At some point a change in Tcl's core (in both 8.5 and 8.4) resulted

in an unintended call to write ZERO bytes to any descriptor, that was
opened for writing, when closing it.

When the descriptor belongs to a pipe, with its other end closed -- such
as when the started process has exited, doing so results in a SIGPIPE.
Tcl's core ignores SIGPIPEs, but TclX' self-tests didn't and began to
fail.

The bug should now be fixed in Tcl, but this patch is needed for TclX to
pass its own tests even when using the uncorrected version of Tcl.

1 files changed, 32 insertions, 0 deletions

diff --git a/lang/tclX/files/patch-signal-test b/lang/tclX/files/patch-signal-test
new file mode 100644
index 000000000000..ffe4c1003dc1
--- /dev/null
+++ b/lang/tclX/files/patch-signal-test
@@ -0,0 +1,32 @@
+--- tests/signal.test	2002-04-02 21:44:21.000000000 -0500
++++ tests/signal.test	2009-11-26 12:19:20.000000000 -0500
+@@ -329,5 +329,5 @@
+     echo Notice: restarting of interrupted system calls is not available on this system.
+ } else {
+-    test signal-3.0 {kill tests} {
++    Test signal-3.0 {kill tests} {
+         # Start program on a pipe, but don't doing anything until we send it
+         # something to make timing issues smaller.
+@@ -346,5 +346,14 @@
+             set resp [gets $sp]
+         } {} {
+-            catch {close $sp}
++	    signal -restart error SIGPIPE
++	    if {[catch {close $sp} e]} {
++		if {$e == "SIGPIPE signal received"} {
++		    puts stderr "$e while closing $sp. This Tcl bug is fixed in recent 8.4 and 8.5 releases"
++		} else {
++		    puts stderr "Unexpected error $e while closing $sp."
++		    append resp - $e
++		}
++	    }
++	    signal default SIGPIPE
+             alarm 0
+         }
+@@ -353,5 +362,5 @@
+         # Note this this is somewhat timing dependent and might fail
+         # on a heavly loaded system.
+-    } {HELLO 1}
++    } 0 {HELLO 1}
+ }
+