User:Totktonada

I'm Alexander Turenko. Tarantool developer.

Nice ideas

• Built-in help (issue TBD).
• Swiss army knife to overview CI results (issue).

Disquiets and fears

Listed things, which should not stall forever in my opinion. The order is random.

• Popen track is stalled.
• No good tool and/or service for deployment of modules to S3 based repositories.
  • Example: attempt to use update_repo.sh involves too much code.
  • It blocks full deprecation of packagecloud.
• net.box connection from older to newer tarantool may fail.
• JDBC track is stalled.
• Not upgraded mode is not tested.
• No recipe to upgrade a partially upgraded instance.
• No tests for net.box and _vcollation bug (it's my debt).
  • There is agreement how to implement them.
• No testing for PRs from external repositories (but it is in our focus, that's good).
• No tarantool-release packages (but I have PoC).
• libev is old.
• libcoro is old (Igor Munkin said that there were ARM related changes).
• libyaml is relatively old.
• Whether out extended SQL result set metainfo fit good with JDBC?
• Performance testing: no powerful owner.
  • Back box testing: no deep understanding of workloads.
  • Perf. testing in docker, what a hell.
  • bench.tarantool.org is far from being usable as everyday tool
• Tuples marshalling without reencoding track is stalled.
• No good LuaJIT testing.
• Still some jobs on Travis-CI, who own the transition plan?
• No good Lua memory introspection tool.
• Not all tests are run in out-of-source build.
• No testing on gcc-4.8.5, which is default on CentOS 7.
• Lua output track looks stalled.
• Stalled PRs and issues in many repositories (modules, connectors).
• No per-push testing of modules and connectors.
• No testing of tarantoolctl rocks.
• We should eventually upstream most of our patches to luarocks.
• test-run does not support Python 3.
• test-run way to assign a port involves races.
• We should consider support of luatest test files in test-run.
• test-run is not convenient for writing a multi instance test.
• Backport LuaJIT upstream changes (at least setup some tracking process).
  • Subtask: general + x86_64.
  • Subtask: ARM (which one?).
• There are no list of prerequisites to announce first LTS release series with SQL.
• Our docker images are in some unclear state (the link is to discussion, not about per-push / per-release deployments).
• Stalled PRs / issues in packpack.
• Still no full merger documentation.
• Binary protocol documentation is far from being convenient for a connector developer.
• No good documentation how to implement a module.
• No good documentation how to implement a connector (pitfalls and so).
• No recipes how to work with RPM / Deb packages (different scenarious).
• Unclear guarantees about upgrade process (it is not based on testing, right?).
• smtp module unable to link with bundled tarantool libcurl.
• Old tarantool versions on try.tarantool.org.
• fstack in avro-schema can be reverted in favor of the new hack provided by Nick Zavaritsky (perf. gain is possible).
• No agreement re Lua/C function comments style (read from words 'The function has the Lua API').
• No general description of SQL type system (AFAIK).
• No SQL type arithmetic AFAIK (an expression type is not result of a type arithmetic, but deduced from a value).
• There were thoughts around treating numeric literals as decimals, but there are no work on this. I guess presence of a more wide numeric type may affect type arithmetic at whole: say, how overflows are handled.
• No support of compound statements in SQL.