Is std::mem::size_for_val the recommended way to calculate the total allocation for SerializedValues?

[Jasperav]

I upgraded Catalytic to the latest version of this driver (0.10) because I saw there where performance improvements by using size_for_val here: #809. Catalytic now also uses them, e.g.: https://github.com/Jasperav/Catalytic/blob/4f9babb197c865819ef5f7537685873f97ac8037/catalytic_table_to_struct/example/src/generated/child.rs#L174

Is this the best way now or am I missing something? I see that there are now new traits coming but that can take awhile: #463

[piodul]

The PR you linked introduces a heuristic, i.e. it sums the sizes of Rust representations of all the columns in tuples/Vec/slice using std::mem::size_of_val. The std::mem::size_of_val function will sometimes correspond to the serialized CQL size of a value, but not always - for example, if you have a Vec column type it will always be counted as 3 * size_of_val::<usize>(), regardless from the number of its elements.

Calculating the exact serialized size up front sometimes requires more complicated calculations, so it would require extending the ValueList/Value trait. We'd rather add this functionality to the new traits instead of investing more time in the old ones.

To sum up - this is a heuristic which should reduce the current number of reallocations; it's possible to do better, but it requires more changes.

[Lorak-mmk]

closing in favor of #385