From: Ian Jackson <ijackson@chiark.greenend.org.uk>
Date: Mon, 25 Nov 2019 00:01:34 +0000 (+0000)
Subject: stest: If one of our secnets dies, call the test a failure
X-Git-Tag: v0.6.0~229
X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ian/git?p=secnet.git;a=commitdiff_plain;h=31c904ba594c34409eacde50020539c7290a460a

stest: If one of our secnets dies, call the test a failure

This involves use of TclX's `signal' facility.  In my tests it was
easy to make Tcl deadlock by doing too much work in the signal
handler.  In particular reaping children is a bad idea.  Also signals
are not blocked during the signal handler so it would have to be
reentrant.

Instead, use `after idle'.  That is quite soon enough for the reap to
run, and in my tests with TclX 8.4 it all works properly.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
---

diff --git a/stest/common.tcl b/stest/common.tcl
index d6250c3..a58ca6f 100644
--- a/stest/common.tcl
+++ b/stest/common.tcl
@@ -217,6 +217,21 @@ proc finish {estatus} {
     exit $estatus
 }
 
+proc reap {} {
+    global pidmap
+    #puts stderr REAPING
+    foreach pid [array names pidmap] {
+	set got [wait -nohang $pid]
+	if {![llength $got]} continue
+	set info $pidmap($pid)
+	unset pidmap($pid)
+	puts stderr "reaped $info: $got"
+	finish 1
+    }
+}
+
+signal -restart trap SIGCHLD { after idle reap }
+
 proc udp-proxy {} {
     global socktmp udpsock
     set u $socktmp/udp