NumFormat

IMPORTANT: all of this package’s functionalities are being merged into Formatting.jl, please use that instead

A way to get around the limitation that @sprintf has to take a literal string argument. The core part is basically a c-style print formatter using the standard @sprintf macro. It also adds functionalities such as commas separator (thousands), parenthesis for negatives, stripping trailing zeros, and mixed fractions.

Usage and Implementation

The idea here is that the package compiles a function only once for each unique format string within the NumFormat.* name space, so repeated use is faster. Unrelated parts of a session using the same format string would reuse the same function, avoiding redundant compilation. To avoid the proliferation of functions, we limit the usage to only 1 argument. Practical consideration would suggest that only dozens of functions would be created in a session, which seems manageable.

Usage

using NumFormat

fmt = "%10.3f"
s = sprintf1( fmt, 3.14159 ) # usage 1. Quite performant. Easiest to switch to.

fmtrfunc = generate_formatter( fmt ) # usage 2. This bypass repeated lookup of cached function. Most performant.
s = fmtrfunc( 3.14159 )

s = format( 3.14159, precision=3 ) # usage 3. Most flexible, with some non-printf options. Least performant.

Alternative approaches suggested by others:

• Formatting.jl.

• Put the macro in a quote block and eval it (very slow)

fmt = "%10d"
n = 1234
s = eval( Expr( :macrocall, symbol( "@sprintf" ), fmt, n ) ) # VERY slow, 1000x penalty

• ccall to libc sprintf. See this gist. The example shows 6-7x speed penalty.

• Set up a lambda with the macro inside. Ok for repeated use. But the lambda goes out of scope quickly so it cannot be reused. @eval would also repeat compilation, even for the same format.

fmt = "%10d"
n = 1234

#method A lambda
l = :( x -> x ) # placeholder lambda
l.args[2].args[2] = Expr( :macrocall, symbol( "@sprintf" ), fmt, :x )
mfmtr = eval(l)

#method B @eval
@eval mfmtr(x) = @sprintf($fmt,x)

s = mfmtr( n ) # quite fast, but the definition is clunky

Speed

sprintf1: Speed penalty is about 20% for floating point and 30% for integers.

If the formatter is stored and used instead (see the example using generate_formatter above), the speed penalty reduces to 10% for floating point and 15% for integers.

Commas

This package also supplements the lack of thousand separator e.g. "%'d", "%'f", "%'s".

Note: "%'s" behavior is that for small enough floating point (but not too small), thousand separator would be used. If the number needs to be represented by "%e", no separator is used.

Flexible format function

This package contains a run-time number formatter format function, which goes beyond the standard sprintf functionality.

An example:

s = format( 1234, commas=true ) # 1,234
s = format( -1234, commas=true, parens=true ) # (1,234)

The keyword arguments are (Bold keywards are not printf standard)

• width. Integer. Try to fit the output into this many characters. May not be successful. Sacrifice space first, then commas.
• precision. Integer. How many decimal places.
• leftjustified. Boolean
• zeropadding. Boolean
• commas. Boolean. Thousands-group separator.
• signed. Boolean. Always show +/- sign?
• positivespace. Boolean. Prepend an extra space for positive numbers? (so they align nicely with negative numbers)
• parens. Boolean. Use parenthesis instead of "-". e.g. (1.01) instead of -1.01. Useful in finance. Note that you cannot use signed and parens option at the same time.
• stripzeros. Boolean. Strip trailing '0' to the right of the decimal (and to the left of 'e', if any ).
  • It may strip the decimal point itself if all trailing places are zeros.
  • This is true by default if precision is not given, and vice versa.
• alternative. Boolean. See # alternative form explanation in standard printf documentation
• conversion. length=1 string. Default is type dependent. It can be one of aAeEfFoxX. See standard printf documentation.
• mixedfraction. Boolean. If the number is rational, format it in mixed fraction e.g. 1_1/2 instead of 3/2
• mixedfractionsep. Default _
• fractionsep. Default /
• fractionwidth. Integer. Try to pad zeros to the numerator until the fractional part has this width
• tryden. Integer. Try to use this denominator instead of a smaller one. No-op if it'd lose precision.

See the test script for more examples.