Saxon

I don't think there's a realistic prospect of handling in-memory documents where the string value of any node is greater than Java's limits on the capacity of a String. The 2Gb limit is too firmly entrenched in the Java infrastructure and in the APIs we use for accessing the tree.

The best we could do would be to make the failure softer.

For example we could try to ensure that LargeStringBuffer.append() allows the size to go beyond 2Gb, but that wouldn't be useful on its own, because the class implements Java's CharSequence which has methods such as charAt() and subSequence() that use int lengths and offsets. In addition there is simply the problem of testing it.

Moving from 32bit offsets in the TinyTree arrays to 64bit offsets would affect memory usage for everyone; we would need to introduce a new tree model for this.

In addition, I'm not convinced that adding extra checks to get a cleaner error message is worth the trouble. The small number of people who hit this problem will know that they are stretching the boundaries, so the mode of failure doesn't really matter too much.