Broken timestamps from Cisco NX-OS 8.2(2) (8 byte timestamps)

[tik-stbuehler]

Hi,

f7a4bc566a6943586aa1426133d86c660fa1f3aa was supposed to fix #80; but

• the fix is not endian-safe: it will pick the second dword in native endianess, whatever that happens to be
• the zero byte padding can occur from both sides: usually the zero bytes are simply following the "real" data, but sometimes the four zero bytes come first. maybe simply adding the two dwords works.

[phaag]

Hmm .. I don't see, what you mean by "simply adding the two dwords works." Can you please elaborate a bit more on this?

[tik-stbuehler]

If one of the dwords is always zero and you want to pick the other (non-zero) value, adding them should do it. (or bitwise or. or bitwise xor.)

A perhaps more stable solution would be preferring one of the dwords and only look at the other if the first was zero (but the preference should be stable across endianess).

[phaag]

nfdump-1.7 uses a rewritten sequencer for byte loading. This should no longer be an issue.