encoded time.Time before Unix Epoch is not decoded correctly about go HOT 7 CLOSED

Thanks Zond.

This may be due to possible overflow issues, because Go's time.Time uses Jan 1, Year 1 as epoch, while binc/msgpack/java/python/etc use Jan 1, 1970 as epoch.

I wouldn't be able to look into it till next week at the earliest. If you can, please look at codec/time.go (a 100-line file which included encodeTime and decodeTime functions) to start debugging.

Thanks.

from go.

#6 created for this issue.

from go.

Hi Zond,

I looked into this issue.

I looked at your pull request solution also. Part of the reason for pruning the sign ext is to make storage of time super-compact. With your pull request, we lost that. So I tried to look at the solution while maintaining the compact storage for time.Time.

The problem as you figured out was that we didn't handle negative tsecs component during decoding. Thus, times before Unix EPOCH was broken. The fix is to pre-populate sign-extend the byte array with 0xff if a negative time.
///////////////// UGORJI NOTE: FIRST IF BLOCK IN DECODETIME
if bd&(1<<7) != 0 {
var btmp [8]byte
n = ((bd >> 2) & 0x7) + 1
i2 = i + n
copy(btmp[8-n:], bs[i:i2])
////////////// UGORJI NOTE: ADD BLOCK BELOW
//if first bit of bs[i] is set, then fill btmp[0..8-n] with 0xff (ie sign extend it)
if bs[i] & (1 << 7) != 0 {
var timeBs0xff = []byte{0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff}
copy(btmp[0:8-n], timeBs0xff)
////// UGORJI NOTE: OR USE LOOP BELOW. STILL HAVE TO TEST WHICH IS FASTER.
//for j,k := byte(0), 8-n; j < k; j++ { btmp[j] = 0xff }
}

I've tested this and it works. If you can verify that it works on your use-cases, that would be great. I'm not comfortable releasing it with just my limited testing, so need you to co-sign off on it, or wait for a few days when I can focus work on it.

from go.

Yeah, I realized that my solution wasn't the most compact, but I am too inexperienced when it comes to bit fiddling with signed numbers to make anything else work in the time I had :/

If my test (just copy it temporarily from https://github.com/zond/go/blob/master/codec/time_test.go if you don't want it in the repo) runs I should be fine.

from go.

Yes, I understand.

I tested with the code you put in the issue description, which is basically same as your time_test.go. So I think it's fine.

I'd do a push within the next few days. I want to tackle the nil slices/maps/pointers-to-structs thing which you alluded to in the other issue and push solutions to both together.

Thanks for your patience.

from go.

Great!

Thanks so much for looking into this - I really enjoy the possibility to have a compact and fast codec that doesn't assume stream processing (like gob).

I'm looking forward to the other solution as well, but don't sweat it - until you have your commits done I will just use github.com/zond/go/codec - the beauty of go dependency management is sweet :)

from go.

This should be fixed now with the latest commit.

Thanks.

from go.