[Rpm-maint] [PATCH] "%post -p /sbin/ldconfig" wrongly skipped
pmatilai at redhat.com
Tue Apr 15 11:04:25 UTC 2008
On Thu, 3 Apr 2008, Michael Schroeder wrote:
> On Thu, Apr 03, 2008 at 08:39:30AM +0200, Pixel wrote:
>> still, the current situation is plain wrong... The patch i propose
>> makes the behaviour correct, at the cost of only one less ldconfig
> Actually with current glibc versions ldconfig is so fast that
> one can consider ripping out its extra handling from rpm altogether.
> Check for /var/cache/ldconfig/aux-cache if you want to know
> whether your ldconfig is using the new code.
Yup, and fast or not, rpm treating hardcoded "/some/binary" specially is
just... wrong. Not to mention that it gets the ldconfig optimization wrong
in some cases like Pixel noticed.
The ldconfig hackery in rpm is going away sooner or later, what I want to
see is some more generic mechanism for packages to queue actions to happen
at end of the transaction. Things like gtk-update-icon-cache only need
running once, so packages just need a way to flag "please run <foo> for me
before heading home." and for these, rpm would only run each unique
command once after the transaction is finished (%posttrans and all).
Several ways to accomplish that, I've just been too busy whacking
potential buffer overflows and similar fun to have a chance to really
think about it :)
- Panu -
More information about the Rpm-maint