[PATCH 0/7] rework run_command error reporting

[Johannes Sixt <j6t@kdbg.org>]

I got tired of thinking about which return values of run_command functions
are exit codes and which are error codes of system call failures.

Furthermore, it is a pitty that almost no run_command callers correctly
report system call failures.

But on top of all I find it unacceptable that recent git-daemon logs
"unable to run 'git-pack-objects'" if a client terminates the connection
early.

This series addresses these issues, and also fixes an error in git-bisect
on Windows.

1/7 Truncate result of run_command that is used in exit() to lowest 8 bits

    This one fixes a breakage of git-bisect on Windows and should be
    applied at any rate.

2/7 MinGW: simplify waitpid() emulation macros
3/7 run_command: return exit code as positive value
4/7 run_command: report system call errors instead of returning error codes

    These address the mentioned issues; 4/7 is the important change.

5/7 run_command: encode deadly signal number in the return value
6/7 run_command: report failure to execute the program, but optionally don't

    These two are nice to have.

7/7 receive-pack: remove unnecessary run_status report

    This one is actually RFC because it removes a status report that users
    might be used to see, or even depend on.


 bisect.c                     |    4 +-
 builtin-add.c                |    2 +-
 builtin-merge.c              |    2 +-
 builtin-receive-pack.c       |   38 ++--------------
 compat/mingw.h               |    5 +-
 convert.c                    |    2 +-
 git.c                        |   10 +---
 ll-merge.c                   |    4 --
 run-command.c                |  105 +++++++++++++++++++++++------------------
 run-command.h                |   13 +----
 t/t5530-upload-pack-error.sh |    5 ++-
 transport.c                  |   16 +------
 12 files changed, 81 insertions(+), 125 deletions(-)