In the first betas of Codename One we had a lot of bugs related to 2D arrays due to XMLVM. We no longer use XMLVM but the recommendation to avoid 2D arrays remains. We still use them in some occasions e.g. in the creation of a DefaultTableModel but the implementation discards them in favor of an ArrayList internally.
To be fair, a 2D array will be faster than an ArrayList so in this case we discarded them due to their lack of flexibility
What’s the Problem?
To understand the problem of 2D arrays we need to understand how they work. When we write something like this:
That sounds like a "small" semantic difference but it’s a huge one!
That means we have 4 arrays where we should have had just 1. That’s far more objects to GC & allocate. That also means that every array lookup has to happen twice. The problem here is that these semantics are often used when processing large data such as screen images that might be pretty big at which case you need to multiply that overhead by 1024.
There is one exception where the length of the individual entries isn’t uniform, however that’s a pretty rare use case. Most such cases use collection classes and not multi-dimensional arrays.
We do a lot of image RGB manipulations and we always use a single dimension array. This is far more performant and pretty easy to use. When we want to access an individual pixel we can do:
pixels[x + y * width] = newValue;
It might seem that the overhead of multiplication and addition would cancel out the benefit of the array lookup. That’s not true as those operations are remarkably fast by comparison especially with AoT compilation. JIT’s might be able to do some magic with 2D arrays but even if they can do that they would still need more GC and allocation work as both of those are unavoidable. So a 1D array would always be superior in Java.
Shai is the co-founder of Codename One. He's been a professional programmer for over 25 years. During that time he has worked with dozens of companies including Sun Microsystems. For more follow Shai on Twitter & github.