Ferenc Wagner writes: $ ghci Prelude> :m Ratio Ratio> [1%2..10%2]

[1 % 2,3 % 2,5 % 2,7 % 2,9 % 2,11 % 2]

Hmmmm, the CVS copy of Hugs seems to suffer from a different problem: Prelude> [0.5,1.5..5.5]::[Rational] [0 % 1,1 % 1,2 % 1,3 % 1,4 % 1,5 % 1] I'm expecting to see: [1 % 2,3 % 2,5 % 2,7 % 2,9 % 2,11 % 2] -- Alastair Reid

Alastair Reid:

Ferenc Wagner writes:

Hmmmm, the CVS copy of Hugs seems to suffer from a different problem:

Prelude> [0.5,1.5..5.5]::[Rational] [0 % 1,1 % 1,2 % 1,3 % 1,4 % 1,5 % 1]

I'm expecting to see:

[1 % 2,3 % 2,5 % 2,7 % 2,9 % 2,11 % 2]

Rationals in Hugs were always a bit obscure. What do you think, what is the Rational form of 2.3 ? (GHCi says 23/10). The answer is: 2589569785738035 % 1125899906842624 (Old Hugs, Feb. 2001) If you look at the Prelude, you will see that the algorithms used for rationals are not always a rocket science. I replaced (for myself) that stuff by the continued fraction expansions which are fast and give decent results. The rational arithmetics can also be optimised by using algorithms in the 2nd volume of Knuth (the favourite book of Ralf Hinze...) I found similar bugs in sequences as above already (if I am not mistaken) about 7 - 8 years ago, when we discussed a bit the usage of Haskell to some numerics. But nobody really cared about it, and it seems that some small but nasty insects are still alive. Jerzy Karczmarczuk

Jerzy Karczmarczuk wrote (on 22-10-02 13:05 +0200):

What do you think, what is the Rational form of 2.3 ? (GHCi says 23/10).

The answer is:

2589569785738035 % 1125899906842624

Er, why? Because 2.3 is not representable using a double precision float or something? -- Frank

In haskell-cafe, Alastair Reid wrote:

Jerzy Karczmarczuk writes:

...

Rationals in Hugs were always a bit obscure. What do you think, what is the Rational form of 2.3 ? (GHCi says 23/10).

...

The answer is:

...

2589569785738035 % 1125899906842624

...

(Old Hugs, Feb. 2001)

I'm afraid the new release won't fix this.

Once the release is out the door and things settle down a bit (both in Hugs and in my personal life), I'd like to cleanup the Hugs' internals which have gotten quite confused by layer upon layer of backward compatability code. Practical benefits I hope for are:

- Make Float mean 'C float' and Double mean 'C double'. Most of the code is actually in Hugs already but it was disabled because of some long-irrelevant issue involving the foreign function interface.

So it seems that double-precision can (and should) be re-enabled. Alastair (in another place) counsels caution. But it's my impression that several people are using a version of Hugs with USE_DOUBLE_PRECISION set to 1 (Jerzy clearly is) with no problems, so this should be safe. It also passes the Hugs test suite (modulo the extra precision). Any testimonials?