GitLab

Bump base upper bound

testsuite/process011: Don't rely on Python

The refactor of `process011` performed in 8cd7e043, where
the test is taught to use `python3` instead of `sh`, has broken the test
on Debian 10. Specifically, for reasons that I don't yet understand,
`python3` exits with code 1 not code 2 when killing itself with
signal 2. Strangely, my tests on other distributions (e.g. NixOS)
suggest that Python's behavior in this case is rather inconsistent.

To avoid this, we now rather use a dedicated C program instead of
Python.

Fixes #241.

Update autoconf to 2.71

This appears to have been an oversight.

fix typos

Phyx/winio permissions and mode fixes

Co-authored-by: Michael Snoyman <michael@snoyman.com>

Fix process tests on OpenBSD

Fix waitForProcess not closing process handles with delegate_ctlc

endDelegateControlC throws UserInterrupt, which would cause
modifyProcessHandle to roll back, leaving the handle open
despite the waitpid call.

Instead, we postpone the call to endDelegateControlC, as is already
the case in getProcessExitCode.

Additionally, remove endDelegateControlC entirely from the Windows-only
OpenExtHandle branch, where it's a no-op and was also in the wrong place.

It turns out that POSIX allows implementations to report
errors "lazily" by exiting the spawned process with code 127.

In particular, OpenBSD's implementation returns successfully if the
requested executable doesn't exist. This severely hampers our ability to
report errors accurately.

In this patch we add a `configure` test for sensible error-reporting
from `posix_spawnp` and disable it if the test fails.

Fixes #224.

Fix #219

It turns out the autoconf upper-cases the symbol name.

When `execvpe` is not available we use `find_executable` to search for
the executable prior to `fork`ing (fixing GHC #19994). However,
previously we failed to account for the fact that `exec` will be called
from a different working directory than that of the spawning process if
the user has set `cwd`.

Fix this by teaching `find_executable` to search relative to the
requested working directory.

Fixes #219.

In contrast to glibc, Darwin's posix_spawnp implementation doesn't set
errno on failure.

Fixes #227.

I suspect this will be OK on other systems that use posix_spawn.

pdksh (OpenBSD /bin/sh) behaves differently compared to most other
shells: https://marc.info/?l=openbsd-tech&m=164153793509651&w=2

The same subprocess behavior can be achieved by a python script.

Fixes process011 and process009 tests on OpenBSD.

Remove reqlib modifier from ghc test