Parsing and formatting for the HTTP Date: header

Multiple HTTP header fields store timestamps. For example, a response created on May 15, 2015 may contain the header Date: Fri, 15 May 2015 15:34:21 GMT. Since the timestamp does not contain any timezone or leap second information, it is equvivalent to writing 1431696861 Unix time.

This crate provides two functions:

• parse to parse an HTTP datetime string to a u64 unix timestamp
• format to format a u64 unix timestamp to an IMF-fixdate

The date header is technically supposed to contain an IMF-fixdate value, but three formats actually exist in the wild. This crate attempts parsing all three when calling parse.

This is a fork of https://github.com/pyfisch/httpdate to fix some things that I found mildily annoying while using it.

Changes include:

• This crate formats into an &mut [u8], for easier use with &[std::io::IoSlice].
• This crate parses from an &[u8] rather than a &str, since that is what you normally have when parsing http headers.
• This crate is no_std.
• This crate performs no allocations.
• This crate does not panic when given timestamps after the year 9999.
• The code for this crate is simpler. (Subjective, but I think most would agree.)
• This crate is more comprehensively tested with proptest and more edge tests.
  • The original crate used fuzz, which I've never used but it seems to only test truly random inputs. This means the fuzz tests never actually exercised the parsing code since none of the inputs were ever a valid value. (The chances of randomly generating 29 bytes that happen to be a valid timestamp are effectively impossible.)
  • The fuzz/prop tests in this crate generate random valid inputs and assert that the parsing succeeds and that parsing/decoding is invariant.
  • I also included more manually-chosen edge case tests around the epoch, the maximum representable date in IMF-fixdate format, and leap years.
• This crate validates correctness of the (redundant) weekday portion of the date header analytically rather than by converting the parsed value into a SystemTime and back into a second parsed value (with the correct weekday), then checking that the two weekdays match.
• Criterion reports improvement on 3 of 4 benchmarks of around -65%, though I'm not really sure why, it seems like too big of an improvement just for not doing the SystemTime conversion.
  • The fourth benchmark also improved by a similar amount, but it appears to not work correctly in the original crate so I don't include it.
  • I didn't fork for performance reasons, so I'm not too concerned about the precise improvements.

Here's a link to pyfisch's blog post on the original crate: https://pyfisch.org/blog/http-datetime-handling/