Hi VN, Serg, all !
Narayanan wrote:
> Thank you for the patient comments and guidance for this issue
This is something you learn in the build team: Become insisting ;-)
>
> Joerg Bruehe wrote:
>> Hi,
>>
>> [[...]]
>>
>> Sergei Golubchik wrote:
>>> Hi!
>>>
>>> On Aug 20, Joerg Bruehe wrote:
>>>> Hi Sergei, all !
>>>>
>>>>
>>>> Sergei Golubchik wrote:
>>>>> Hi!
>>>>> On Aug 18, Narayanan wrote:
>>>>>> [[...]]
>>
>> [[...]]
>>
>>>
>>>> 3) If this check is done in release builds, in all configurations,
>>>> it is
>>>> just a waste of time and disk space.
>>>
>>> Waste of how much time and how much space ? It's far less than 1 per
>>> cent of either.
>>
>> I have not taken any time measurements. VN, do you have any ?
> I did a "time make abi_check", below are three time measurements
>
> real 0m0.235s
> user 0m0.052s
> sys 0m0.004s
>
> real 0m0.164s
> user 0m0.052s
> sys 0m0.016s
>
> real 0m0.166s
> user 0m0.056s
> sys 0m0.024s
Ok, I had assumed it to take much longer.
What I think would be even more informative is to compare that to a full
build time (compile + link, no test), as your machine might be extremely
fast, but even so I expect it will be tolerable.
Seems I was too pessimistic in my fears.
>
>>
>>>
>>> [[...]]
>>
>> In a release build, we do it six times (for the different
>> configurations).
>>
>> I would like to know the build times (compile + link, no tests) both
>> with and without the ABI check.
>> The answer should tell how important it is to avoid repeating the
>> check when building different configurations of the same version on
>> the same platform.
>
> If I have done the right things in finding the time, I think this is not
> much.
See above.
>
>
>>
>> [[...]]
>>>
>>> Right, but VN has this WL pushed in the team tree, and it's tested on
>>> all pushbuild platforms already. As far as I remember this includes
>>> Solaris and OS X with gcc.
>>
>> It is in the team tree before having two reviews ? No comment.
>
> I thought Serg had super powers, and that his review was enough.
He has, and even though I may have a different opinion than he I will
never doubt his competence.
It is just weird, then, that I was asked to review this in addition.
Never mind, not your fault.
>
>>
>> The stronger you argue to "run it on as many platforms as possible",
>> the more important it gets to make the "diff" binary configurable.
>> Currently, "diff" output for the same file difference in the test
>> suite differs by platform. (A weird sentence, but you get the meaning.)
>
> But I have made it uniform across all platforms in pushbuild. I have
> been struggling to get this working
> on all platforms for the past month.
From a build team point, "all platforms" includes several non-Linux
non-gcc platforms, which typically are the ones causing more trouble
(because Linux is the typical developer platform).
We don't have enough of these machines to use them in pushbuild, so they
lack coverage, and any specific problems show up in the release builds only.
Currently, Solaris belongs to that class, I hope that will change soon
(= we will use more Solaris boxes in pushbuild).
>
>
>>
>>>
>>>> [[...]]
>>>
>>> A perceived possibility of a hypothetical compiler that supports
>>> -E -dI -nostdinc ? I'm afraid, it's over-engineering, adding the
>>> complexity without necessity.
>>
>> Same argument as above: If you would like to cover all platforms, then
>> include the mechanism now, so that we can add other compilers if we
>> detect the need and the possibility (by adding a branch then, not
>> changing the general approach).
>
> We can always change the autoconf rule to do what we want, can't we?
Sure we can.
It just is easier to have a general mechanism from the very start, so
that all we need to do is add a branch, than to start with a fixed
command (switched off on the platforms that don't fit) and later change
that approach to support choice.
See the other subthread for my final summary - I will stop objecting.
Regards,
Jörg
--
Joerg Bruehe, MySQL Build Team, joerg@stripped (+49 30) 417 01 487
Sun Microsystems GmbH, Sonnenallee 1, D-85551 Kirchheim-Heimstetten
Geschaeftsfuehrer: Thomas Schroeder, Wolfgang Engels, Dr. Roland Boemer
Vorsitzender des Aufsichtsrates: Martin Haering Muenchen: HRB161028
