Floating point conversion tests

[la-wu]

This PR adds floating point tests for floatDec and doubleDec. It expects implementations to match base show functionality wrt rounding and (non) shortest-path.

These are based on the original C tests with modifications to match show.

[Bodigrim]

Awesome!

[sjakobi]

It expects implementations to match base show functionality wrt rounding and (non) shortest-path.

How about a property test that compares the floatDec and doubleDec output to show?

BTW, where does the requirement to match show come from? It sounds like a reasonable idea, but is there an explicit promise that we'll do that or was there a discussion or something like that?

[la-wu]

How about a property test that compares the floatDec and doubleDec output to show?

Do you mean e.g

bytestring/tests/builder/Data/ByteString/Builder/Tests.hs

Line 571 in 5ab40a8

, testBuilderConstr "floatDec" dec_list floatDec

This test tends to not cover enough cases to exposes differences reliably but is a good statement of expectation.

BTW, where does the requirement to match show come from? It sounds like a reasonable idea, but is there an explicit promise that we'll do that or was there a discussion or something like that?

There was some discussion in the original C PR #222 (comment) and mostly the idea was that it would be 'quite unexpected to diverge from show'.

[Bodigrim]

BTW, where does the requirement to match show come from?

Diverging from it means a breaking change, and of a nasty nature. I think it's desitable to keep show and both ByteString and Text builders in line with each other.

Ideally I'd love to see show fixed in base, but it's likely to take ages, so the shortest path to get ryu merged is to mimic all infelicities.

[sjakobi]

How about a property test that compares the floatDec and doubleDec output to show?

Do you mean e.g

bytestring/tests/builder/Data/ByteString/Builder/Tests.hs

Line 571 in 5ab40a8

, testBuilderConstr "floatDec" dec_list floatDec

Thanks for making me aware (again) of this test! :)

BTW, where does the requirement to match show come from?

Diverging from it means a breaking change, and of a nasty nature. I think it's desitable to keep show and both ByteString and Text builders in line with each other.

Agreed. It might make sense to explicitly say that floatDec and doubleDec match the show output in their Haddocks. But this PR probably isn't the right place for this.

[Bodigrim]

Thanks @Lumaere!