leftjoin! is actually copying reference instead of value?!

[mzy2240]

MWE:

a = DataFrame(x=[1,1])
b = DataFrame(x=[1], y=[[1,2,3]])
leftjoin!(a, b, on=:x)
deleteat!(a[1,:y], 1)
println(a)

And out of surprise, you would see that both vectors in the y column got updated ([1,2,3] -> [2,3])! If I iterate over each row using eachrow, then changing the first row would also affect the second row. I don't know whether it is a desired behavior (I am sure in Pandas it does not behave like this), and I am wondering what to do if I only want to change the first vector in the y column.

Thanks!

[bkamins]

I am sure in Pandas it does not behave like this

Pandas behaves the same way:

>>> df1 = pd.DataFrame({'key':[1,1]})
>>> df2 = pd.DataFrame({'key':[1], 'value':[[1,2,3]]})
>>> res = pd.merge(df1, df2, how='left', left_on='key',right_on='key')
>>> res['value']
0    [1, 2, 3]
1    [1, 2, 3]
Name: value, dtype: object
>>> res['value'][0].append(4)
>>> res['value']
0    [1, 2, 3, 4]
1    [1, 2, 3, 4]
Name: value, dtype: object
>>> df2['value']
0    [1, 2, 3, 4]
Name: value, dtype: object

if you want a copy you have to do it manually after a join by doing a.y = copy.(a.y) or a.y = deepcopy(a.y).