flexi_logger 0.19.6 has utc time

[incker]

flexi_logger 0.19.6 has utc time flexi_logger 0.19.5 had local time

Maybe it is not about flexi_logger crate. But chrono and time ones But how to switch back to local?

By the way trying to do UtcOffset::current_local_offset().unwrap();

Cause panic

thread 'main' panicked at 'called Result::unwrap() on an Err value: IndeterminateOffset', src/main.rs:264:58

[emabee]

0.19.6 switched form chrono to the current version of time, due to an unhandled CVE (see https://github.com/emabee/flexi_logger/issues/97).

I was hoping that this change would not impact users. What exactly did you do to encounter problems?

[incker]

0.19.5 -> 0.19.6 migration makes some breaking changes

logger
    .format(custom_format)
    .start()
    .unwrap();

[incker]

But I can not handle why I have utc time instead of local..

[emabee]

0.19.5 -> 0.19.6 migration makes some breaking changes Correct, I overlooked that, sorry.

But I can not handle why I have utc time instead of local.. What environment are you running in? Plain linux?

[incker]

What environment are you running in? Plain linux?

linux mint

[emabee]

So in your environment, will OffsetDateTime::now_local().unwrap() panic?

I'm asking because DeferredNow::now() calls OffsetDateTime::now_local().unwrap_or_else(|_| OffsetDateTime::now_utc()), thus silently ignores errors from OffsetDateTime::now_local()

[incker]

Have asked a question here also:

https://github.com/time-rs/time/discussions/389

I'm asking because DeferredNow::now() calls OffsetDateTime::now_local().unwrap_orelse(|| OffsetDateTime::now_utc()), thus silently ignores errors from OffsetDateTime::now_local()

True, that is why I have utc time

I do not know why, but 0.19.5 version with chrono had local time

[emabee]

Would you please try out if their unsound_local_offset pseudo-feature helps? It is described in the last section of https://docs.rs/time/0.3.4/time/#feature-flags.

[incker]

Yes, using RUSTFLAGS="--cfg unsound_local_offset" gives correct local time

Looks like chrono and time have same issue Chrono: https://github.com/chronotope/chrono/issues/499 Time: https://github.com/time-rs/time/issues/293

So reading changelog https://github.com/emabee/flexi_logger/blob/master/CHANGELOG.md#0196---2021-10-26 nearly nothing have changed)

But just let me know which crate you'll use in flexi_logger in future releases? time or chrono?

[emabee]

Thanks for checking. The situation is really bad, we're caught between a rock and a hard place, either we face the risk of unexpected crashes (I'm not sure how likely they are, though) or we fall back to UTC unexpectedly unless end consumers use this RUSTFLAGS="--cfg unsound_local_offset" hack.

I could (1) try factoring out the time vs chrono use, and allow choosing between them via a feature switch. Or we could (2) hope that the fix for time comes fast, which is not very likely. So let me see how (1) works out.

[incker]

Just want to say that features in crates does not have to be "mutually exclusive" I mean that adding feature time + chrono in a same time does not have cause compile errors in your crate

[incker]

Maybe it will be better solution to give choise UTC or "local time on your own risk" not sure

And time will show which crate time or chrono is more maintained and actual

[emabee]

I published 0.20.0 now, which uses the new time 0.3.5, which should allow using local time in single-threaded programs. I changed the time use such that I fetch the UTC offset only once, when flexi_logger is initialized, and then fetch only local time and add the remembered offset explicitly. This should then also work if the program switches to multi-threading after the logger is initialized,

[emabee]

Please let me know if this works for you.

[incker]

Everything works correct in 0.20.0 Time is local (not UTC) printing timezone using formatter is also correct

[incker]

Update: time is not local if not set export RUSTFLAGS="--cfg unsound_local_offset"

[emabee]

That's too bad - and not in line with what time promises, I think. Just to be sure: are you initializing flexi_logger in your main? Is it a plain main, or is it an async main?

[emabee]

For some additional information, see time-rs/time#380. According to the discussion there, time 0.3.5 uses /proc/self/task on Linux to determine if multiple threads are running.

I could add an optional feature to use chrono for determining the offset. This brings back the original situation, where, if I got it right, the application must refrain from modifying the environment in concurrent threads.

[jhpratt]

The time crate will never return UTC as the local offset if it can't determine it; it'll return the error value.

[emabee]

That's the right thing to do, definitely. I just hope that it will be possible soon to determine the local offset in a safe manner on all important platforms.

[jhpratt]

Likewise! The primary thing that isn't implemented is a way to determine the number of threads on Mac. Windows works unconditionally and Linux works when single-threaded. Other platforms are, of course, less common. I'm open to implementations, though.

[snaar]

This was mentioned earlier in this issue, but what about introducing option to always use UTC with no offset? I kinda always wanted that feature regardless.

[tdudz]

hi @emabee

it looks like latest release of time 0.3.6 supports local offset on all platforms now

[emabee]

Thanks for triggering; with version 0.22.3 I updated the dependency to time to version 0.3.7.