(original) (raw)

On Fri, Feb 6, 2009 at 5:01 PM, R�mi Forax <forax@univ-mlv.fr> wrote:

Kevin Bourrillion a �crit :

Hello,

A few thoughts.

First, this functionality is badly needed. �Absolutely everyone
rewrites this, in hundreds of different ways. �At Google we're no
exception, we have our own hand-rolled Join.java class full of static
methods, and it has thousands of callers in our private codebase.

Your selection of overloads (Iterable, Object\[\], and
(Object,Object...)) is exactly correct. �Kudos. �We'd added Iterator
versions as well, but almost no one uses them, so forget that.

My main concerns with your current API are:

\* I don't think that ",".join(collection) is a good idea. �It clashes
too strongly with the StringBuilder etc. versions. �It also plain
"looks weird". �A static String.join(",", collection) is preferable.
I don't think anyone would be surprised by this.
�

Use the receiver or not:
Pro:
\- join works with split and split use the receiver
\- Python use the receiver too

Cons:
\- static seems least surprising.
\- Perl's join and PHP's join are functions

I'm convinced, I will modify the patch.

\* It should be considered separately whether to support other
Appendables via static methods (somewhere) like:

�static A join(A appendable, String delimiter,
Iterable tokens) throws IOException

This provides a superset of what your StringBuilder etc. methods can
do, but your methods are still good because of their convenience
especially in chains of method calls, and because they don't need to
throw any exception.
�

Or perhaps, in JDK8 (when closure will be integrated) the compiler will
be able to infer that because join is used with String, it can't throw an IOE
That's not what BGGA propose but I will love to see that :)
But that's another story.

\* The biggest problem we've had with our Join class is that it turns
out that everyone has a different idea of how they'd like nulls
handled.

� - output as "null" (your, and our, current behavior)
� - skip?
� - skip only if trailing?
� - throw NPE? (often the best thing)
�

In my opinion (i am perhaps wong) join will be mostly used
just before print the String. That's why I have chosen the the print(ln) behavior.

We now support everyone by having the concept of an instantiable
Joiner (I've seen this in other languages). �Example:

� �private static final Joiner JOINER = Joiner.with(", ").skipNulls();
� � �. . .

� � � �return JOINER.join("Harry", "Ron", "Hermione");

Just to give everyone food for thought, our Joiner's API currently
looks like this:

�static Joiner with(String separator)

�Joiner skipNulls()
�Joiner useForNull(String nullText)

� A appendTo(A appendable, Iterable parts)
throws IOException
� A appendTo(A appendable, Object\[\] parts)
throws IOException
� A appendTo(A appendable, Object first,
Object... rest) throws IOException

�StringBuilder appendTo(StringBuilder builder, Iterable parts)
�StringBuilder appendTo(StringBuilder builder, Object\[\] parts)
�StringBuilder appendTo(StringBuilder builder, Object first, Object... rest)

�String join(Iterable parts)
�String join(Object\[\] parts)
�String join(Object first, Object... rest)

�MapJoiner withKeyValueSeparator(String keyValueSeparator)

�class MapJoiner {
� � A appendTo(A appendable, Map map)
throws IOException
� �StringBuilder appendTo(StringBuilder builder, Map map)
� �String join(Map map)

� �MapJoiner useForNull(String nullText)
�}

End of random thoughts.
�

Cool, i don't think this should be integrated in the JDK,
After all, If i want a full featured joiner, I can download �the google collection jar :)

\--
Kevin Bourrillion @ Google
internal: �http://go/javalibraries
google-collections.googlecode.com
google-guice.googlecode.com
�

R�mi