GitLab

Signed-off-by: Herbert Valerio Riedel <hvr@gnu.org>

The Hugs CPP conditionals in `System.Process.Internals` make little
sense, as that module is only listed in `exposed-modules` when
`impl(ghc)` and moreover, it's `import` in `System.Process` is guarded
by an `#ifndef __HUGS__`.
Signed-off-by: Herbert Valerio Riedel <hvr@gnu.org>

Since ad1a5c88

 `process` supports GHC>=7.2 (which implies `unix>=2.5`),
therefore `MIN_VERSION_unix(2,5,0)` is always true.
Signed-off-by: Herbert Valerio Riedel <hvr@gnu.org>

These automake aux files are not needed (anymore).
Signed-off-by: Herbert Valerio Riedel <hvr@gnu.org>

The problem is in the exception handler in readProcess. When it receives
an asynchronous exception, it tries to clean up by closing the pipes.
However the attempt to close outh blocks because the reader thread
(reading with hGetContents) is blocking on the handle.

This fixes #8483.
Signed-off-by: Austin Seipp <austin@well-typed.com>

A comma in AC_CHECK_FUNCS prevented setitimer from being detected.
Signed-off-by: Austin Seipp <austin@well-typed.com>

Supporting GHC 7.2 again was just a single `#ifdef`. Otoh, supporting
GHC 7.0 would require reverting part of d10b168a

.

This also updates the build-dependancy version constraints to reflect
the current reality. The code is `-Wall` clean for GHC 7.4 and later.
Signed-off-by: Herbert Valerio Riedel <hvr@gnu.org>

This includes declaring all potentially used language pragmas in
`other-extensions` and declare the `CPP` in the `.hs` files instead of
relying on `default-extensions`.

The confusing NHC98/HUGS conditionals should be removed, as the package
most probably won't work w/ those compilers anyway these days.
Signed-off-by: Herbert Valerio Riedel <hvr@gnu.org>

Signed-off-by: Edward Z. Yang <ezyang@mit.edu>

Two fixes:

    - Properly reimplement 'cat' using the 'write' system
      call instead of printf; printf was expecting a NULL-terminating
      string and was running over into the stack.

    - Use the process007_fd shim program for the second cat
      as well; essentially, apply the e380844b

 fix to all cases.
Signed-off-by: Edward Z. Yang <ezyang@mit.edu>

dash doesn't support 2-digit FD numbers with the 0<&10 syntax, and
some ways were using FDs that high. So now we have a little C program
to do the job instead.

This error was breaking validate on Linux for me.

true and false aren't in /bin on OS X. Now we don't hardcode the path.

On non-Windows, the child thread now comunicates any errors back to the
parent thread via pipes.

It's had a comment reminding us to do so since 2008.