Timely article in the Register today: http://www.theregister.co.uk/2016/07/05/linux_letting_go_32bit_builds_on_... I've been thinking about this as i686 is so often broken that I've now stopped bothering to test it in the libguestfs tests that I do on Rawhide: http://pkgs.fedoraproject.org/cgit/rpms/libguestfs.git/commit/?id=aa63cef... If you need to run an i686 virtual machine based on Rawhide, my experience is that it's more likely than not that it won't boot, and no one cares. Do we have stats for the relative proportion of i686 vs x86-64 downloads?

On Tue, Jul 05, 2016 at 10:04:03AM +0100, Peter Robinson wrote: > On Tue, Jul 5, 2016 at 9:57 AM, Richard W.M. Jones <rjones(a)redhat.com&gt; wrote: > > Timely article in the Register today: > > http://www.theregister.co.uk/2016/07/05/linux_letting_go_32bit_builds_on_... > > > > I've been thinking about this as i686 is so often broken that I've now > > stopped bothering to test it in the libguestfs tests that I do on > > Rawhide: > > > > http://pkgs.fedoraproject.org/cgit/rpms/libguestfs.git/commit/?id=aa63cef... > > > > If you need to run an i686 virtual machine based on Rawhide, my > > experience is that it's more likely than not that it won't boot, and > > no one cares. > > > > Do we have stats for the relative proportion of i686 vs x86-64 downloads? > > No really because of mirrors etc, but mirror manager stats from Feb > (FPL DevConf talk) list i686 as around 20% unique IP hits, that > doesn't take into account proxies/NAT using same IP etc. What clients are requesting from MirrorManager can also be seen here: https://admin.fedoraproject.org/mirrormanager/statistics/2016-07-05/archs

On Tue, Jul 05, 2016 at 10:04:03AM +0100, Peter Robinson wrote: > On Tue, Jul 5, 2016 at 9:57 AM, Richard W.M. Jones <rjones(a)redhat.com&gt; wrote: >> Timely article in the Register today: >> http://www.theregister.co.uk/2016/07/05/linux_letting_go_32bit_builds_on_... >> >> I've been thinking about this as i686 is so often broken that I've now >> stopped bothering to test it in the libguestfs tests that I do on >> Rawhide: >> >> http://pkgs.fedoraproject.org/cgit/rpms/libguestfs.git/commit/?id=aa63cef... >> >> If you need to run an i686 virtual machine based on Rawhide, my >> experience is that it's more likely than not that it won't boot, and >> no one cares. >> >> Do we have stats for the relative proportion of i686 vs x86-64 downloads? > > No really because of mirrors etc, but mirror manager stats from Feb > (FPL DevConf talk) list i686 as around 20% unique IP hits, that > doesn't take into account proxies/NAT using same IP etc. What clients are requesting from MirrorManager can also be seen here: https://admin.fedoraproject.org/mirrormanager/statistics/2016-07-05/archs

On 07/05/2016 11:09 AM, Adrian Reber wrote: > > On Tue, Jul 05, 2016 at 10:04:03AM +0100, Peter Robinson wrote: >> >> On Tue, Jul 5, 2016 at 9:57 AM, Richard W.M. Jones <rjones(a)redhat.com&gt; >> wrote: >>> >>> Timely article in the Register today: >>> >>> http://www.theregister.co.uk/2016/07/05/linux_letting_go_32bit_builds_on_... >>> >>> I've been thinking about this as i686 is so often broken that I've now >>> stopped bothering to test it in the libguestfs tests that I do on >>> Rawhide: >>> >>> >>> http://pkgs.fedoraproject.org/cgit/rpms/libguestfs.git/commit/?id=aa63cef... >>> >>> If you need to run an i686 virtual machine based on Rawhide, my >>> experience is that it's more likely than not that it won't boot, and >>> no one cares. >>> >>> Do we have stats for the relative proportion of i686 vs x86-64 >>> downloads? >> >> >> No really because of mirrors etc, but mirror manager stats from Feb >> (FPL DevConf talk) list i686 as around 20% unique IP hits, that >> doesn't take into account proxies/NAT using same IP etc. > > > What clients are requesting from MirrorManager can also be seen here: > > https://admin.fedoraproject.org/mirrormanager/statistics/2016-07-05/archs These statistics do not cover package downloads of i686 packages which are part of the x86_64 repositories, do they? I think the numbers are also skewed by the fact that EPEL 7 is not available for i686, which is not of direct relevance to Fedora. (The reason why it's missing is not lack of demand, but lack of a publicly available build root for Red Hat Enterprise Linux 7 on i686.)

Florian -- devel mailing list devel(a)lists.fedoraproject.org https://lists.fedoraproject.org/admin/lists/devel@lists.fedoraproject.org

>>>> On Tue, Jul 5, 2016 at 9:57 AM, Richard W.M. Jones <rjones(a)redhat.com&gt; >>>> wrote: >>>>> >>>>> Timely article in the Register today: >>>>> >>>>> http://www.theregister.co.uk/2016/07/05/linux_letting_go_32bit_builds_on_... >>>>> >>>>> I've been thinking about this as i686 is so often broken that I've now >>>>> stopped bothering to test it in the libguestfs tests that I do on >>>>> Rawhide: >>>>> >>>>> >>>>> http://pkgs.fedoraproject.org/cgit/rpms/libguestfs.git/commit/?id=aa63cef... >>>>> >>>>> If you need to run an i686 virtual machine based on Rawhide, my >>>>> experience is that it's more likely than not that it won't boot, and >>>>> no one cares. >>>>> >>>>> Do we have stats for the relative proportion of i686 vs x86-64 >>>>> downloads? >>>> >>>> >>>> No really because of mirrors etc, but mirror manager stats from Feb >>>> (FPL DevConf talk) list i686 as around 20% unique IP hits, that >>>> doesn't take into account proxies/NAT using same IP etc. >>> >>> >>> What clients are requesting from MirrorManager can also be seen here: >>> >>> https://admin.fedoraproject.org/mirrormanager/statistics/2016-07-05/archs >> >> >> These statistics do not cover package downloads of i686 packages which are >> part of the x86_64 repositories, do they? >> >> I think the numbers are also skewed by the fact that EPEL 7 is not available >> for i686, which is not of direct relevance to Fedora. (The reason why it's >> missing is not lack of demand, but lack of a publicly available build root >> for Red Hat Enterprise Linux 7 on i686.) >> > > Here is a graph for just Fedora OS from time immemorial of Fedora > using a 7 day moving average. > > https://smooge.fedorapeople.org/simple_stats/fedora-hardware-full-ma.png > > I hope this is helpful.. [I am working on ways to make this available > regularly but am up to my neck in spam accounts so don;'t expect > soon.] That doesn't seem to distinguish between "ppc" and "ppc64" at first glance, which I would think we'd want it to. Particularly since "ppc" isn't a thing that has existed in installable form for a while now.

Further, it doesn't distinguish between "ppc64" and "ppc64p7", which are actually separate architectures in the koji sense.

If you need to run an i686 virtual machine based on Rawhide, my experience is that it's more likely than not that it won't boot, and no one cares.

On Tue, Jul 5, 2016 at 11:56 AM, Florian Weimer <fweimer(a)redhat.com&gt; wrote: > On 07/05/2016 10:57 AM, Richard W.M. Jones wrote: > >> If you need to run an i686 virtual machine based on Rawhide, my >> experience is that it's more likely than not that it won't boot, and >> no one cares. > > > Well, that's independent for the state as primary vs secondary architecture. > > If we remove i686 as a primary architecture, we will not have i686 packages > in the x86_64 repository. Is this what we want? We're in the process of redefining what constitutes a secondary arch and this is part of that consideration. There's a bunch of proprietary common third party tools/apps that people rely on that still need i686 around.

On Tue, Jul 05, 2016 at 12:56:37PM +0200, Florian Weimer wrote: > On 07/05/2016 10:57 AM, Richard W.M. Jones wrote: > >> If you need to run an i686 virtual machine based on Rawhide, my >> experience is that it's more likely than not that it won't boot, and >> no one cares. > > Well, that's independent for the state as primary vs secondary architecture. > > If we remove i686 as a primary architecture, we will not have i686 > packages in the x86_64 repository. Is this what we want? IMHO that would be bad.