Allow converting a Timestamp back into a snowflake ID

[inklesspen]

According to the API docs, for timestamps within the legal range, you can convert them into a snowflake ID as (timestamp_ms - DISCORD_EPOCH) << 22. This is useful for pagination; you can search for messages before or after a specific time.

In my project I've currently done a very half-assed version of this:

trait FromDateTime<Tz: TimeZone> {
    fn from_datetime(value: DateTime<Tz>) -> Result<Self>
    where
        Self: Sized;
}

/// Discord's epoch starts at "2015-01-01T00:00:00+00:00"
const DISCORD_EPOCH: u64 = 1_420_070_400_000;
const MAX_DISCORD_TIMESTAMP: u64 = 5_818_116_911_103;

impl<Tz: TimeZone> FromDateTime<Tz> for NonZeroU64 {
    fn from_datetime(value: DateTime<Tz>) -> Result<Self>
    where
        Self: Sized,
    {
        let unix_millis: u64 = value.timestamp_millis().try_into()?;
        if unix_millis <= DISCORD_EPOCH || unix_millis > MAX_DISCORD_TIMESTAMP {
            anyhow::bail!("Unable to convert this timestamp to a snowflake");
        }
        let snowflake = (unix_millis - DISCORD_EPOCH) << 22;
        Ok(NonZeroU64::new(snowflake).expect("already checked"))
    }
}
impl<Tz: TimeZone> FromDateTime<Tz> for AuditLogEntryId {
    fn from_datetime(value: DateTime<Tz>) -> Result<Self>
    where
        Self: Sized,
    {
        Ok(NonZeroU64::from_datetime(value)?.into())
    }
}

Additionally, you can pass 0 as the snowflake for "the beginning of time"; however this would seem to pose a larger problem due to the snowflake ID types relying on NonZeroU64.

[inklesspen]

Since the next branch no longer uses NonZeroU64, passing 0 for "the beginning of time" is no longer a problem there.

If anyone would be willing to convert this discussion into an issue, I'd be happy to send a PR for it.

[arqunis]

Created #3485