Implement FromStr for UDecimal/Decimal.

[jeandudey]

Solves #140

[apoelstra]

Can you add a fuzz test that tries to parse a decimal, and when it succeeds tries to reserialize it (and check that it roundtrips correctly)?

[jeandudey]

Done

[apoelstra]

Nice, thanks! Unfortunately it looks like it failed trying to roundtrip the number '6' (which I kinda expected to happen ... that should obviously parse as a 6, but then I'd expect it to serialize as 6.00000000, so I guess the fuzz test needs to be more accepting... perhaps it should parse, then serialize, then parse again, and check that both parses gave the same Decimal / UDecimal?)

Also, not a big deal, but in future I'd appreciate if fuzz-tests would appear in their own commit, because they're generally independent of other changes and can be reviewed independently.

[jeandudey]

Done 👍, fuzz tests are now in their own commit.

[jeandudey]

Build is failing because rust-bitcoinconsensus doesn't build on 1.14.0, the cc crate was updated and doesn't build on 1.14.0 anymore.

[jeandudey]

Tracking issue: alexcrichton/cc-rs#336

[apoelstra]

Looks like there is also a failing fuzz test case (hex 2d3735363539383732343037353635393837323434 which you should be able to copy into your fuzz binary's unit test, then run cargo test in the fuzz/ directory to reproduce)

[jeandudey]

That hex string is the number -75659872407565987244 (as a string), it's too large to fit in the mantissa field. The Decimal/UDecimal parsing functions now return Error::TooBig when an overflow occurs.

[jeandudey]

Ok, don't know why tests are failing I'm feeling stupid right now :P, I'm going to sleep and check that tomorrow.

[apoelstra]

I don't mind the churn, but you may find life easier if you run ./travis-fuzz.sh locally before pushing (and maybe edit the file to do more iterations).

I'm glad we're finding this stuff though ... I expect all of these issues existed in the old code, we just never sussed them out.

[jeandudey]

@apoelstra I can't do fuzzing right now on my PC (some reasons, not related to this), that's why I'm pushing and fixing.

[jeandudey]

Also the tests are failing with the bytes 302e333434343034333334343333303333303333302039333330203b3333 (a very large fp number), but crashes when dec.to_string is called, I found out that this line in the Display implementation is causing this error (an overflow panic):

        let ten = 10i64.pow(self.exponent as u32);

Not sure how to handle this though.

[apoelstra]

I think we should just forbid exponents larger than 18. I guess this could cause issues for Dogecoin or something but we don't support anything like that now.

[jeandudey]

lol, the build is finally green

[apoelstra]

Thanks so much! Seems like most of the bugs were latent and had been hanging around since 2014 so it's awesome to finally clear them out.