Fix off by one error in frontier req server

[orhanhenrik]

The current implementation will send 1 too many items, since count starts at 0.

[orhanhenrik]

Any idea why the test is failing? I cannot see how it would be related to my code. The test active_difficulty.recalculate_work is failing on the macos build.

[cryptocode]

frontier_req_server::sent_action increments the count before calling send_next. Doesn't that make the original code correct?

[orhanhenrik]

If the wanted count is 1, and you run the code, the following happens:

1. send_next (count=0), 0<=1, ok
2. sent_action, increment count=1
3. send_next (count=1), 1<=1, ok
4. sent_action, increment count=2
5. send_next (count=2), 2<=1, not ok

So the server will send 2 results and then exit if I have understood the code correctly.

There should probably be a test for this as well, but I'm not very good at C++ and I couldn't even get the project to build on my mac, so it's a bit hard to write and check a test..

[orhanhenrik]

Tests are passing on travis here: https://travis-ci.org/orhanhenrik/nano-node, so I'm not sure what's wrong with this test run. Maybe a race condition in the code?

[cryptocode]

I see, send_next is initially called from the visitor. This looks like it'll fix it.

I think clients currently always use max here (2^31-1), which might be why tests doesn't pick it up. We can look into fixing the test in a separate PR. The test failure is unrelated, probably one of the tests failing intermittently on CI.

[orhanhenrik]

Yeah, I noticed that max is always used. I’m writing a separate node implementation and was getting the wrong number of results, so that’s how I found this minor bug.

[SergiySW]

LGTM