Add insert_all

[ruslandoga]

This PR adds a helper NIF that can be used to bulk-insert "unlimited" number of rows.

Should be enough to clоse #129 once it gets into ecto_sqlite3

Regarding performance: XQLite uses this approach in insert_all benchmark in https://github.com/ruslandoga/elixir-sqlite-bench

[warmwaffles]

This looks good.

[warmwaffles]

I do not like how often the windows test flaps.

[warmwaffles]

Ugh

c_src\sqlite3_nif.c(703): error C2057: expected constant expression c_src\sqlite3_nif.c(703): error C2466: cannot allocate an array of constant size 0 c_src\sqlite3_nif.c(703): error C2133: 'types_array': unknown size 

[warmwaffles]

Use enif_alloc for this. This way the VM will know about the memory allocated and will count it correctly. Also enif_free to free the memory.

[ruslandoga]

Right, I did it in the very next commit: cf69cd5

[warmwaffles]

ah I was walking the changes.

[warmwaffles]

We are missing cleanup code here. We need to enif_free(types_array) before all of these error returns.

[ruslandoga]

Ah, right. And not just here, but for every return.

[ruslandoga]

I think I'll need to find another way to pass the types information. The current approach is getting complicated without VLA.

[ruslandoga]

Sorry!

[warmwaffles]

This is just the nature of C.

One way would be to use a common cleanup statement with goto like this.

{ // ... if (!enif_get_list_cell(env, head, &param, &head)) { goto bad_arg_exception; } // .. enif_free(types_array); return am_done; bad_arg_exception: enif_free(types_array); return raise_badarg(env, param); }

[ruslandoga]

This is just the nature of C.

I think it should be possible to implement insert_all or a similar function for bulk inserts with no manual memory allocation. It's only used in a small part of the function -- just to know what types to insert -- but ends up affecting all control flow.

[warmwaffles]

I wonder how often this feature would be used and if it is worth it to solve in the NIF.

[ruslandoga]

All Repo.insert_all calls would go through this NIF and the performance improvement is quite significant (seems close to x2 from my benchmarks, but potentially higher because it allows reusing a single prepared statement for all inserts into a table) so the I think it's worth it.