installation process can crash due to an issue with one package when choosing "install updates" as part of the install

Exception during installation:
Jan 28 03:17:46 ubuntu plugininstall.py: SystemError: E:Sub-process /usr/bin/dpkg returned an error code (1)
Jan 28 03:17:46 ubuntu plugininstall.py:

Jan 28 03:17:27 ubuntu ubiquity: dpkg-deb (subprocess): data: internal bzip2 read error: 'DATA_ERROR'
Jan 28 03:17:27 ubuntu ubiquity: dpkg-deb: error: subprocess <decompress> returned error exit status 2
Jan 28 03:17:27 ubuntu ubiquity: dpkg: error processing /target/var/cache/apt/archives/libreoffice-core_1%3a3.5.0~beta2-2ubuntu4_i386.deb (--unpack):
Jan 28 03:17:27 ubuntu ubiquity: short read on buffer copy for backend dpkg-deb during `./usr/lib/libreoffice/program/libooxlo.so'
Jan 28 03:17:27 ubuntu ubiquity: No apport report written because MaxReports is reached already

Status changed to 'Confirmed' because the bug affects multiple users.

My bug just got marked as a dupe of this one. I'd like to note two observations:

- I'd explicitly unchecked "install updates during install". I've noticed for the past 2-3 releases that Ubiquity completely ignores what you tell it in this regard and downloads updates during the install process anyway.
- If Ubiquity sees that a package doesn't match the checksum, why doesn't it wipe it out and try to re-download it, rather than just crashing?

It doesn't actually have anything directly to do with "install updates during install". This isn't downloading updates, it's downloading language packs.

Checking package checksums is supposed to be apt's responsibility, not ubiquity's. apt normally *does* check the checksum, and in my tests a proxy that deliberately corrupts packages on the way through causes an exception to be raised, which at worst ought to cause ubiquity to display an error message but continue largely normally. This isn't happening in the cases in this bug and its duplicates. Rather than adding another layer of error checking which duplicates a layer of error checking that's supposed to be there already, I would like to figure out why the current layer of error checking isn't working.

There isn't even a particularly straightforward interface for ubiquity to double-check the checksums itself. I think this can only be sensibly handled as an apt bug.

Actually, while I do still think this is an apt bug, I've found a workaround in ubiquity which doesn't seem too unreasonable; the effect will be that any corrupted downloads will be removed and the entire set of packages currently attempting to be installed will be skipped; that will normally mean language packs don't get installed, which is not ideal but not a showstopper. I think that will be enough to let me take this off the rls-p-tracking list.

This bug was fixed in the package ubiquity - 2.10.12

---------------
ubiquity (2.10.12) precise-proposed; urgency=low

  * Fix download verification to handle epochs correctly, and add a little
    more logging so that we can debug any further errors along these lines.
  * GTK frontend:
    - Disable WebKit's cache, since the slideshow doesn't have a significant
      browsing interface and the cache is memory-hungry. This may help with
      LP #979350.
 -- Colin Watson <email address hidden> Fri, 13 Apr 2012 19:49:13 +0100

This is the relevant changelog entry:

ubiquity (2.10.11) precise-proposed; urgency=low

  * Work around occasional undetected download errors in apt by doing our
    own verification pass before trying to install packages (LP: #922949).
  * Update translations from Launchpad.

 -- Colin Watson <email address hidden> Fri, 13 Apr 2012 17:16:33 +0100

The installer crashed trying to reinstall previous applications, while migrating from 12.04 LTS

The Precise Pangolin has reached end of life, so this bug will not be fixed for that release