Re: pg_upgrade test for binary compatibility of core data types

On 9/13/21 9:20 AM, Andrew Dunstan wrote:
> On 9/12/21 2:41 PM, Andrew Dunstan wrote:
>> On 9/11/21 8:51 PM, Justin Pryzby wrote:
>>> @Andrew: did you have any comment on this part ?
>>>
>>> |Subject: buildfarm xversion diff
>>> |Forking https://www.postgresql.org/message-id/20210328231433.GI15100@telsasoft.com
>>> |
>>> |I gave suggestion how to reduce the "lines of diff" metric almost to nothing,
>>> |allowing a very small "fudge factor", and which I think makes this a pretty
>>> |good metric rather than a passable one.
>>>
>> Somehow I missed that. Looks like some good suggestions. I'll
>> experiment. (Note: we can't assume the presence of sed, especially on
>> Windows).
>>
>>
> I tried with the attached patch on crake, which tests back as far as
> 9.2. Here are the diff counts from HEAD:
>
>
> andrew(at)emma:HEAD $ grep -c '^[+-]' dumpdiff-REL9_* dumpdiff-REL_1*
> dumpdiff-HEAD
> dumpdiff-REL9_2_STABLE:514
> dumpdiff-REL9_3_STABLE:169
> dumpdiff-REL9_4_STABLE:185
> dumpdiff-REL9_5_STABLE:221
> dumpdiff-REL9_6_STABLE:11
> dumpdiff-REL_10_STABLE:11
> dumpdiff-REL_11_STABLE:73
> dumpdiff-REL_12_STABLE:73
> dumpdiff-REL_13_STABLE:73
> dumpdiff-REL_14_STABLE:0
> dumpdiff-HEAD:0
>
>
> I've also attached those non-empty dumpdiff files for information, since
> they are quite small.
>
>
> There is still work to do, but this is promising. Next step: try it on
> Windows.
>
>

It appears to do the right thing on Windows. yay!

We probably need to get smarter about the heuristics, though, e.g. by
taking into account the buildfarm options and the platform. It would
also help a lot if we could make vcregress.pl honor USE_MODULE_DB.
That's on my TODO list, but it just got a lot higher priority.

cheers

andrew

--
Andrew Dunstan
EDB: https://www.enterprisedb.com