Description of problem: Latest bash fails to execute configure script Version-Release number of selected component (if applicable): bash-3.1-2 How reproducible: 100% Steps to Reproduce: 1. Try to build postgresql in beehive 2. Notice that configure script for the "pgtcl" component fails Actual results: ... checking if 64bit support is enabled... no checking if 64bit Sparc VIS support is requested... no checking system version (for dynamic loading)... ./configure: line 5698: syntax error near unexpected token `(' ./configure: line 5698: ` case `(ac_space=' '; set | grep ac_space) 2>&1` in' [root@ls20-bc1-13 Pgtcl]# The "checking system version" message is at line 3546; I'm not sure how control suddenly jumps two thousand lines further down, but running the script under sh -x confirms that's what it's doing. Expected results: successful execution of script Additional info: The postgresql specfile forcibly re-autoconfs this sub-package, so the failing script has just been built by autoconf. I first thought that autoconf might be broken, but this same package has been built successfully since the last autoconf update, so the blame seems to rest with bash. I've assigned this "high" priority because it's blocking me from updating postgresql, which I really wanted to do before fc5test2 freeze :-(
The script seems a bit large to attach to this bz entry, but I've left a copy as ~tgl/configure on porkchop.
Buggy script at line 3561: system=MP-RAS-`awk '{print }' /etc/.relid'` should read system=MP-RAS-`awk '{print }' /etc/.relid` i.e. take out the third single-quote.
Argh, sorry to bother you with such a stupid thing. It's weird though that this script did not fail before. Have bash's quoting rules changed recently?
Indeed, bash-3.0 does not fail on that code. It must just be some small parser change. I think it's just good luck that it didn't fail before.