rodeos and eosio-tester

[tbfleming]

Change Description

This builds on #8750

Consensus Changes

• Consensus Changes

API Changes

• API Changes

Documentation Additions

• Documentation Additions

[spoonincode]

I realize this is an opinion since we haven't come to a strategy as a team, but it'd be nice to not add new external deps unless needed. In this case, it seems like gflags has great documented support for being included as a git submodule and simply built as part of our eosio cmake "super project". Could we try that instead of bringing in a new external dep?

[tbfleming]

Yep. That could solve some namespace differences I'm seeing between the different distros.

[tbfleming]

The question is: how can I point rocksdb's cmake to it without modifying rocksdb's cmake

[spoonincode]

actually, do we even need gflags support in rocksdb? Appears it can be built without it
CMAKE_DEPENDENT_OPTION(WITH_GFLAGS "build with GFlags" ON

[spoonincode]

We may need to install rocksdb's license file. See toward end of root CMakeLists.txt.

Actually there may be two license files that we need, not sure about that leveldb license file.

[tbfleming]

I suspect we need both

[tbfleming]

@leordev is going to add rabbitmq support. We need to decide how to bring in that library. submodule?

[spoonincode]

I think we need to place the rapidjson in use by fc in to a separate namespace as it will be a different version of what abieos uses and may in theory cause an ODR violation? I will take care of that. Just blocking PR until I get that fix in..

[tbfleming]

We also need to make sure that every piece of code which uses

#include <rapidjson/reader.h>

pulls in the right one.

removing rapidjson block

[spoonincode]

A couple build issues I've noticed:

abieos cmake lists cmake 3.11 as required. Is it actually required? using cmake 3.10 seemed to work for me. eosio current minimum is 3.8 so this change would move that requirement.
I also wonder if we really need the set(CMAKE_OSX_DEPLOYMENT_TARGET 10.14) in abieos any more that is kinda hacky.

gcc8 & 9 are failing due to line 53 in state_history_plugin.cpp -- another one of those member name same as type name issues.

gcc7 ICEs.

[spoonincode]

For rocks, we're still building tests (WITH_TESTS) and benchmarks (WITH_BENCHMARKS_TOOLS) and tools (WITH_TOOLS). Do we need all of these?

[spoonincode]

bind_front_handler, tcp_stream, and make_strand are boost 1.70 features and will fail with current eosio 1.67 minimum requirement.

[tbfleming]

If I remember right, make_strand came with a change to the threading interface used by beast. If we switch it to the previous interface, it won't build with 1.70. Maybe I'm remembering wrong.

[arhag]

We will address the gcc7 ICE by raising minimum gcc requirements to 8. We will address the Boost issues by increasing minimum Boost requirements to 1.70. These will be handled in other PRs.

This approval does not mean that I fully reviewed all of the new rodeos and tester code. Rather this PR review was focused on build matters and ensuring that all the new code and changes do not impact the behavior of nodeos.