Front page | perl.perl5.porters |
Postings from November 2016
Re: hv.h hek definition
Thread Previous
|
Thread Next
From:
demerphq
Date:
November 3, 2016 07:39
Subject:
Re: hv.h hek definition
Message ID:
CANgJU+X5M4VuGyVcVojEJrxkvH7mo5N5GE6ReiGm3UU7i4WUrQ@mail.gmail.com
On 3 November 2016 at 08:38, demerphq <demerphq@gmail.com> wrote:
> On 3 November 2016 at 01:21, Tony Cook <tony@develop-help.com> wrote:
>> On Mon, Oct 24, 2016 at 05:45:22PM +0200, demerphq wrote:
>>> Well I have pushed the patch. I dont think it should matter, and if we
>>> find it does then we can tweak it further, or revert if necessary.
>>
>> Unfortunately this broke the build on HP-UX.
>>
>> SV keys are stored as pointers in hek_key:
>>
>> #define HeKEY_sv(he) (*(SV**)HeKEY(he))
>>
>> #define HeKEY(he) HEK_KEY(HeKEY_hek(he))
>>
>> #define HEK_KEY(hek) (hek)->hek_key
>>
>> So setting/fetching a SV key produced a bus error early in the build
>> (when running configpm) since PA-RISC requires aligned access.
>>
>> Reverting d3148f758506efd28325dfd8e1b698385133f0cd fixed the problem,
>> which I've pushed to blead for now.
>
> Er, reverting under this type of situation is not that helpful, even
> if gets things building again. If we want to try another iteration of
> this issue then we will just have to do another "shot in the dark" and
> hope it doesnt break HP-UX.
>
> Better I think would be if you could please try unreverting and change
> the hek_flags field to U32, so we at least know if that fixes the
> alignment problem you have just revealed.
Specifically what alignment is required. Does it have to be an 8 byte
line, or is a 4 byte line ok?
Yves
--
perl -Mre=debug -e "/just|another|perl|hacker/"
Thread Previous
|
Thread Next