stringtofloat

Minimal dependency, easy to understand string to floating point number conversion.

Requirements

• C compiler with unsigned 64-bit integer type and basic arithmetic on it (comparision, addition, subtraction, bit shifts).
• <stdint.h> header file or its equivalent defining int32_t, uint32_t, uint64_t types.
• It is assumed that int type is at least 16-bit wide.
• Size of other types and system endianess are not important.
• The C standard library is not required, unless a compiler needs it to provide the above features.
• No other libraries are required.

Features

• Converts an ASCII string to IEEE 754 double precision number.
• Supports scientific notation, all common abbreviations (no zero before dot, ommitted plus signs).
• The code is easy to understand, algorithm is explained in details.
• The code is splitted into parser and converter parts, which are independent. Parser is easy to modify to support non-ASCII or wide chars.
• Parser reads character by character, so can work with pipe style inputs. There is no limit for length of string, string is not buffered.
• Small in size. On most platforms the compiled code is less than 2 kB.
• Uses no floating point arithmetic by itself.
• Covers full range of double precision numbers with precise limit checking.
• Numbers out of range are converted to infinity with proper sign.
• Very small numbers are converted to zero with proper sign.

Limitations

• No support for explicit "+Inf", "-Inf", "NaN" strings.
• Denormalized floats are never generated, zero of proper sign is returned instead.
• The rounding mode is always round towards zero.

Tested platforms

• Linux on x86_64
• Linux on x86
• Linux on ARM
• Windows on x86_64
• MorphOS on PowerPC (32-bit)
• AmigaOS on M68k