Prepare to merge dqlite-next into master

[cole-miller]

Let's do all our work on one branch, with experimental stuff (like thread pool and vfs2 integration) guarded behind #ifdef DQLITE_NEXT. This will avoid a big merge headache down the line.

Signed-off-by: Cole Miller [email protected]

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 80.56%. Comparing base (4d74efa) to head (ec33e4f).
Report is 3 commits behind head on dqlite-next.

Additional details and impacted files

@@             Coverage Diff              @@
##           dqlite-next     #642   +/-   ##
============================================
  Coverage        80.56%   80.56%           
============================================
  Files              196      196           
  Lines            28300    28304    +4     
  Branches          5297     5300    +3     
============================================
+ Hits             22799    22804    +5     
+ Misses            3808     3767   -41     
- Partials          1693     1733   +40     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[cole-miller]

@just-now I think this is good to go, would appreciate you confirming that I didn't miss anything important when adding ifdefs.

[letFunny]

It is a little bit tricky to review but I tried to have master and this PR side by side to see if they had the same behavior when DQLITE_NEXT was not defined. Looks good to me apart from one comment I included, thanks!

[letFunny]

I may be wrong here but I think the label that exists in master at the moment is err_after_loop_init so the if def should be:

....
err_after_raft_transport_init:
	raftProxyClose(&d->raft_transport);
#ifdef DQLITE_NEXT
err_after_pool_init:
	pool_close(&d->pool);
	pool_fini(&d->pool);
#endif
err_after_loop_init:
	uv_loop_close(&d->loop);
....

[cole-miller]

That way of doing it doesn't quite work as-is because on dqlite-next there is a goto err_after_pool_init in the error branch after raftProxyInit above. We would have to change that to

        rv = raftProxyInit(&d->raft_transport, &d->loop);
        if (rv != 0) {
#ifdef DQLITE_NEXT
                goto err_after_pool_init;
#else
                goto err_after_loop_init;
#endif
        }

The way I've written it looks kind of weird but gets away with only one ifdef, which I think is nice. Thoughts?

[cole-miller]

Pushed 24a4879 to codify that dqlite-next relies (will rely) on using the bundled raft sources and headers, since we will be making incompatible changes like #639.

[just-now]

Looks good to me.