ipsplit tests fails due to endianess

[jelu]

@nicki-krizek can you fix ipsplit tests using overwrite or make overwrite handle endianess correctly?

They fail on other endian systems such as s390x:
https://bugzilla.redhat.com/show_bug.cgi?id=2323980#c7

[nicki-krizek]

Heh, I actually found a note about this issue when I wrote these tests a few years back: https://github.com/DNS-OARC/dnsjit/blob/develop/src/test/test_ipsplit.lua#L2

This doesn't matter for the ipsplit filter functionality (as noted in the docs, the bytes are written in host order).

On BE-platform, I'd recommend disabling / commenting out the related checks if needed, or just skipping the ipsplit test. IMO this is not worth fixing. It could be commented out in the upstream test suite if you insist.

[jelu]

I know it doesn't matter for ipsplit, but what is the purpose of overwrite and shouldn't it be consistent no matter the byte order?

[nicki-krizek]

The overwrite is to create a sequential 4 byte client ID for each unique input address. This is then used in shotgun which keeps track of each of those clients and has to have a memory structure allocated for each of them. This client ID then basically serves as an index into the allocated array. The reason it's done in ipsplit is because it's more efficient to do this when pre-processing the data, rather than trying to do this mapping at shotgun runtime.

In practice, the byte order doesn't matter as long as you don't attempt to use the pre-processed data from BE platform to run shotgun on LE platform.

It could be fixed to always be consistent. The other part that does the decoding which would have to be adjusted is in https://gitlab.nic.cz/knot/shotgun/-/blob/master/replay/dnssim/src/output/dnssim.c#L176

[jelu]

OK. I'll fix the test and mark the overwrite function as non-portable (once I'm back).