RCORE-2223 reduce unnecessary table selections in replication

[ironage]

This uses the idea in #7734 and takes it a bit further. Although mutations on embedded objects are now ignored, the selections on their tables still made it into the history. In the test download this leads to a history like this:

Select table: class_top
Create object: ObjKey(1)
Select table: class_top__meta
Select table: class_top
Select table: class_top__meta
Select table: class_top
Select table: class_top_embedded1
Select table: class_top
Select table: class_top_embedded1_embedded2
Select table: class_top_embedded1
Select table: class_top_embedded1_embedded2
Select table: class_top_embedded1_embedded2_embedded5
Select table: class_top_embedded1_embedded2
Select table: class_top_embedded1_embedded2_embedded5
Select table: class_top_embedded1_embedded2_embedded6
Select table: class_top_embedded1_embedded2
Select table: class_top_embedded1_embedded2_embedded6
Select table: class_top_embedded1_embedded3
Select table: class_top_embedded1
Select table: class_top_embedded1_embedded3_embedded4
Select table: class_top_embedded1_embedded3
Select table: class_top_embedded1_embedded3_embedded4
Select table: class_top_embedded1_embedded3
Select table: class_top_embedded1_embedded3_embedded6
Select table: class_top_embedded1_embedded3
Select table: class_top_embedded1_embedded3_embedded6
Select table: class_top_embedded1
Select table: class_top_embedded1_embedded4
Select table: class_top_embedded1
Select table: class_top_embedded1_embedded4
Select table: class_top_embedded1
Select table: class_top_embedded1_embedded5
Select table: class_top_embedded1
Select table: class_top_embedded1_embedded5
Select table: class_top
Create object: ObjKey(2)
...

And with these changes we have the equivalent log:

Select table: class_top
Create object: ObjKey(1)
Create object: ObjKey(2)

...

In this particular test, I observe the following size difference:
Before changes:
Logical file size: 294M
History size: 34.5M
After changes:
Logical file size: 262M
History size: 1.5M

☑️ ToDos

• 📝 Changelog update
• 🚦 Tests (or not relevant)
• C-API, if public C++ API changed
• bindgen/spec.yml, if public C++ API changed

[ironage]

Edit: Fixed. This is in draft because I think collections are missing insertions for some reason. Details below:

Before:

Select table: flx_pending_bootstrap
Create object: ObjKey(1)
Select table: flx_pending_bootstrap_changesets
Select table: flx_pending_bootstrap
Select table: flx_pending_bootstrap_changesets
--------------------------------------------
Select table: flx_subscription_sets
Modify object: state on ObjKey(1)
Modify object: error on ObjKey(1)
--------------------------------------------
Select table: flx_pending_bootstrap_changesets
Select table: flx_pending_bootstrap
Select collection: changesets on ObjKey(1)
Collection insert at 1
Select table: flx_pending_bootstrap_changesets
--------------------------------------------
Select table: flx_pending_bootstrap_changesets
Select table: flx_pending_bootstrap
Select collection: changesets on ObjKey(1)
Collection insert at 2
Select table: flx_pending_bootstrap_changesets
Select table: flx_pending_bootstrap
Select collection: changesets on ObjKey(1)
Collection insert at 3
Select table: flx_pending_bootstrap_changesets
--------------------------------------------
Select table: flx_pending_bootstrap_changesets
Select table: flx_pending_bootstrap
Select collection: changesets on ObjKey(1)
Collection insert at 4
Select table: flx_pending_bootstrap_changesets
Select table: flx_pending_bootstrap
Select collection: changesets on ObjKey(1)
Collection insert at 5
Select table: flx_pending_bootstrap_changesets
Select table: flx_pending_bootstrap
Select collection: changesets on ObjKey(1)
Collection insert at 6
Select table: flx_pending_bootstrap_changesets
Select table: flx_pending_bootstrap
Select collection: changesets on ObjKey(1)
Collection insert at 7
Select table: flx_pending_bootstrap_changesets
--------------------------------------------
Select table: flx_pending_bootstrap_changesets
Select table: flx_pending_bootstrap
Select collection: changesets on ObjKey(1)
Collection insert at 8
Select table: flx_pending_bootstrap_changesets
Select table: flx_pending_bootstrap
Select collection: changesets on ObjKey(1)
Collection insert at 9
Select table: flx_pending_bootstrap_changesets
Select table: flx_pending_bootstrap
Select collection: changesets on ObjKey(1)
Collection insert at 10
Select table: flx_pending_bootstrap_changesets
Select table: flx_pending_bootstrap
Select collection: changesets on ObjKey(1)
Collection insert at 11
Select table: flx_pending_bootstrap_changesets
--------------------------------------------

After:

Select table: flx_pending_bootstrap
Create object: ObjKey(1)
--------------------------------------------
Select table: flx_subscription_sets
Modify object: state on ObjKey(1)
Modify object: error on ObjKey(1)
--------------------------------------------
Select table: flx_pending_bootstrap
Select collection: changesets on ObjKey(1)
Collection insert at 1
--------------------------------------------
Select table: flx_pending_bootstrap
Select collection: changesets on ObjKey(1)
Collection insert at 2
--------------------------------------------
Select table: flx_pending_bootstrap
Select collection: changesets on ObjKey(1)
Collection insert at 6
--------------------------------------------
Select table: flx_pending_bootstrap
Select collection: changesets on ObjKey(1)
Collection insert at 10

[coveralls-official]

Pull Request Test Coverage Report for Build james.stone_582

Details

• 304 of 359 (84.68%) changed or added relevant lines in 3 files are covered.
• 68 unchanged lines in 13 files lost coverage.
• Overall coverage decreased (-0.006%) to 91.094%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
src/realm/replication.cpp	34	44	77.27%
test/test_replication.cpp	263	308	85.39%

Files with Coverage Reduction	New Missed Lines	%
src/realm/mixed.cpp	1	86.46%
src/realm/sort_descriptor.cpp	1	94.06%
test/fuzz_tester.hpp	1	57.73%
src/realm/replication.cpp	2	80.55%
src/realm/replication.hpp	2	81.0%
src/realm/sync/noinst/client_impl_base.cpp	2	83.24%
src/realm/sync/client.cpp	3	90.96%
src/realm/sync/network/network.cpp	3	87.95%
src/realm/sync/noinst/protocol_codec.hpp	4	73.82%
src/realm/util/assert.hpp	4	87.1%

Totals
Change from base Build 2537:	-0.006%
Covered Lines:	217067
Relevant Lines:	238289

---

💛 - Coveralls

[tgoyne]

Reading the values here and passing them to do_select_collection() rather than passing the collection in is a bunch of extra instructions for no benefit and the structure of this means that the compiler can't inline it away.

The idea behind this division is that select_collection() gets inlined into each of the callers so that no function call is needed in the common case of the collection already being selected, but the less common case is outlined rather than being repeated for each called. This is a fairly dubious optimization that we haven't actually measured in the last decade, but it theoretically could be useful.

[ironage]

Good observation. I cleaned it up a bit, hopefully this makes it easier for the compiler to make optimizations.

[tgoyne]

There doesn't appear to be any obvious reason we have to check this every time rather than caching it and only checking when selecting an object?

[ironage]

That is a valid optimization, changed.