1# From arnold@f7.net Fri Nov 26 11:53:12 2004 2# X-Envelope-From: james@nocrew.org 3# X-Envelope-To: <arnold@skeeve.com> 4# To: bug-gawk@gnu.org 5# Subject: gawk 3.1.4: reproducible hang, regression from 3.1.3 6# From: James Troup <james@nocrew.org> 7# Date: Fri, 26 Nov 2004 03:14:05 +0000 8# Message-ID: <877jo9qp36.fsf@shiri.gloaming.local> 9# User-Agent: Gnus/5.1006 (Gnus v5.10.6) Emacs/21.3 (gnu/linux) 10# MIME-Version: 1.0 11# Content-Type: text/plain; charset=us-ascii 12# 13# 14# Hi, 15# 16# A Debian user reported[0] gawk 3.1.4 broke a (relatively) complex 17# program that makes extensive use of awk, called 'apt-move'. I finally 18# managed to reduced the problem down to a 3 line test case, enclosed 19# below[1]. 20# 21# I believe the problem comes from the following code, introduced in 22# 3.1.4: 23# 24# [io.c, line 560] 25# | for (rp = red_head; rp != NULL; rp = rp->next) { 26# | if ((rp->flag & RED_EOF) && tree->type == Node_redirect_pipein) { 27# | if (rp->pid != -1) 28# | wait_any(0); 29# | } 30# 31# The problem is that, if we have an existing redirect which is a simple 32# file redirect[b] and it's hit EOF and we try to create a new '|' 33# redirect[c], this new code will try to wait(2) and if there are any 34# other redirects which _did_ spawn a child (like [a]) the wait() will 35# hang indefinitely waiting for it to exit. 36# 37# Hope that makes sense :) 38# 39# -- 40# James 41# 42# [0] http://bugs.debian.org/263964 43# 44# [1] 45# ================================================================================ 46#!/usr/bin/gawk -f 47 48BEGIN { 49 printf "" | "cat" # [a] 50 getline line < "/dev/null" # [b] 51 "true" | getline line # [c] 52} 53# ================================================================================ 54