H. S. Teoh 
| It's been a while since the last "D rocks!" post. So here's one.
I'm guessing that most D users don't realize extent of the flexibility of std.format -- I know I didn't until I discovered this little gem hidden in the docs (and then only implicitly!).
But first, a motivating example. Suppose you have some kind of data structure, let's call it S, and at some point in your program, you want to output it. The most obvious way, of course, is to implement a toString() method:
struct S {
... // my sooper sekret data here!
string toString() const pure @safe {
// Typical implementation to minimize overhead
// of constructing string
auto app = appender!string();
... // transform data into string
return app.data;
}
}
void main() {
auto s = S();
... // do wonderful stuff with s
writeln(s);
}
This is the "traditional" implementation, of course. A slight optimization that's possible is to realize that there's an alternative signature of toString() that alleviates the overhead of doing any string allocations at all:
struct S {
// This method now takes a delegate to send data to.
void toString(scope void delegate(const(char)[]) sink) const
{
// So you can write your data piecemeal to its
// destination, without having to construct a
// string and then return it.
sink("prelude");
sink(... /* beautiful prologue */);
sink("concerto");
sink(... /* beautiful body */);
sink("finale");
sink(... /* beautiful trailer */);
// Look, ma! No string allocations needed!
}
}
So far so good. This is (or should be) all familiar ground.
But suppose now you want to write your data to, say, a backup file in one format, but output your data to the user in another format. How would you do this?
You could make toString() output one format, say the on-disk format,
then add another method, say toUserReadableString() for outputting the
other format. But this is ugly and non-extensible. What if you have a
whole bunch of other formats that need to be output? You'd be drowning
in toNetworkString(), toDatabaseString(), toHtmlEscapedString(), etc.,
etc., which bloats your data's API and isn't very maintainable to boot.
Here's where a little known feature of std.format comes in. Note that when you write:
S s;
writeln(s);
This actually ultimately gets translated to the equivalent of:
S s;
writefln("%s", s);
Where the %s specifier, of course, means "convert to the standard string representation". What is less known, though, is that this actually translates to something like this:
Writer w = ... /* writer object that outputs to stdout */
FormatSpec!Char fmt = ... /* object representing the meaning of "%s" */
s.toString((const(char)[] s) { w.put(s); }, fmt);
In human language, this means that "%s" gets translated into a FormatSpec object containing "s" in its .spec field (and if you write, say, "%10s", the 10 gets stored in the .width field, etc.), and then this FormatSpec object gets passed to the toString method of the object being formatted, if it is defined with the correct signature. To see this in action, let's do this:
struct S {
void toString(scope void delegate(const(char)[]) sink,
FormatSpec!char fmt) const
{
// This is for probing how std.format works
// under the hood.
writeln(fmt.spec);
}
}
void main() {
S s;
// Wait -- what? What on earth are %i, %j, %k, and %l?!
writeln("%i", s); // Hmm, prints "i"!
writeln("%j", s); // Hmm, prints "j"!
writeln("%k", s); // Hmm, prints "k"!
writeln("%l", s); // Hmm, prints "l"!
}
Do you see what's going on? The format specifiers are not hard-coded into the library! You can invent your own specifiers, and they get passed into the toString method. This allows us to do this:
struct S {
void toString(scope void delegate(const(char)[]) sink,
FormatSpec!char fmt) const
{
switch(fmt.spec) {
// Look, ma! I invented my own format specs!
case "i":
// output first format to sink
break;
case "j":
// output second format to sink
break;
case "k":
// output third format to sink
break;
case "l":
// output fourth format to sink
break;
case "s":
// output boring default string format
break;
default:
throw new Exception(
"Unknown format specifier: %" ~
fmt.spec);
}
}
}
Of course, FormatSpec contains much more than just the letter that defines the specifier. It also contains field width, precision, etc.. So you can implement your own handling for all of these parameters that are specifiable in a writefln format string.
Here's a somewhat silly example to show the flexibility conferred:
import std.format;
import std.stdio;
struct BoxPrinter {
void toString(scope void delegate(const(char)[]) sink,
FormatSpec!char fmt) const
{
if (fmt.spec == 'b') {
// Draws a starry rectangle
foreach (j; 0..fmt.precision) {
foreach (i; 0..fmt.width) {
sink("*");
}
sink("\n");
}
} else {
// Boring old traditional string representation
sink("BoxPrinter");
}
}
}
void main() {
BoxPrinter box;
writefln("%s", box);
writefln("%6.5b", box);
writefln("%3.2b", box);
writefln("%7.4b", box);
}
Here's the output:
BoxPrinter
******
******
******
******
******
***
***
*******
*******
*******
*******
As you can see, the width and precision parts of the custom %b specifier has been reinterpreted into the dimensions of the box that will be printed. And when you specify %s, a traditional innocent-looking string is printed instead. In effect, we have implemented our own custom format specifier.
D rocks!!
Oh, and did I mention that D rocks?
T
--
Do not reason with the unreasonable; you lose by definition.
|
Permalink
Reply
