"com.lihaoyi" %% "sourcecode" % "0.1.3" // Scala-JVM
"com.lihaoyi" %%% "sourcecode" % "0.1.3" // Scala.js
sourcecode
is a small Scala library for that provides common "source code"
context to your program at runtime, similar to Python's __name__
, C++'s
__LINE__
or Ruby's __FILE__
. For example, you can ask for the file-name
and line number of the current file, either through the ()
syntax or via an
implicit:
val file = sourcecode.File()
assert(file.endsWith("/sourcecode/shared/src/test/scala/sourcecode/Tests.scala"))
val line = implicitly[sourcecode.Line]
assert(line == 16)
This might not be something you want to use for "business logic", but is very
helpful for things like debugging, logging or
providing automatic diagnostics for DSLs.
This information is also available via an implicit
, letting you write functions
that automatically pull it in.
Using SourceCode on code dealing with lots of anonymous functions or anonymous classes can easily turn what you see in your debug printouts from this:
To this:
By capturing source information you can use to give your objects and function meaningful names that tell you where they were defined, automatically without needing you to manually assign a string-ID to every anonymous function or anonymous class you define all over your code bas.
The kinds of compilation-time data that sourcecode
provides are:
sourcecode.File
: full path of the current file where the call occurssourcecode.Line
: current line numbersourcecode.Name
: the name of the nearest enclosing definition:val
,class
, whatever.sourcecode.FullName
: the name of the nearest enclosing definition:val
,class
, whatever, prefixed by the names of all enclosingclass
s,trait
s,object
s orpackage
s. Note that this does not include other enclosingdef
s,val
s,var
s orlazy val
s`sourcecode.Enclosing
: the name of the nearest enclosing definition:val
,class
, whatever, prefixed by the names of all enclosingclass
s,trait
s,object
s orpackage
s,def
s,val
s,var
s orlazy val
s`sourcecode.Text[T]
: when you want to take a value of typeT
, but also want to get the "source text" of that particular value. Note that if you have multiple statements in a{}
block,sourcecode.Text
will only capture the source code for the last expression that gets returned. This implicit is slightly experimental; be sure to report any bugs you find!sourcecode.Args
: the arguments that where provided to the nearest enclosing methodsourcecode.Name.Machine
,sourcecode.FullName.Machine
andsourcecode.Enclosing.Machine
which are similar tosourcecode.Name
,sourcecode.FullName
andsourcecode.Enclosing
except they do not filter out synthetic method names; e.g. if you want to see the<init>
names or<local foo>
names as part of the path, use these instead.
All these are available both via ()
and as implicits, e.g. sourcecode.File
can be summoned via sourcecode.File()
or implicitly[sourcecode.File].value
.
This also means you can define functions that pull in this information
automatically:
def foo(arg: String)(implicit file: sourcecode.File) = {
... do something with arg ...
... do something with file.value ...
}
foo("hello") // the implicit sourcecode.File is filled in automatically
sourcecode
does not rely on runtime reflection or stack inspection, and
is done at compile-time using macros. This means that it is both orders of
magnitude faster than e.g. getting file-name and line-numbers using stack
inspection, and also works on Scala.js where reflection and stack inspection
can't be used.
Here are a few examples of sourcecode
's core functions being used in a
variety of contexts. Hopefully they will give you an idea of how the various
implicits behave:
package sourcecode
object Implicits {
def implicitRun() = {
val name = implicitly[sourcecode.Name]
assert(name.value == "name")
val fullName = implicitly[sourcecode.FullName]
assert(fullName.value == "sourcecode.Implicits.fullName")
val enclosing = implicitly[sourcecode.Enclosing]
assert(enclosing.value == "sourcecode.Implicits.implicitRun enclosing")
val pkg = implicitly[sourcecode.Pkg]
assert(pkg.value == "sourcecode")
val file = implicitly[sourcecode.File]
assert(file.value.endsWith("/sourcecode/shared/src/test/scala/sourcecode/Implicits.scala"))
val line = implicitly[sourcecode.Line]
assert(line.value == 20)
lazy val myLazy = {
trait Bar{
val name = implicitly[sourcecode.Name]
assert(name.value == "name")
val fullName = implicitly[sourcecode.FullName]
assert(fullName.value == "sourcecode.Implicits.Bar.fullName")
val file = implicitly[sourcecode.File]
assert(file.value.endsWith("/sourcecode/shared/src/test/scala/sourcecode/Implicits.scala"))
val line = implicitly[sourcecode.Line]
assert(line.value == 34)
val enclosing = implicitly[sourcecode.Enclosing]
assert(enclosing.value == "sourcecode.Implicits.implicitRun myLazy$lzy Bar#enclosing")
}
val b = new Bar{}
}
myLazy
}
}
Note that in "normal" usage you would not directly call implicitly
to summon
up sourcecode
values; rather, you would add implicit parameters of these
types to your functions. That would make these values automatically available
to your functions without needing to manually keep passing them in. Apart from
summoning them via implicits, you can also use the apply
method on each type
to pull them in using the ()
syntax:
package sourcecode
object Apply {
def applyRun() = {
val name = sourcecode.Name()
assert(name == "name")
val fullName = sourcecode.FullName()
assert(fullName == "sourcecode.Apply.fullName")
val enclosing = sourcecode.Enclosing()
assert(enclosing == "sourcecode.Apply.applyRun enclosing")
val pkg = sourcecode.Pkg()
assert(pkg == "sourcecode")
val file = sourcecode.File()
assert(file.endsWith("/sourcecode/shared/src/test/scala/sourcecode/Apply.scala"))
val line = sourcecode.Line()
assert(line == 20)
lazy val myLazy = {
trait Bar{
val name = sourcecode.Name()
assert(name == "name")
val fullName = sourcecode.FullName()
assert(fullName == "sourcecode.Apply.Bar.fullName")
val file = sourcecode.File()
assert(file.endsWith("/sourcecode/shared/src/test/scala/sourcecode/Apply.scala"))
val line = sourcecode.Line()
assert(line == 34)
val enclosing = sourcecode.Enclosing()
assert(enclosing == "sourcecode.Apply.applyRun myLazy$lzy Bar#enclosing")
}
val b = new Bar{}
}
myLazy
}
}
By default, the various implicits all ignore any synthetic <init>
or
<local Foo>
methods that might be present:
package sourcecode
object NoSynthetic {
def run() = {
class EnumValue(implicit name: sourcecode.Name){
override def toString = name.value
}
object Foo extends EnumValue
assert(Foo.toString == "Foo")
object Bar{
assert(sourcecode.Name() == "Bar")
assert(sourcecode.FullName() == "sourcecode.NoSynthetic.Bar")
assert(sourcecode.Enclosing() == "sourcecode.NoSynthetic.run Bar")
}
Bar
}
}
If you want these synthetic methods to be shown, use the .Machine
versions
of each of these instead:
package sourcecode
object Synthetic {
def run() = {
class EnumValue(implicit name: sourcecode.Name.Machine){
override def toString = name.value
}
object Foo extends EnumValue
assert(Foo.toString == "<init>")
object Bar{
assert(sourcecode.Name.Machine() == "<local Bar>", sourcecode.Name())
assert(sourcecode.FullName.Machine() == "sourcecode.Synthetic.Bar.<local Bar>")
assert(sourcecode.Enclosing.Machine() == "sourcecode.Synthetic.run Bar.<local Bar>")
}
Bar
}
}
Hopefully this has given you a reasonable feel for what* sourcecode does. You may still be wondering why we would want any of this: what could we possibly use these things for? Why would we want to write code that depends on our package paths or variable names? The section below will provide use cases that you will hopefully be able to relate to.
At first it might seem strange to make use of these source-level details in your program: shouldn't a program's meaning not change under re-formatting and re-factoring?
It turns out that there are a number of entirely valid use cases for this sort of information that is both extremely handy, and also would not be surprising at all to a developer using your API. Here are a few example use cases:
You can use sourcecode.File
and sourcecode.Line
to define log
functions
that automatically capture their line number and file-name
def log(foo: String)(implicit line: sourcecode.Line, file: sourcecode.File) = {
println(s"${file.value}:${line.value} $foo")
}
log("Foooooo") // sourcecode/shared/src/test/scala/sourcecode/Tests.scala:86 Fooooo
This can be handy for letting you see where the log lines are coming from,
without tediously tagging every log statement with a unique prefix.
Furthermore, this happens at compile time, and is thus orders of magnitude
faster than getting this information by generating stack traces, and works
on Scala.js where stack-inspection does not. Lastly, if you want additional
information such as method names, class names, or packages to be provided to
your logging function, you can easily do so by asking for the sourcecode.Name
or sourcecode.FullName
or sourcecode.Pkg
implicits.
You can use sourcecode.Name
to define an enumeration-value factory function
that automatically assigns names to the enum values based on the name of the
val
that it is assigned to
package sourcecode
object EnumExample {
def run() = {
case class EnumValue(name: String){
override def toString = name
}
class Enum{
def value(implicit name: sourcecode.Name) = EnumValue(name.value)
}
object MyEnum extends Enum{
val firstItem = value
val secondItem = value
}
assert(MyEnum.firstItem.toString == "firstItem")
assert(MyEnum.secondItem.toString == "secondItem")
}
}
This is very handy, and this functionality is used in a number of libraries such as FastParse and Scalatags to provide a boilerplate-free experience while still providing good debuggability and convenience.
Sometimes you want to make sure that different enum values in differently
named enums (or even an enum of the same name in a different package!) are
given unique names. In that case, you can use sourcecode.FullName
or
sourcecode.Enclosing
to capture the full path e.g.
"com.mypkg.MyEnum.firstItem"
and "com.mypkg.MyEnum.secondItem"
:
package sourcecode
object EnumFull {
def run() = {
case class EnumValue(name: String){
override def toString = name
}
class Enum{
def value(implicit name: sourcecode.FullName) = EnumValue(name.value)
}
object MyEnum extends Enum{
val firstItem = value
val secondItem = value
}
assert(MyEnum.firstItem.toString == "sourcecode.EnumFull.MyEnum.firstItem")
assert(MyEnum.secondItem.toString == "sourcecode.EnumFull.MyEnum.secondItem")
}
}
You can also use sourcecode.Name
in an constructor, in which case it'll be
picked up during inheritance:
class EnumValue(implicit name: sourcecode.Name){
override def toString = name.value
}
object Foo extends EnumValue
println(Foo.toString)
assert(Foo.toString == "Foo")
How many times have you written tedious code like
object Bar{
def foo(arg: String) = {
println("Bar.foo: " + arg)
}
}
Where you have to prefix every print statement with the name of the enclosing
classes, objects or functions to make sure you can find your print output
2-3 minutes later? With source.Enclosing
, you can get this for free:
def debug[V](value: sourcecode.Text[V])(implicit enclosing: sourcecode.Enclosing) = {
println(enclosing.value + " [" + value.source + "]: " + value.value)
}
class Foo(arg: Int){
debug(arg) // sourcecode.DebugRun.main Foo [arg]: 123
def bar(param: String) = {
debug(arg -> param)
}
}
new Foo(123).bar("lol") // sourcecode.DebugRun.main Foo#bar [arg -> param]: (123,lol)
You can easily vary the amount of verbosity, e.g. by swapping the
sourcecode.Enclosing
for a `sourcecode.Name if you think it's too verbose:
def debug[V](value: sourcecode.Text[V])(implicit name: sourcecode.Name) = {
println(name.value + " [" + value.source + "]: " + value.value)
}
class Foo(arg: Int){
debug(arg) // Foo [arg]: 123
def bar(param: String) = {
debug(param -> arg)
}
}
new Foo(123).bar("lol") // bar [param]: lol
Or leaving it out entirely:
def debug[V](value: sourcecode.Text[V]) = {
println("[" + value.source + "]: " + value.value)
}
class Foo(arg: Int){
debug(arg) // [arg]: 123
def bar(param: String) = {
debug(param -> arg)
}
}
new Foo(123).bar("lol") // [param]: lol
Thus you can easily configure how much information your debug
helper method
needs, at its definition, without having to hunt all over your codebase for the
various debug
call-sites you left lying around and manually tweaking the
verbosity of each one. Furthermore, if you want additional information like
sourcecode.Line
or sourcecode.File
, that's all just one implicit away.
The PPrint
library provides a pprint.log
method that does exactly this: prints out the
value provided (in this case pretty-printing it with colors and nice formatting
& indentation) together with the enclosing context and line number, so you
can easily distinguish your individual prints later:
scala> class Foo{
| def bar(grid: Seq[Seq[Int]]) = {
| // automatically capture and print out source context
| pprint.log(grid, tag="grid")
| }
| }
defined class Foo
scala> new Foo().bar(Seq(0 until 10, 10 until 20, 20 until 30))
pkg.Foo#bar "grid":12
List(
Range(0, 1, 2, 3, 4, 5, 6, 7, 8, 9),
Range(10, 11, 12, 13, 14, 15, 16, 17, 18, 19),
Range(20, 21, 22, 23, 24, 25, 26, 27, 28, 29)
)
pprint.log
is itself defined as
def log[T: PPrint](value: T, tag: String = "")
(implicit cfg: Config = Config.Colors.PPrintConfig,
path: sourcecode.Enclosing,
line: sourcecode.Line) = ...
Using sourcecode.Enclosing
and sourcecode.Line
to provide the context to
be printed. You can, or course, define your own log
method in the same way,
customizing it to print or not-print exactly what you want to see via the
implicits that sourcecode
provides!
sourcecode.Args
can be used to access all parameters that where provided
to a method:
def debug(implicit name: sourcecode.Name, args: sourcecode.Args): Unit = {
println(name.value + args.value.map(_.map(a => a.source + "=" + a.value).mkString("(", ", ", ")")).mkString(""))
}
def foo(bar: String, baz: Int)(p: Boolean): Unit = {
debug
}
foo("baz", 42)(true) // foo(bar=baz, baz=42)(p=true)
The Scala programming is a popular choice to embed domain-specific languages: that means that you start with some external language, e.g. this MathProg example
param m;
param n;
param l;
set I := 1 .. m;
set J := 1 .. n;
set K := 1 .. l;
param c{J};
param d{K};
param a{I, J};
var x{J} integer, >= 0;
var y{K} >= 0;
The linked slides has more detail about what exactly this language does (it describes mathematical optimization problems). For a variety of reasons, you may prefer to write this as part of a Scala program instead: for example you may want Scala's IDE support, or its ability to define functions that help reduce boilerplate, or maybe you like the way the compiler provides type errors when you do the wrong thing.
A first attempt at converting this to Scala may look like this:
val m = param("m")
val n = param("n")
val l = param("l")
val I = set("I") := 1 to m
val J = set("J") := 1 to m
val K = set("K") := 1 to m
val c = param("c", J)
val d = param("d", K)
val a = param("a", I, J)
val x = xvar("x", J).integer >= 0
val y = xvar("y", K) >= 0
There's a bunch of duplication around the names of the val
s: each val
has its name repeated in a string that gets passed to the expression on the
right. This is for the program to use the name of the val
later: for example
when printing error messages, or the results of the computation, you want to
see which val
s are involved! Thus you end up duplicating the names over and
over and over.
With sourcecode, you can easily define param
set
and xvar
as taking
implicit sourcecode.Name
s, thus eliminating all the boilerplate involved in
duplicating names:
val m = param
val n = param
val l = param
val I = set := 1 to m
val J = set := 1 to m
val K = set := 1 to m
val c = param(J)
val d = param(K)
val a = param(I, J)
val x = xvar(J).integer >= 0
val y = xvar(K) >= 0
The popular FastParse parser-combinator library uses sourcecode for exactly this use case
import fastparse.all._
val A = P( "aa" )
val B = P( "bb" )
val C = P( (A | B).rep(1) )
C.parse("aabb") // Success((), 4)
C.parse("X") // Failure((A | B):1:1 ..."X")
As you can see, the names of the rules A
and B
are embedded in the error
messages for parse failures. This makes debugging parsers far easier, while
saving you the effort of duplicating the name of the parser in possibly
hundreds of rules in a large parser. In this case, it is the P(...)
function
which takes an implicit sourcecode.Name
that does this work:
def P[T](p: => Parser[T])(implicit name: sourcecode.Name): Parser[T] =
parsers.Combinators.Rule(name.value, () => p)
And forwards the name on to the actual Rule
object, which can make use of it
in its .toString
method.
- Add scala 2.12.x support, thanks to Lars Hupel
-
Add
sourcecode.Args
implicit, which can be used to capture debugging information about the nearest enclosing function call for logging/debugging, thanks to Benjamin Hagemeister -
Attempted fix for #17 and #13, thanks to Simeon H.K. Fitch
-
Ignore
<local foo>
and<init>
symbols when determiningsourcecode.Name
,sourcecode.FullName
orsourcecode.Enclosing
. If you want these, use thesourcecode.Name.Machine
/sourcecode.FullName.Machine
/sourcecode.Enclosing.Machine
implicits instead. -
Add
sourcecode.Text
implicit to capture source code of an expression -
Add implicit conversions to
sourcecode.*
, so you can pass in aString
to manually satisfy and implicit wanting asourcecode.Name
orsourcecode.FullName
orsourcecode.File
, anInt
to satisfy an implicit asking forsourcecode.Line
-
sourcecode.Enclosing
has been simplified to take a singleString
rather than the previousVector[Chunk]
. -
Added the
sourcecode.Pkg
implicit, which provides the current enclosing package without any of theclass
s/object
s/def
s/etc.. Can be subtracted fromsourcecode.Enclosing
if you only want theclass
s/object
s/def
s/etc.
- First release