ldb test test_guid_indexed_v1_db fails on ppc64le

[lslebodn at fedoraproject.org]

On (27/08/19 06:27), Andrew Bartlett wrote:
>On Mon, 2019-08-26 at 13:29 +0200, Lukas Slebodnik via samba-technical
>wrote:
>> ehlo,
>> 
>> The repack unit test failed just only on ppc64le with ldb-2.0.5
>> Other architectures are ok
>> https://koji.fedoraproject.org/koji/taskinfo?taskID=37286196
>> 
>https://kojipkgs.fedoraproject.org//work/tasks/6219/37286219/build.log
>> 
>> BTW it would be good to suppress error messages
>> "MDB_INVALID: File is not an LMDB file".
>> I could see them also on x86_64
>> 
>
>Sadly LMDB is not portable across endian or word length boundaries. 
>
>The only way to fix this is to store a LE version of the LDB file and
>detect which to use.  We don't hit the word length issue because we
>lock this to 64 bit only so far.
>

IIUC:
* x86_64  is little endian and 64 bit
* ppc64le is little endian and 64 bit

So I do not see any difference and therefore I am confused here.
And there was not any failure on s390x which is big endian.

>Do you think you could have a go at that?
>

Is your comment related to the error message "MDB_INVALID: File is not an LMDB
file"  or to the failure in the test?

I am sorry but I am lost.
Moreover, the error " (1, 'Failure during prepare_write): Corrupt database -> Operations error')"
and my knowledge of ldb/tdb/lmdb is not sufficient for debugging
I already gave up after an hour. I am sorry I do not have days for that,

But I have an access to the ppc64le machine and I can help
with providing debug output with some trace messages if it helps.

LS