[PATCH] crecord: fix issue when backgrounding editor would leave artefact

[lc2817 at columbia.edu]

# HG changeset patch
# User Laurent Charignon <l.charignon at gmail.com>
# Date 1437017963 25200
#      Wed Jul 15 20:39:23 2015 -0700
# Node ID aacc79aaf710f69c3e060920bcf3255d88640a87
# Parent  35fa7c77c754aa4d156c42abfdb61ca178468872
crecord: fix issue when backgrounding editor would leave artefact

Before this patch:
- if a user was entering a commit message after having ran the curses
interface
- and then uses ctrl-z, followed by fg to put the editor in the
background/foreground
- then the curses interface would leave artefact on the screen of
the editor, making entering the commit message a difficult task

This happened because ncurses registers a signal handler for SIGTSTP and
does not restore the original signal handler after running.
More info at:
http://stackoverflow.com/questions/31440392/
curses-wrapper-messing-up-terminal-after-background-foreground-sequence/
31441709#31441709

This patch restores the original value of the signal handler after
running the curses interface and therefore fixes this issue.
It don't know how to add a test for this issue, I tested the scenario
above manually and it works correctly with the patch.

diff -r 35fa7c77c754 -r aacc79aaf710 mercurial/crecord.py
--- a/mercurial/crecord.py	Sun Jul 12 17:59:25 2015 +0900
+++ b/mercurial/crecord.py	Wed Jul 15 20:39:23 2015 -0700
@@ -482,7 +482,10 @@
     """
     ui.write(_('starting interactive selection\n'))
     chunkselector = curseschunkselector(headerlist, ui)
+    f = signal.getsignal(signal.SIGTSTP)
     curses.wrapper(chunkselector.main)
+    # ncurses does not restore signal handler for SIGTSTP
+    signal.signal(signal.SIGTSTP, f)
 
 def testdecorator(testfn, f):
     def u(*args, **kwargs):