commit_conv_vec() writes the elements one by one, so a reader can
still see a mix of old and new values while the loop runs. Same as
before this series, so no regression, but the changelog could say
this doesn't make the write atomic either. Otherwise someone later
will expect the vector to change all at once.

The copy is limited by table->maxlen and only vectors pay for it.

Fine by me.

Reviewed-by: Bradley Morgan <[email protected]>
Thanks!

Reply via email to