[Java] Users are required to initialize VariableLengthVectors.offsetVector before calling VariableLengthVectors.mutator.getSafe

[asfimport]

https://github.com/apache/arrow/blob/master/java/vector/src/main/codegen/templates/VariableLengthVectors.java#L492

Here VariableLengthVectors.getMutator().setSafe() calls:

offsetVector.getAccessor().get(index)

however, index 0 of offsetVector (which is always 0) is not initialized by VariableLengthVectors.

As a result, user of the VariableLengthVectors needs to manually initialize
the class by calling:

VariableLengthVectors.getOffsetVector().getMutator().setSafe(0, 0)

I wonder if this is necessary or should VariableLengthVectors initialize this for the user

Reporter: Li Jin / @icexelloss

_{Note: This issue was originally created as ARROW-485. Please see the migration documentation for further details.}

[asfimport]

Jacques Nadeau / @jacques-n:
There should be better documentation on this. In order to use vectors, the correct order of operations are:

1. allocateNew() (allocate memory for the vector)
2. Set one or more values using getMutator().setSafe(i, val). Note, this has to be monotonically increasing position but allows index skips.
3. call set valueCount(n) where n is the number of valid indices in the vector
4. read or serialize data

I believe that if you follow these operations, you will not have a problem here. I'm guessing you're trying to use a vector before allocating (1).

[asfimport]

Li Jin / @icexelloss:
Thanks, this is super helpful. I am going to follow the steps.