Robert Elz <kre%munnari.OZ.AU@localhost> writes: > Date: Thu, 12 Dec 2013 18:05:00 +0400 > From: Aleksej Saushev <asau%inbox.ru@localhost> > Message-ID: <8738ly5fn7.fsf%inbox.ru@localhost> > > | Shouldn't that be "make >& OUT"? > > No .. or that is it could be, but that's a hack for cshcompatability, > and not recommended, the >&file syntax is ambiguoys with the >&N syntax > for rearranging file descriptors, so bash uses &> for the csdh'ism >& > (of course, ">FILE 2>&1" works too, and works in all sh's so I just use that. Bash uses "&>" as a shortcut, but I ran with the POSIX-specified make > /dev/null 2>&1 & and got the hang. ps alxw says that scons is at fault. > Greg, run the make again (no need to clean), in foreground, and see > what happens. There is a need to clean; pkgsrc make won't actually run the build target if work/.build_done is present. (I actually tried this.) If I redirect stdin as well, the problem does not occur: make < /dev/null > /dev/null 2>&1 & so the problem is that scons does something to stdin. After about 20 print statements, I present the following minimal test program, which blocks if run in the background but runs fine in the foreground. (scons hangs where it does "import readline".) So this is either a readline bug, or a bug in scons for invoking readline when the input should not be read.
#!/usr/pkg/bin/python2.7
import sys
sys.stderr.write("before readline\n")
import readline
sys.stderr.write("after readline\n")
Attachment:
pgpC93iIsVShH.pgp
Description: PGP signature