Comment 28 for bug 25620

Revision history for this message
Debian Bug Importer (debzilla) wrote :

Message-ID: <email address hidden>
Date: Sun, 13 Nov 2005 16:40:12 +0100
From: =?iso-8859-1?q?Frank_K=FCster?= <email address hidden>
To: <email address hidden>
Cc: Joey Hess <email address hidden>, Norbert Preining <email address hidden>,
 <email address hidden>, <email address hidden>,
 teTeX maintainers <email address hidden>
Subject: Re: Bug#338587: tex-common fails to install

Eric Valette <email address hidden> wrote:

> I (of course!) took the postinst of the failing packages aka tex-common.
> (Do not assume I'm stupid please).=20

Sorry, I didn't want to offend; I just had a problem understanding your
sentence, and that might well be due to my english being not good
enough, not yours.=20

> When I ran "postinst configure" it
> prompted a dialog window asking to either keep my installed version of a
> configuration file (THAT configuration file name I do not remember)=20

upon rereading, I get the meaning, sorry.

> and
> I selected replace by maintainer version as I wanted to be sure the
> problem was not comming from incompatible version of this configuration
> file. The "postins" command suceeded and the "apt-get -f install" after i=
t.
>
> So if the problem was involing only "debconf", I suspect the run by hand
> "postinst configure" would have not solved it magically. You can
> continue to play ping pong between tex-common and debconf or close the
> bug. I guess it will resurface later via another bug.

If there is a bug in tex-common, then there are two bugs, one in
debconf, and one in tex-common. After running the tex-common postinst
script again, it might be that the debconf code that caused the error
message was no longer called.

But I still do not understand things here.

$ grep "db_" tex-common/debian/* | grep -v 'true$'
tex-common/debian/config.in:db_version 2.0
tex-common/debian/config.in:db_go
tex-common/debian/config.in: db_go
tex-common/debian/postinst.in:db_version 2.0

which means that all lines with calls to db_* end with || true except
for db_version and db_go. This seems to indicate to me that *if* we
made an error with debconf, it wouldn't reveal itself as a failing
maintainer script.

Second, how is it possible that a conffile dialog (in fact it must be a
ucf configuration file dialog, but that looks similar) pops up during a
second invocation of the postinst script, unless you have changed the
file meanwhile? ucf is called only once in the postinst script, and
this is

- after sourcing /usr/share/debconf/confmodule and calling db_version,
  but before any other debconf use

- only conditional on the first parameter being configure or
  reconfigure.=20=20

I don't see how a configuration file could be on your system that is
neither any old known nor the new version unless one of two things
happened:

- it was a problem with your local debconf installation, and it was in
  fact fixed by reinstalling it, although you first said it wasn't (how
  did you check?). In this case, the postinst script failed right while
  loading confmodule, and never got to the file changed dialog.

- or you have edited some configuration file below /etc/texmf between
  the different postinst invocations. But in this case I cannot see how
  changing the conffile would affect later debconf invocations - we only
  use db_get and db_fset, and we do that without checking any files.

> BTW the bug leading for the change in tex-common 0.9->0.10 looks very
> similar as my debconf default config is "dialog/LOW".

Which bug do you mean - #337073? Why do you think it looks similar -
it's about an error in a chmod invocation, while yours involves a Perl
error in Debconf.

Regards, Frank

--=20
Frank K=FCster
Inst. f. Biochemie der Univ. Z=FCrich
Debian Developer