If this is still valid, could you post the output of:
urpme --test libspeech_tools1

Keywords: (none) => NEEDINFO

urpme --test libspeech_tools1
To satisfy dependencies, the following 4 packages will be removed (7.7MB):
  festival-1.96-12mdv2010.1.i586
   (due to missing libestbase.so.1.2.96.1,
    due to missing libestools.so.1.2.96.1,
    due to missing libeststring.so.1.2)
  gnome-speech-driver-festival-0.4.25-4.mga1.i586
   (due to missing festival)
  libspeech_tools1-1.2.96-12mdv2010.1.i586
   (due to unsatisfied speech_tools == 1.2.96-12mdv2010.1)
  speech_tools-1.2.96-12mdv2010.1.i586
   (due to missing libestbase.so.1.2.96.1,
    due to missing libestools.so.1.2.96.1,
    due to missing libeststring.so.1.2)
Remove 4 packages? (y/N) n

Why didn't festival get updated to festival-2.1-2.mga1 ?

Attach the urpmi-log.txt from executing:
urpmi --auto-update --debug 2>&1 | tee /tmp/urpmi-log.txt

Created attachment 356 [details]
Output of urpmi --auto-select --debug

Please test next festival and speech_tools packages, I think/hope it'll fix this issue...

Created attachment 377 [details]
Output of urpmi --auto-update --debug

Still doesn't force removal of libspeech_tools1-1.2.96-12mdv2010.1.i586

The log indicates you are using --keep option or have "keep" in /etc/urpmi/urpmi.cfg. This prevents the removal of libspeech_tools1.

There is still a bug in urpmi, though, as it shouldn't try to upgrade festival when "--keep" prevents removal of libspeech_tools1, which conflicts with festival upgrade.

Please run the failing urpmi command "--debug bug701" option, and then provide the created ./bug701 subdirectory in a tarball or zip, to ensure that we can reproduce this "--keep" bug in the future. This is also needed if you are not in fact using --keep and this is something else.

CC: (none) => anssi.hannula

Turns out I previously had --keep in the command, and in urpmi.cfg.

In response to a different bug, I had removed if from the command, but had
forgotten to remove it from urpmi.cfg.

# urpmi --auto-select --debug bug701
Error: can't use --auto-select along with package list.

Checking the man page, the debug option doesn't take any parameters.
What is it specifically you want me to run, and should I run it with the --keep
parameter?

Sorry, that's a typo in my part.

I meant "--bug bug701", and yes, I'd like it to be run with "--keep", since the urpmi bug is related to --keep.

The zip is too large for bugzilla, so I've uploaded it to
http://www.ody.ca/~dwhodgins/bug701.zip
(Please let me know when you have a copy, so I can delete it)

The actual command I ran was
urpmi --auto-select --resume --noclean --allow-force --allow-nodeps --keep --debug --bug bug701 2>&1 | tee /tmp/urpmi-log.txt

I moved the urpmi-log.txt into the bug701 directory prior to creating the zip.

The urpmi command displayed a mess on the screen, as shown in the log.txt,
and appeared to hang.

Using strace on a seperate konsole showed it was waiting on a read from stdin,
so I switched back to the konsole it was running in, pressed y and enter.

If it did display the message asking whether or not to continure, it was not
readable.  Should I now test without the --keep?

I have the zip (and I'll also keep it in http://onse.fi/files/bug701.zip).

Yes, no more information is needed so you can now test normally without --keep.

Thanks for your help. (note that --keep bugs are currently quite low priority, but we'll get to this bug eventually)

Created attachment 391 [details]
urpmi without --keep

Even without the --keep (removed from both the command and urpmi.cfg),
it's still failing.

Thanks to Anssi as usual.


IINM, you'll have to resolve the issue manually, you used --keep which installed newer packages, e.g. new festival package got installed and in the same time the old libspeech_tools1-1.2.96-12mdv2010.1 was kept, which wouldn't/shouldn't have happened as new festival requires new libeststring1.2 which conflicts with the old libspeech_tools1-1.2.96-12mdv2010.1.

Nope, the new festival was never installed.

Looks like something strange is happening here, I'll have to take a closer look later (I expect it to not be a trivial bug).

In the meantime, feel free remove libspeech_tools1 manually.

I've removed it manually, as well as other packages that were causing problems, that
were probably due to the previous use of the keep option.

Should the bug be closed, or left open for the keep problem?

Let's keep the bug open. And it is not just a --keep problem anymore, as per comment #12.

Keywords: NEEDINFO => (none)
Assignee: bugsquad => anssi.hannula
Source RPM: festival-2.1-2.mga1.src.rpm => perl-URPM

any news on that bug? shall we keep it as release_blocker  ?

CC: (none) => ennael1

I think that should be up to Anssi, but in my opinion, this bug should not be
considered a release blocker, now.

Anssi any news on this one? If nothing new before the end of this week we will decrease priority

As somewhat expected, I didn't have the time during the week to look at this (due to exams every day), but I'll look at this today or tomorrow.

OK, this doesn't seem to be a "regular" dependency solver algorithm issue, but somehow urpmi forgets the removals from the transaction.

I'll continue to debug this.

It is caused by the "allow-force: 1" you have in urpmi.conf.

When it is specified, urpmi never tries to remove packages due to conflicts (since it could be that the user specified --allow-force to avoid some invalid removal of lots of packages; mdv soft commit 13557).

The fix for this is non-trivial as sometimes when --allow-force is used we do not indeed want to remove any conflicting packages. So some semantic changes to the options are needed if we want to allow the "casual" use of allow-force.

This is not a release blocker, as this doesn't happen without "allow-force: 1" and changing the semantics of urpmi behavior now is a bit too last minute for mga1.

Removing the blocking of bug #56, this is not strictly related to mdv upgrade.

I'll get back to this bugreport at a later time (i.e. at least not before MGA 1).
In the meantime, ideas on how to handle this situation are welcome.

Status: NEW => ASSIGNED
Blocks: 56 => (none)

Probably the best way to handle this, would be an errata entry explaining
--keep and --allow-force (or the equivalent urpmi.cfg entries) should not
be used, since they prevent package removal causing conflicts later.

any news on this bugreport ?

CC: (none) => dmorganec

I think comment 22 indicates this can be closed.  The problem only happens
when people use --allow-force and --allow-nodeps, which I had done while
using Mandriva, to get around bug 2317 type problems, and dependency
problems with backports.

As Mageia is working around bug 2317 by linking needed dependencies, and
will be doing basic qa on backports, there shouldn't be any reason for
people to use --allow-force or --allow-nodeps.

Anssi, if you think the bug should be reopened, please do so.  In the meantime,
I'll go ahead an close the bug as invalid.

Status: ASSIGNED => RESOLVED
Resolution: (none) => INVALID