On Fri, Jun 01 2012, Jameson Graef Rollins <jrollins at finestructure.net> wrote:
> There's a bug in the configure script that is causing auto-reruns of > ./configure to not inherit original command line options if there was > more than one. For instance, if I run: > > ./configure --with-gmime-version=2.4 --prefix=/home/jrollins/opt/notmuch > > Then in Makefile.config I get: > > configure_options = > --with-gmime-version=2.4--prefix=/home/jrollins/opt/notmuch > > This means that auto-reruns of configure will not get the proper > options. > > I tracked this down to an issue with IFS and /bin/sh. The first line of > ./configure is: > > readonly DEFAULT_IFS=$IFS > > DEFAULT_IFS is then used to reset IFS after it is modified within the > script. The problem is that /bin/sh is setting DEFAULT_IFS to be NULL > (i.e. ''), which leads to no separation between variables when "$@" is > expanded (which is itself problematic since I don't think "@" should use > the IFS when expanded). So this might be a bug in dash. I don't know. > In any event we need to fix this somehow. dash manual states that readonly name=value should work, but it seems to have some side effects. I tried some alternatives on command line: $ readonly FOO1=$IFS $ echo "$FOO1" | od -f x1 0000000 0a 0000001 $ readonly FOO2="$IFS" $ echo "$FOO2" | od -f x1 0000000 20 09 0a 0a 0000004 $ FOO3=$IFS $ readonly FOO3 $ echo "$FOO4" | od -f x1 0000000 20 09 0a 0a 0000004 I.e. in dash 'readonly name=$value' $value gets expanded like it was given in command line and not when just doing variable assignment. Easiest is just to drop the 'readonly'; too bad it caused problems :( (I have to go now -- but I'll make a patch later unless someone beats me by doing it sooner ;) Tomi > I see two obvious solutions: > > * use /bin/bash. This a one line diff that fixes the problem > immediately. > > * replace: > > IFS=$DEFAULT_IFS > > with: > > unset IFS > > Unsetting IFS also resets IFS to the default, without going through > this intermediate step. > > I'm ok with either solution, but I'll wait for some feedback since I > imagine someone will have an opinion. > > jamie. > _______________________________________________ > notmuch mailing list > notmuch at notmuchmail.org > http://notmuchmail.org/mailman/listinfo/notmuch