On Thu, Jul 17, 2014 at 10:57:55AM +0200, Jonas Smedegaard wrote: > Quoting Agustin Martin (2014-07-15 18:11:06) > > On Fri, Jul 11, 2014 at 11:37:14PM +0200, Jonas Smedegaard wrote: > >> Please consider applying the bugfix also to a stable point release. > > > > The underlying bug is debconf database corruption and it is not fixed > > by dictionaries-common (nor known to be caused by it). > > > > Previously, dictionaries-common complained loudly (as above) when > > noticing it, pointing users to fix debconf database corruption (and > > giving advice on that), so underlying bug can be fixed and have things > > working as expected again. > > Uhm, please correct me if I'm wrong, but my experience is that > dictionaries-common not only "complains loudly" but *fails*, which > causes the whole package upgrade to fail.
Hi, That is what I meant by "complain loudly". Original debconf bug report is old, http://bugs.debian.org/198297, more than 10 years ago. All this was intended to try to provide as much valuable info as possible to debconf maintainers in a rather imperative way, I admit. But it was not a nonsense at that time. dictionaries-common was acting like a police officer enforcing a reaction, so debconf maintainers got as much info as possible. This had the benefit of getting a consistent debconf database at a small price. Most database corruption was apparently caused by write errors when saving databases. Long ago debconf was improved to deal better with this case and bug report was closed. However, seems that yet another database corruption is happening, causing the error message to reappear. > Seems that _this_ bug is not about the underlying corruption but how > "loudly" as you call it it is treated by dictionaries-common (why else > close it for newer package releases without the underlying issue > fixed?). In the meantime we have changed our mind that this should be dealt with more gracefully, by means of a debconf question warning and a better default selection in case of debconf database corruptiom (old was based on debconf values that might no longer be present), and that is how things work now, dictionaries-comon is now the good citizen that warns and gives advice. But old behavior had been there for years. > > Since those changes do not fix the real underlying problem (debconf > > database corruption), but only change the package complains about it, > > I do not think it is eligible for a stable point release. Do not know > > what stable release team would think about this. > > I believe the issue of dictionaries-common failing - thereby > interrupting upgrade of upnrelated packages - is in itself serious > enough for a stable fix, even if underlying issue is still unfixed. Current status involves a number of changes to make things work better in the general case. And this is what I consider too much for a stable fix, and I do not think the stable release team would like it. The advantage for this part is that it is better tested. To ask the release team about this, I think I should prepare first a minimal changeset including the debconf question and the way possible values are extracted. The drawback of this is that is less tested. so I fear about introducing new bugs in stable unless I test this extensively. I had time to do a first cut for such minimal changeset, attached. I still have to test it better before asking the release team. > Since you seem to understand the underlying issue (I don't) did you file > appropriate bugreport about that? If so, please mention bug number > here, for others stumbling across this bugreport and (unlike me) > interested concerned not about package upgrade interruption but the > underlying corruption issue. I already mentioned that original bug report is old and closed. Besides some stable boxes, I run a couple of testing and sid boxes that are upgraded in a nearly daily basis. I have never run into this debconf database corruption, neither the old nor the new one, so I did not open a new bug report. I would not have real info to provide myself. -- Agustin
diff --git a/debian/changelog b/debian/changelog index 296af42..baf4040 100644 --- a/debian/changelog +++ b/debian/changelog @@ -1,3 +1,13 @@ +dictionaries-common (1.12.11+deb7u1) unstable; urgency=low + + * Improve general behavior on debconf database corruption: + - update-default.in: Use /var/lib/dictionaries-common info as last + resource to try getting a default in case of debconf database + corruption. This should work around installation error better. + - Use a debconf question to warn about debconf database corruption. + + -- Agustin Martin Domingo <[email protected]> Tue, 22 Jul 2014 15:52:50 +0200 + dictionaries-common (1.12.11) unstable; urgency=low * debian/control: Add Breaks against ancient myspell-cs-cz, diff --git a/debian/dictionaries-common.templates b/debian/dictionaries-common.templates index 965ec8f..b20d568 100644 --- a/debian/dictionaries-common.templates +++ b/debian/dictionaries-common.templates @@ -1,3 +1,19 @@ +Template: dictionaries-common/debconf_database_corruption +Type: note +Description: Possible debconf database corruption + The setting for "${question}" is missing, but packages providing + candidates are installed. + . + This may be due to corruption in the debconf database. See + "/usr/share/doc/dictionaries-common/README.problems" on "Debconf + database corruption". + . + In this case, running "/usr/share/debconf/fix_db.pl" can help to put + the debconf database in a consistent state. + . + Some questions are likely to be asked after this message in order to + leave the dictionaries system in a (provisionally) working state. + Template: dictionaries-common/invalid_debconf_value Type: note _Description: An invalid debconf value [${value}] has been found @@ -108,4 +124,3 @@ _Description: Default values for ispell dictionary/wordlist are not set here will not set the default ispell dictionary. . Use instead 'select-default-ispell' or 'select-default-wordlist' scripts. - diff --git a/scripts/system/update-default.in b/scripts/system/update-default.in index 8333713..6b1bbe2 100644 --- a/scripts/system/update-default.in +++ b/scripts/system/update-default.in @@ -74,17 +74,13 @@ if ( $ret == 0 && $value ){ $manual++; } else { # Question is empty, but $class elements are installed. Two possibilities. - if ( -x "/usr/bin/apt-extracttemplates" ){ - # apt-utils is installed. Other possibility is debconf database corruption. - my $shared_question = "shared/packages-$class"; - print STDERR "$program: Question empty but elements installed for class \"$class\"\n"; - print STDERR " $question: return code: \"$ret\", value: \"$value\"\n"; - print STDERR " Choices: " . metaget ($question, "choices") . "\n"; - my ($shared_ret,$shared_owners) = metaget ($shared_question, "owners"); - print STDERR " $shared_question: return code: \"$shared_ret\" owners/error: \"$shared_owners\"\n"; - print STDERR " Installed elements: " . join(', ',sort keys %{$dictionaries}) . "\n\n"; - print STDERR " Please see \"/usr/share/doc/dictionaries-common/README.problems\", section\n"; - print STDERR " \"Debconf database corruption\" for recovery info.\n\n"; + if ( -x "/usr/bin/apt-extracttemplates" ){ # apt-utils is installed. + # This seems to be debconf database corruption. Warn loudly about it. + my $debconf_dbase_corruption_question = "dictionaries-common/debconf_database_corruption"; + subst($debconf_dbase_corruption_question,"question",$question); + input("high",$debconf_dbase_corruption_question); + title ("dictionaries-common: [I:ispell dictionaries:I][W:wordlists:W]"); + go (); } else { # apt-utils is not installed, so $class templates may not have been parsed at preconfigure stage. # Delay settings until ispell dictionaries/wordlists are configured and their debconf templates parsed. @@ -96,22 +92,24 @@ if ( $ret == 0 && $value ){ unless ( $manual ){ # Handle invalid debconf values if ( not exists $dictionaries->{$value} ){ - my @available_keys = (); - foreach ( split (/\s*,\s*/, metaget ($question, "choices")) ){ - # strip leading/trailing whitespace and create a list of available keys - s/^\s+//; - s/\s+$//; - push (@available_keys,$_) if ( defined $dictionaries->{$_} ); - } + # debconf database is presumably corrupted. Try after stored values. + my @available_keys = sort {lc $a cmp lc $b} keys %$dictionaries; my $choices = join (', ', sort {lc $a cmp lc $b} @available_keys); + print STDERR "$program: Trying to get a default value from \"$choices\"\n"; my $forced_key = $available_keys[0] || die "$program: Selected [I:ispell dictionary:I][W:wordlist:W]" . " \"$value\" \n" . "does not correspond to any installed package in the system\n" . "and no alternative [I:ispell dictionary:I][W:wordlist:W] could be selected.\n"; - subst($iquestion,"value",$value); - fset ($iquestion,"seen","false"); - input("high",$iquestion); # Warn about what happened + + # Warn about what might have happened if not debconf database corruption + if ( $value ){ + subst($iquestion,"value",$value); + fset ($iquestion,"seen","false"); + input("high",$iquestion); + } + + # Prepare a selection box to get a default. subst ($question, "choices", $choices); # Put sane values in debconf choices field subst ($question, "echoices", $choices); # Put sane values in debconf echoices field set ($question, $forced_key); # Set debconf value to a sane one
