This is the mail archive of the
mailing list for the Cygwin project.
Re: Packages that change without incrementing the version/release
On Sat, Nov 20, 2004 at 05:10:53PM -0000, Max Bowsher wrote:
>Christopher Faylor wrote:
>>On Sat, Nov 20, 2004 at 03:14:52PM -0000, Max Bowsher wrote:
>>>And one important consequence of this is that packagers must *never*
>>>files without changing the release number, once they have gone to the
>>>mirrors - because setup.exe will be telling users that something odd has
>>>happened - not just silently ignoring the inconsistency.
>>I don't agree with this. As you noticed, I recently repackaged some of
>>the xorg stuff to test a theory about why X installs were failing. If
>>I'd bumped the version number then it would have been an indication of a
>>new version of the software -- which was not the case.
>>In this case, setup.ini correctly represented files available for the
>>distribution and the files on the users disk were not wrong. I don't
>>think that setup should complain about this unless the user specifically
>>requests a redownload or reinstall.
>It may not have been a new version of the software, but it was a new
>version of the package.
>But, philosophical issues aside - how can setup tell the difference between
>a corrupt package and one that has been 'replaced', and so no longer
>corresponds to the current info in setup.ini?
setup.ini is always correct. If the package changed in setup.ini, that
should not trigger a download unless the user specifically asks for it.
I don't see any reason to check md5sums for packages that aren't being
installed. When a package is specifically requested for reinstallation
an informational message could be displayed indicating that the package