Replace humantime crate with jiff

[oherrala]

Rationale for this change

The crate humantime appears to be unmaintained. There's open PR in RustSec's advisory-db about this: rustsec/advisory-db#2249

The crates clap and env_logger have already made the switch from humantime to jiff:

• chore(deps): Switch humantime to jiff clap-rs/clap#5944
• chore(deps): Switch humantime to jiff rust-cli/env_logger#352

Since new version of env_logger is already published the object_store crate is now second largest user of humantime crate by crates.io downloads counts and the largest user after new version of clap is released.

What changes are included in this PR?

Replace humantime crate with jiff. The functionality should stay the same.

Are there any user-facing changes?

There should be none.

Fixes #7264

[tustvold]

Thank you, as jiff isn't 100% compatible with humantime I think this constitutes a breaking change and will therefore have to wait for the next breaking release in likely ~6 months time...

That being said now that an advisory has been published we might want to just push ahead with this to avoid undue angst... I'll let other maintainers weigh in

Edit: tbh given how small humantime actually is we probably could just vend the parser locally as a temporary measure... AFAICT there isn't actually an issue with the code

[BurntSushi]

AFAICT there isn't actually an issue with the code

It depends on what you're doing, but humantime will happily treat 1 month as 30.44 days, regardless of which month it is. Similarly for other variable length units, like years and days. The discussion in the Jiff docs shows examples, and specifically how Jiff avoids this pitfall.

With that said, in this PR, the impl Parse for Duration would previously accept 1month (translating it to 30.44 24-hour days), but the implementation in this PR will reject anything with non-zero units greater than hours. So 1 day would be rejected here, for example. In order to parse calendar units, you need to use Span. And to resolve that to a real duration in time, you need a relative datetime. (Which is more annoying and presumably why humantime decided to implicitly convert calendar units to their "average" value, despite it producing unintuitive results.)

[tustvold]

Given these are timeouts I suspect few users are configuring anything larger than an hour. Are there any differences when parsing such?

[BurntSushi]

Given these are timeouts I suspect few users are configuring anything larger than an hour. Are there any differences when parsing such?

That's a good point and a good question. The major difference is definitely that parsing into SignedDuration will reject any duration with non-zero year/month/week/day units. But aside from that, I think it depends on which direction of compatibility you're interested in. If it's just a matter of what humantime supports in terms of parsing, we can take a look at its implementation:

        let (mut sec, nsec) = match &self.src[start..end] {
            "nanos" | "nsec" | "ns" => (0u64, n),
            "usec" | "us" => (0u64, n.mul(1000)?),
            "millis" | "msec" | "ms" => (0u64, n.mul(1_000_000)?),
            "seconds" | "second" | "secs" | "sec" | "s" => (n, 0),
            "minutes" | "minute" | "min" | "mins" | "m"
            => (n.mul(60)?, 0),
            "hours" | "hour" | "hr" | "hrs" | "h" => (n.mul(3600)?, 0),
            "days" | "day" | "d" => (n.mul(86400)?, 0),
            "weeks" | "week" | "w" => (n.mul(86400*7)?, 0),
            "months" | "month" | "M" => (n.mul(2_630_016)?, 0), // 30.44d
            "years" | "year" | "y" => (n.mul(31_557_600)?, 0), // 365.25d
            _ => {
                return Err(Error::UnknownUnit {
                    start, end,
                    unit: self.src[start..end].to_string(),
                    value: n,
                });
            }
        };

Jiff's "friendly" duration format has a grammar, and indeed, for hour/minute/second/millisecond/microsecond/nanosecond units, Jiff can parse all of the units that humantime can parse (plus more, including things like µs). In general, in terms of things that humantime can parse but Jiff can't, I believe the only thing is M for months. Jiff requires at least mo or mos.

The other direction is different, that is, can humantime parse what Jiff outputs. I'm not sure if you need that form of compatibility, but if you do, Jiff has a special HumanTime mode for it.

[alamb]

That being said now that an advisory has been published we might want to just push ahead with this to avoid undue angst... I'll let other maintainers weigh in

I think this is a good change and that the removal of a RUSTSEC is worth any potential issue with parsing questionable durations.

Given @BurntSushi 's long history of maintaining core crates in the Rust ecosystem and the current adoption of jiff I think this is a good choice

https://crates.io/crates/jiff

BTW if/when this PR is merged, I think we can/should consider backporting this change to the object_store 0.11 line so others can pick up the change before 0.12.0 percolates through the ecosystem

[yhx-12243]

humantime seems resurrected now (1 hour ago).