Null pointer exceptions, also known as NPEs, are pretty common errors.
Heck, two days ago I couldn’t buy a bus ticket because I got a nice “Object reference not set to an instance of an object” in the payment page.
The good news? Crystal doesn’t allow you to have null pointer exceptions.
Let’s start with the simplest example:
Compiling the above program gives this error:
Error in foo.cr:1: undefined method 'foo' for Nil nil.foo ^~~
nil
, the only instance of the Nil class, behaves just like any other class in Crystal.
And since it doesn’t have a method named “foo”, an error is issued at compile time.
Let’s try with a slightly more complex, but made up, example:
Can you spot the bug?
Compiling the above program, Crystal says:
Error in foo.cr:20: undefined method 'value' for Nil puts box.value ^~~~~ ================================================================================ Nil trace: foo.cr:19 box = make_box n ^ foo.cr:19 box = make_box n ^~~~~~~~ foo.cr:9 def make_box(n) ^~~~~~~~ foo.cr:10 case n ^
Not only it tells you that you might have a null pointer exception (in this case, when n is not one of 1, 2, 3, 4, 5, 6),
but it also shows you where the nil
originated. It’s in the case
expression, which has a default empty else
clause, which has a nil
value.
One last example, which might well be real code:
Can you spot the bug now? It turns out that TCPSocket#gets
(IO#gets, actually),
returns nil
at the end of the file or, in this case, when the connection is closed.
So capitalize
might be called on nil
.
And Crystal prevents you from writing such a program:
Error in foo.cr:10: undefined method 'capitalize' for Nil puts socket.gets.capitalize ^~~~~~~~~~ ================================================================================ Nil trace: std/file.cr:35 def gets ^~~~ std/file.cr:40 size > 0 ? String.from_cstr(buffer) : nil ^ std/file.cr:40 size > 0 ? String.from_cstr(buffer) : nil ^
To prevent this error, you can do the following:
This last program compiles fine. When you use a variable in an if
’s condition, and because the only
falsy values are nil
and false
, Crystal knows that line
can’t be nil inside the “then” part of the if
.
This is both expressive and executes faster, because it’s not needed to check for nil
values at runtime at every method call.
To conclude this post, one last thing left to say is that while porting the Crystal parser from Ruby to Crystal, Crystal refused to compile because of a possible null pointer exception. And it was correct. So in a way, Crystal found a bug in itself :-)
comments powered by Disqus