This stage failed.

@Jens-G thanks for pointing out the failures. Looking at several of them it appears to be because I added to test/StessTest.thrift but only updated test/cpp/src/StessTest.cpp and NOT lib/d/test/stress_test_server.d so there is an unimplemented function.
The fix I made was only under the c++ server; let me see if adding a test under lib/cpp/test/ would be better contained and cover only the c++ server.

This looks promising and relevant. @slicking , there is a small favor that would be very helpful in moving this forward. Could you provide a separate PR with a test case, that shows that indeed client messages would bring remainingMessageSize_ to zero (or below), thereby causing an EOD_OF_FILE exception? If this can be demonstrated in a small test, it would point to a severe problem in current thrift, and nicely motivate this PR.

Do you think you could provide a test?

Do you think Message Size check is implemented the right way ?
Is this not a protocol concern, rather than a distributed protocol / transport responsibility ?
The functionality is implemented through transport base class function definition. Don't you think it would be better to move these to a different object which responsibility would be to check message size validity ?
Several functions in the implementation are never called (in production code, I haven't checked the tests), code is hard to read and some paths are never reached (resetConsumedMessageSize(long newSize = -1) with actual newSize value).
I do not know enough about the whole Virtual protocol / transport implementation to propose relevant changes to the code, but the current one (I use version 0.14.3) broke my program !

I've moved from 0.11.0 to 0.20.0 and encountered the issue described here, fix in this PR seems to solve it, is there a plan for this change to be merged?

I merged the fix part in 86b05bf, but left the rest, as there are unfortunately still open questions etc.

Thank you, I guess this will be in next thrift release, when do you plan it?

Is the remainder of this PR still valid? Anyone able to comment on the remaining issues?

Thanks @Jens-G for merging the fix; I'm glad it could be of help to a few others.
I've not had time to get back to this issue and update the test; with the fix merged though I think this PR can be closed. The changes to update the test in this PR need to be reworked to be specific to the c++ server and, if anyone is able, can be taken up in a separate PR.