On Thu, 29 Mar 2018 15:32:06 +0200, Stephen Gallagher
<sgallagh(a)redhat.com> wrote:
On Thu, Mar 29, 2018 at 9:16 AM Stephen Gallagher <sgallagh(a)redhat.com>
wrote:
> On Thu, Mar 29, 2018 at 8:43 AM Kevin Kofler <kevin.kofler(a)chello.at>
> wrote:
>> Hi,
>>
>> on #fedora-kde, we discussed a huge increase to the size of the KDE
>> live
>> image:
>> F27 KDE GA: 1656752111 bytes
>> F27 KDE Respin: 1744830464 bytes
>> F28 KDE Beta: 2069889024 bytes
>>
>> It turns out the LXQt live image is also hit:
>> F27 LXQt GA: < 1 GiB
>> F28 LXQt Beta: 1.4 GiB
>> and I am pretty sure that this is a global issue affecting ALL the live
>> images.
>>
>> While there is certainly more than one cause (e.g., the difference
>> between
>> F27 KDE GA and the F27 KDE Respin must be caused by package updates
>> and/or
>> added dependencies), it is striking that between a recent F27 KDE
>> Respin
>> (which has several of the KDE package updates that are also in the F28
>> Beta)
>> and the F28 Beta, there is a size increase of more than 300 MB!
>>
>> Therefore, my question:
>> Is this size increase caused by Annobin?
>>
>> If not, then WHAT causes this size increase? And can the offending
>> change(s)
>> be reverted in time for F28 Final?
>>
>> If yes, then IMHO it is time to enact the contingency plan, i.e.:
>> 1. drop the annobin requirement from redhat-rpm-config, AND
>> 2. perform an emergency mass rebuild to actually get rid of the bloat.
>>
>> A global live image size increase of 19%-40% is just not acceptable.
>>
>> Kevin Kofler
>
>
> Have you done a comparison between individual package sizes and count?
> I wonder if the problem is that some subset of >>critical packages
> gained a whole lot of dependencies (intentionally or not) or if we're
> actually seeing that built >>binaries are suddenly carrying a huge
> additional amount of content. Or even if for some reason the binaries
> aren't being >>stripped properly.
I just checked several packages that compiled the same sources on F26
and F28:
* sscg: 4.31% increase in size
* nodejs 8: 22% *decrease* in size
* libmodulemd: 12% *decrease* in size
* sssd-common: 2.58% increase in size
* boost-system: 1.00% increase in size
Granted, this is just a spot-check, but it seems to me like (on the
whole) improvements in the compiler are at least >balancing out the
increase caused by annobin. It definitely doesn't indicate a 19-40%
increase across the board. So I'd >guess it's more likely to be
dependency bloat or one or two core packages growing significantly.
Just for fun, I wrote a patch comparing only _unrebased_ (probably just
rebuilt) packages between F27 and F28.
For compiled packages, the size difference is just a bunch of percent.
Some packages grew considerably, but these are mostly fonts
(google-noto-sans-fonts by 1012%).
The diff is attached. Order of data is: percentage, package name, f27 RPM
size, f28 RPM size.
Martin