write(1, "\"704: blah: unspecified\"\n", 25"704: blah: unspecified") = 25
--- SIGVTALRM (Virtual timer expired) @ 0 (0) ---
sigreturn()                             = ? (mask now [])
gettimeofday({1303958431, 174266}, NULL) = 0
select(7, [], [6], NULL, {0, 0})        = 1 (out [6], left {0, 0})
write(6, "345\n15346\n15347\n15348\n15349\n1535"..., 8096

The program is blocked trying to write to git-check-attr, and
git-check-attr is in turn blocked waiting for its output to be read.

I've skipping over missingh and filing this bug directly on ghc because
I think it's unlikely missingh is at fault. IIRC, pipeBoth works by
sparking off a helper thread, which is used to write input to a command.
Unless it made a bad assumption about that being a safe thing to do,
this must be a bug in GHC?

FWIW, I have worked around this in my code by forking a process, not a
thread, to do the writing. Which works fine, just a little more
heavyweight than needed. I'm concerned about all the other potential
callers of pipeBoth out there, however.