[OE-core] gcc 5.2 failures

[Bruce Ashfield]

On 15-07-27 09:50 AM, Paul Eggleton wrote:
> On Monday 27 July 2015 09:31:01 Bruce Ashfield wrote:
>> On 15-07-27 09:20 AM, Bruce Ashfield wrote:
>>> On 15-07-27 05:30 AM, Richard Purdie wrote:
>>>> I've run a gcc 5.2 test build on the autobuilder:
>>>>
>>>> http://errors.yoctoproject.org/Errors/Search/?items=10&query=3628c3c06fa4
>>>> 195003ac655bcc791acfac775173&limit=50
>>>>
>>>>
>>>> 41 errors (with a few more pending).
>>>>
>>>> The good news is that if we tweak the security flags, the poky-lsb gcc,
>>>> elfutils, coreutils and iptables issues can be removed and I have a
>>>> patch for this. This leaves:
>>>>
>>>> 3.14 kernel failures for edgerouter, genericx86-64, qemuarm, beaglebone,
>>>> mpc8315e-rdb
>>>
>>> Gah. I had all these building with 5.1 .. chasing gcc is a pain
>>> with this older kernel.
>>>
>>>> openssl issue for p1022ds
>>>>
>>>> u-boot on imx28evk, p1022ds, mpc8315e-rdb
>>>>
>>>> xf86-video-imxfb-vivante on imx6qsabresd
>>>>
>>>> linux-imx issue on imx53qsb
>>>>
>>>> Some kind of "random" qemu runtime issue (4 cases).
>>>>
>>>> At this point I think we likely need to enter bugs into the bugzilla for
>>>> each of these. If we want to switch 1.9 to use this (which I think is
>>>> desirable), we need to get this fixed as a priority.
>>>>
>>>> Bruce: How do you want to handle the 3.14 issues? Switch to 4.1? or fix
>>>> 3.14?
>>>
>>> Now that 4.1 is in place, and I can't really see a large user base that
>>> needs gcc 5.x with the linux-yocto 3.14 kernel (other folks using
>>> master with their own kernel's will obviously have to deal with the
>>> issue in their trees) .. join that with the fact that we need to update
>>> all the reference boards to 4.1 anyway, my suggestion is that we open
>>> bugs for the h/w reference updates (and I'll get the appropriate Wind
>>> River eyes on them) and walk away from burning more cycles on gcc 5.x
>>> and the 3.14 kernel.
>>
>> And of course, I remembered that we updated all the reference
>> boards to 3.19, so if 3.19 builds with gcc 5.2 (it may not), then
>> all we need to do is move the lsb reference to 4.1 and we'll also
>> be free of backporting :)
>
> It would be nice if we were able to point to a set of fixes that people can
> apply onto their older kernels in the migration guide for 1.9 though - I
> guarantee we'll get people asking...

There's unfortunately no clean set of fixes, the mips one for
example, is a bit of a knarly backport in the assembly code.

If it was easy to point them out, we would have just done all
the ports :)

Bruce

>
> Cheers,
> Paul
>