On Thu, Sep 07, 2006 at 05:26:53PM -0500, Luis Rodrigo Gallardo Cruz wrote: > On Thu, Sep 07, 2006 at 08:27:43PM +0200, kku wrote: > > >You are programaticaly managing a configuration file in /etc. You > > >should look into using ucf, that already handles this. > > > > I've had a look at ucf, but I do not think it fits my need (unless I have > > overlooked something). > > > > I have a template file > > "/usr/share/doc/personalbackup/personalbackup.apache" and the parameter > > "webalias" > > is being replaced in that template during postinst. Depending on the user's > > choose for apache/apache2 this file will be placed in > > '/etc/apache/conf.d/personalbackup' or > > '/etc/apache2/sites-enabled/personalbackup' > > > > Now to detect that the user did (not) change the conf file, I check the > > md5sum of the content of the appropriate conf file 'except' for the Alias > > line. > > And as far as I can see this is something ucf cannot handle...or am I wrong > > here? > > Uhh. From apt-cache show ucf
Luis, I think you missed a critical bit of what's being done with the config file -- the md5sum of only part of the config file is being checked, which (unless it's has had some *very* interesting features added lately) ucf can't do. This doesn't, however, change the fact that this "only compare parts of the config file" approach isn't policy compliant. If I change *any* part of a config file, I expect those changes to be preserved. Debconf Is Not A Registry is a sacred and hallowed principle, and it's still applicable here. I don't care if I answered a certain question a certain way when the package was first installed -- the Unix philosophy is that I go and edit config files directly when I want things changed. If the package goes and puts things back how they were because of a debconf setting (overriding my carefully hand-crafted changes) then that's a bug. - Matt -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
