[scc-dev] C99 standard for a particular floating point format

From: Tim Kelly <gtkelly_at_dialectronics.com>
Date: Wed, 08 Mar 2023 08:16:53 -0500

Perhaps someone on the list with more access to the C99 standard and/or
experience can enlighten me on how

double ddb = 0x400921FB54442D18;

should be interpreted during parsing. At first glance, one would be
inclined to say the above is an integer, but in reading the draft of
C99, I can not rule out that it should be interpreted as an IEE-754
representation of a real number. If it was shorter, I would concur that
it is an integer, but it is of the correct length for a double.

Although not common, it might be possible the programmer already knows
the floating point representation of the number in question. The above
example is pi to 15 digits past the decimal. I can not recall precisely
where, but in one of cases when porting software to Minix 3.1.6, I have
came across some manually encoded floating point values.

 From the standard, I can glean that hexadecimal representation is
completely legal, before and after the decimal point, but only base 10
numbers can be in the exponent (to avoid conflicting with the suffix 'F'
to denote single precision).

Beyond that, I am unable to find a specific statement saying hexadecimal
numbers must contain a decimal point, which means the above number is
ambiguous. However, I do not find the draft to be particularly succinct
in its discussions, and aspects appear to be contained in several
different sections.

Base 10 numbers are not required to have a decimal point when being

double ten = 10;

is perfectly legal, obviously. Therefore, I can infer neither is it a
requirement that base 16 representation requires a decimal point.

An argument could be made to follow what clang and/or pcc do, but at
least with clang, the method seems to be do what gcc does. I would
rather be compliant with the standard, and if there is ambiguity about
implementation, do the opposite of what gcc does (it's not really a
choice if all the options do the same thing).

I can say that clang and pcc appear to require a 'P' if the base 16
number contains a decimal point and interprets the number as an integer
if it is not present, but I am unable to find the specific clause
requiring this. If the decimal place and P are not required and P is
omitted, I would make the assumption that the exponent is 0, i.e.

double dec = 0x3FF3C28.F5C28F5C3;

would be interpreted as

double dec = 0x3FF3C28.F5C28F5C3P0;

As a result, if the decimal place is not required, the original example
of pi in IEEE-754 is ambiguous.

Does anyone know what the standard is requiring regarding the specific
format above?

Thank you,

"We believe that when you create a machine to do the work of a man,
you take something away from the man."
					-- Sojef
To unsubscribe send a mail to scc-dev+unsubscribe_at_simple-cc.org
Received on Wed 08 Mar 2023 - 14:16:53 CET

This archive was generated by hypermail 2.3.0 : Fri 21 Apr 2023 - 16:20:39 CEST