bad outputs for f2s

[timotheecour]

I know it's still WIP, but just for reference:

  template test2(x, s) =
    let s2 = f2s(x)
    check s2 == s
  test2 0.0, "0.0" # current: 0e0
  test2 1.0, "1.0" # current: 1e0
  test2 1.3, "1.3" # current: 1.3e0
  test2 0.12345678901234567890123456789, "0.12345678901234568" # should truncate to same as python3, I think; current: 1.2345679e-1 which doesn't have enough digits

• some of these are real errors (not roundtrip safe), like 0.12345678901234567890123456789
• some of these are cosmetic errors; eg e0 should just be stripped
• the .0 shouldn't be stripped before an exponent

that's for consistency with almost all languages; if needed i can do some more research to double check what's printed in other langs

[disruptek]

This is not a printf format code; it's for stringifying a float.

[disruptek]

It's easiest if you just PR the tests; you can do that right from the web interface.

[timotheecour]

It's easiest if you just PR the tests; you can do that right from the web interface.

I can add a PR with failing tests, sure; obviously would have to use when/if false since they'd fail...

This is not a printf format code; it's for stringifying a float.

well in my mind this repo should eventually (looks like this could be soon given what you already have) replace system.addFloat, which is needed to properly close nim-lang/Nim#13196 (among other benefits) (and araq seems to agree with that direction see nim-lang/Nim#13364 (comment))
Obviously there are still things to figure out for this to work (and be importable in system without bloating dependencies; which is feasible via a magic IMO), but if that's (one of) the goals, f2s should be suitable for that. It doesn't have to be the only format option though, and I think f2s could get a format string (with maybe few initial supported options, subset of what printf options allow, to grow as needed over time) eg:

f2s(result: var string, x: float, fmt: string) = ...
# or maybe via static string if benchmarking shows it can affect performance in any way
f2s(result: var string, x: float, fmt: static string) = ...

doAssert f2s(1.2) == "1.2" # implies default fmt = "%g"
doAssert f2s(0.12, "%f") == "1.2e-1"
doAssert f2s(0.12, "%e") == "0.12"
doAssert f2s(0.12, "%g") == "0.12" # Use the shortest representation: %e or %f

[disruptek]

If they don't fail, then the tests aren't useful.

There's a printf component to Ryu that we can port for printf-like behavior that is consistent with Ryu itself. This would be suitable for adoption in addFloat, etc.