Let's do .each do |key, value| here and rename the existing key to section and the existing value to contents.

Last one, promise, Crystal actually interprets \n within single quotes since they're used to distinguish the type, not whether escape sequences are interpreted ;)

You should add an actual example in code which does not rely on the INI.parse and the contents of an additional file. The example in the docstring would do.

Is a configurable space really useful? And why is it not a single whitespace by default?

I do this because there are INI files with and without spaces around =. It's not clear what is the good default. For example, systemd services has without spaces and php.ini has one space.

space should be IMO " " (space!) from default.

This method (and the overload below) should also take NamedTuple(String, String) as an ini argument.

Hard to decide between no spaces and one space. By looking around, the specifications of Microsoft and Cloanto tend to use no spaces, but Apache and PHP use one around =.
Anyone has others official sources for INI specifications?

In fact, removing the type restriction from the method signature, the code should instantly work for named tuples and other payload types than String.

We can add Boolean, Number, String according to the different implementations listed in Wikipedia. Array isn't always supported.

No, just remove all type restrictions. This way, you can pass in any type and the builder just prints the result of to_s. This is the most developer-friendly solution.

@straight-shoota I know, but adding NamedTuple(String, String) raises can only instantiate NamedTuple with named arguments.

By the way regardless of the types, the mapping should follow the Key(Value, Key(Key, Value) structure.

That's OK, we will just need to add verifications and raise custom errors to avoid incorrect INI files if we accept everything.

Just do it like this:

module INI
  def self.build(io : IO, ini, space : String = "")
    ini.each do |section, contents|
      io << '[' << section << "]\n"
       contents.each do |key, value|
         io << key << space << '=' << space << value << '\n'
       end
       io.puts
     end
  end
end

INI.build(STDOUT, {"main": {foo: "bar", baz: 12, regex: /regex/}})

No need for type restrictions or custom errors. If a data structure that does not provide a matching interface, it will automatically result in a compile time error.

You should use a String::Builder instead of Memory::IO, because this is optimized for creating a string.
And please add a method overload INI.build(io : IO, hash) that can directly write to an IO instead of allocating a string.

I've first found IO::Memory.new if the Crystal performance guide. So we can further improve perfomance using String::Builder for strings; I will make a PR for this, thanks.

I've not found yet a case when IO::Memory.new is slower than the String.build approach, using Benchmark.ips. Anyone has a sample to demonstrate this? We will be able to add it to the Performance Guide.

require "benchmark"

Benchmark.ips do |x|
  x.report("String.build") do
    String.build do |io|
      100.times do
        io << "hello world"
      end
    end
  end
  x.report("IO::Memory") do
    io = IO::Memory.new
    100.times do
      io << "hello world"
    end
    io.to_s
  end
end

Output:

String.build 475.55k (   2.1µs) (± 1.25%)       fastest
  IO::Memory  252.7k (  3.96µs) (± 1.47%)  1.88× slower

whats your benchmark?

@larubujo Thanks! I haven't used 100.times before. I had just added a block similar to ones already present in the guide. So I've faced strange issues, memory related I think.

This line could be replaced by io.puts

Better use String.build instead of instantiating a builder directly.

The last argument should be space

good catch, miss this one

ditto

Why is this in the spec??

Sorry, I've pasted this by mistake. I revert this ASAP

It's probably better not to change existing test data. You don't really need to use this file.

That's for Char type testing, that should be converted to a String.

Personally I think that having the space be configurable as any string is a bit too much and makes it possible to create invalid INI files. Why not make it a boolean which switches between "" and " ". Regardless, I think 1 space should be the default, not zero. We can always change this default later. Actually nvm, let's maap whitespace false by default.

I'd rather

io << key
io << ' ' if space
io << '='
io << ' ' if space
io << value << '\n'

I've done it with io << key << (space ? " = " : '=') << value << '\n', less verbose. Is it fine?

redundant blank line.