As per sajolida's request on
https://redmine.tails.boum.org/code/issues/14754#note-38

https://labs.riseup.net/code/issues/15282#note-10

Downloading an IUK can take a long time (up to hours) so users will likely use
Tails for other purposes meanwhile. However, the current behavior is to abruptly
disable the network as soon as the IUK is downloaded, possibly breaking work
for them.

We've stopped shipping Vidalia in March 2016.

The process that's applying the upgrade runs as a different user, so we cannot
abort it from the frontend ⇒ neither the "Cancel" button nor --auto-kill make
sense in this context. Besides, tails-install-iuk may be already running when
one clicks "Cancel", so even if we could abort that process, the resulting
system could be in a broken state.

Passing --auto-close to "zenity --pulsate" has two issues:

1. It does not make any sense at all: progress won't ever reach 100% so
   this --auto-close can't possibly happen for good reasons.

2. For some reason, it prevents the dialog from being displayed at all.
   This leads users to reboot after the download is completed, before the
   upgrade is fully applied, which can result in confusion ("I've upgraded but
   Tails keeps telling me I should upgrade") or worse, in partially upgraded
   systems that break in all kinds of interesting ways.

Regarding how this regression was introduced: I suspect that some upgrade of
zenity or IPC::Run has made zenity either close the dialog as soon as it's
extremely briefly displayed (not seen that but I can have missed it) or
decide it's not worth displaying anything (since progress has reached 100%
already, or there's no input stream connected to STDIN).

It's concerning that it took us at least months, if not years, to notice that.

The only call to this method has been removed in commit
d0e9ceaf.

They can't be run under pbuilder/sbuild and it's a shame that this prevents
us from running the rest of the test suite while building the package.

Otherwise it's broken on current Debian testing/sid.

msgfmt --check --statistics --verbose -o bn_BD.mo bn_BD.po
bn_BD.po:98: a format specification for argument 'free_memory' doesn't exist in 'msgstr'
bn_BD.po:188: a format specification for argument 'name' doesn't exist in 'msgstr'
msgfmt: found 2 fatal errors

Partially reverts aeaea596.

Unless running under FCGI, it's totally useless but adds quite some complexity
that might cause trouble.