Comment author: administrator

Hi Damien,

Damien Doligez wrote (2003-07-26T20:51:04+0200):

Float of string accepts the strings "nan", "inf" and "-inf".

Not on my machine. I guess it depends on the C standard library.

Yes. I took a look at the source. I thought there was a special case
for those tokens. It depends on the C standard library.

Ocaml(top) always display those value with a dot at the end. eg.,
[...]
Could you get rid of the excess dot?

This is now done in the working source.

Thanks. The dot has obvious lexical problems. I like to be able to
take the output from an ocaml session and use it as input. This is
handy for regression tests and other such stuff. I was happy when
3.05 displayed all floats like a lexical float.

Oh, the ocamltop lexer treats "nan" as an identifier and there happens
to be a Pervasives.nan which is a handy coincidence. I almost was
going to mention that if the tokens "infinity" and "neg_infinity" were
used instead of "inf" and "-inf" this might be considered a good
thing. Most likely this is not a good thing.

I'm not sure I understand what you mean here. It would be easy
to define "inf" in pervasives, but for "-inf" we would need to hack
the lexer or the parser. I'm not sure we want to do that.

Maybe the values that string_of_float ought to return a "nan",
"infinity" and "neg_infinity"? I suspect that once again this is
under the control of the standard C library.

-cheers

Comment author: administrator

Hi John,

Float of string accepts the strings "nan", "inf" and "-inf".

Not on my machine. I guess it depends on the C standard library.

Ocaml(top) always display those value with a dot at the end. eg.,
[...]
Could you get rid of the excess dot?

This is now done in the working source.

Oh, the ocamltop lexer treats "nan" as an identifier and there happens
to be a Pervasives.nan which is a handy coincidence. I almost was
going to mention that if the tokens "infinity" and "neg_infinity" were
used instead of "inf" and "-inf" this might be considered a good
thing. Most likely this is not a good thing.

I'm not sure I understand what you mean here. It would be easy
to define "inf" in pervasives, but for "-inf" we would need to hack
the lexer or the parser. I'm not sure we want to do that.

-- Damien

Comment author: administrator

While we're on the topic, this behavior is good:

float_of_string "nan";;

• : float = nan.

float_of_string "inf";;

• : float = inf.

float_of_string "-inf";;

• : float = -inf.

Why does entering those values only work for nan?

nan;;

• : float = nan.

inf;;

Unbound value inf

By coincidence. The strings "nan", "inf", "-inf" depend on the C
library implementation of printf for floats. Hopefully, the same C
library can read them back using strtod, which corresponds to
float_of_string in Caml. But other C libraries may chose different names.

So, we can't guarantee that these names will match those of the float
constants nan, infinity, neg_infinity defined in Pervasives. Moreover,
"-inf" would cause a slight parsing/typing problem...

However, in the spirit of "the Caml toplevel should be able to read
back what it prints", I'll change the toplevel printing of floats so
that the strings "nan", "infinity" and "neg_infinity" are used for the
corresponding float values.

• Xavier Leroy

Comment author: administrator

Display of nan and inf: fixed by DD 2003-07-26
Top-level display of nan and inf changed to allow top-level readback by XL
2003-07-29.