Iox #59 memory abstraction modularization roudi fixes

[marthtz]

No description provided.

[elBoberido]

I think most of the changes in iceoryx_utils/test/moduletests/test_relative_pointer.cpp are due to merge errors

@elfenpiff @MatthiasKillat can you please check

[elBoberido]

@elfenpiff please check the windows build ... btw, when do we get our CI ;)

[Karsten1987]

btw, when do we get our CI ;)

Have you considered setting up GitHub Actions for it? Given that Iceoryx easily compiles with colcon, we could technically also set up the ROS2 Action CI for it.

[elBoberido]

btw, when do we get our CI ;)

Have you considered setting up GitHub Actions for it? Given that Iceoryx easily compiles with colcon, we could technically also set up the ROS2 Action CI for it.

@mossmaurice already has taken some steps for CI, I just don`t know how far we are. But running all the ROS2 test with iceoryx is definitely appealing

[dkroenke]

I checked on Raspberry Pi4 and there 5 testcases failing, will take a deeper look on that on monday.

[dkroenke]

I checked on Raspberry Pi4 and there 5 testcases failing, will take a deeper look on that on monday.

ok i checked the examples and they are at least running. For following up i created a new issue: #63

[dkroenke]

Tests are failing but this is adressed here: #63

[elfenpiff]

The issue states a problem with the relative ptr on 32-bit #63. This could mean that the examples are just running by accident. I would not merge this pull request until it is clarified if this is the case or till every unit test is running

[elfenpiff]

Operating system:
Gentoo Linux 5.5.8 (arm 64bit)

Compiler version:
GCC 9.2.0

The following tests are failing in utils_moduletests:
AtomicRelocatablePointer_test.memoryRelocation by signal SIGBUS (Misaligned address error)
ObjectPool_test.parameter_construct by signal SIGSEGV (Address boundary error)

relativeptrtests/0.ConstrTests
/home/larry/iceoryx/iceoryx_utils/test/moduletests/test_relative_pointer.cpp:127 : Failure
Expected equality of these values:
rp.getId()
Which is: 2
1
/home/larry/iceoryx/iceoryx_utils/test/moduletests/test_relative_pointer.cpp:143 : Failure
Expected equality of these values:
rp.getId()
Which is: 2
1
/home/larry/iceoryx/iceoryx_utils/test/moduletests/test_relative_pointer.cpp:152 : Failure
Expected equality of these values:
rp.getId()
Which is: 2
1

relativeptrtests/1.ConstrTests

• same is above
  relativeptrtests/2.ConstrTests
• same is above

The following tests are failing in posh_moduletests:
[ RUN ] PosixShmMemoryProvider_Test.CreateMemory
Reserving 1024 bytes in the shared memory [/FuManchu]
fish: “./posh_moduletests” terminated by signal SIGBUS (Misaligned address error)

[elfenpiff]

Operating system:
Gentoo Linux 5.5.9 (x86_64)

Compiler version:
GCC 9.2.0

The following tests are failing in utils_moduletests:
relativeptrtests/0.ConstrTests
/home/elchris/Development/marthtz-iceoryx/iceoryx_utils/test/moduletests/test_relative_pointer.cpp:127: Failure
Expected equality of these values:
rp.getId()
Which is: 2
1
/home/elchris/Development/marthtz-iceoryx/iceoryx_utils/test/moduletests/test_relative_pointer.cpp:143: Failure
Expected equality of these values:
rp.getId()
Which is: 2
1
/home/elchris/Development/marthtz-iceoryx/iceoryx_utils/test/moduletests/test_relative_pointer.cpp:152: Failure
Expected equality of these values:
rp.getId()
Which is: 2
1

relativeptrtests/1.ConstrTests

• same is above
  relativeptrtests/2.ConstrTests
• same is above

The following tests are failing in posh_moduletests:
/home/elchris/Development/marthtz-iceoryx/iceoryx_posh/test/moduletests/test_mq_interface.cpp:121: Failure
Actual function call count doesn't match EXPECT_CALL(*time_MOCK::mock, clock_gettime(_, _))...
Expected: to be called twice
Actual: called once - unsatisfied and active
[ FAILED ] CMqInterface_test.MqInterfaceCreator_TimedReceive

/home/elchris/Development/marthtz-iceoryx/iceoryx_posh/test/moduletests/test_mq_interface.cpp:121: Failure
Actual function call count doesn't match EXPECT_CALL(*time_MOCK::mock, clock_gettime(_, _))...
Expected: to be called twice
Actual: called once - unsatisfied and active
[ FAILED ] CMqInterface_test.MqBase_TimedReceive

/home/elchris/Development/marthtz-iceoryx/iceoryx_posh/test/moduletests/test_mq_interface.cpp:121: Failure
Actual function call count doesn't match EXPECT_CALL(*time_MOCK::mock, clock_gettime(_, _))...
Expected: to be called twice
Actual: called once - unsatisfied and active
[ FAILED ] CMqInterface_test.MqInterfaceUser_TimedReceive

[MatthiasKillat]

As for the relative pointer tests in relativeptrtests/0.ConstrTests you are seemingly missing an important fix in pointer_repository.hpp line 62 and 79:
m_info[id].endPtr = reinterpret_cast<ptr_t>(reinterpret_cast<uint64_t>(ptr) + size);
has to be changed to
m_info[id].endPtr = reinterpret_cast<ptr_t>(reinterpret_cast<uint64_t>(ptr) + size - 1);

Then this issue should not occur (we already have this on develop).

[MatthiasKillat]

AtomicRelocatablePointer_test.memoryRelocation by signal SIGBUS (Misaligned address error)

is interesting and I cannot reproduce it. I suggest disabling this test for now as the capability that is tested here (relocation of a memory segment by copy) is not used currently.
I suspect it has something to do with the fact that the memory segment copied by memcpy contains structures which are therefore indirectly copied as well. However, this memcpy does not automatically satisfy the alignment restrictions of the objects which leads to the error on use.
This would be interesting insofar as it shows that arbitrary relocation (which we are not performing in iceoryx) is not possible in general. It is probably sufficient to copy in a way that the first structure stored in the memory segment is correctly aligned which will ensure that all others are correctly aligned as well (since they were before the copy).

I will take a look at this issue in more detail but until then disabling the test is recommended.

[elBoberido]

As for the relative pointer tests in relativeptrtests/0.ConstrTests you are seemingly missing an important fix in pointer_repository.hpp line 62 and 79:
m_info[id].endPtr = reinterpret_cast<ptr_t>(reinterpret_cast<uint64_t>(ptr) + size);
has to be changed to
m_info[id].endPtr = reinterpret_cast<ptr_t>(reinterpret_cast<uint64_t>(ptr) + size - 1);

Then this issue should not occur (we already have this on develop).

It's already there, but I reverted to much from the test. This is fixed now.

[elBoberido]

AtomicRelocatablePointer_test.memoryRelocation by signal SIGBUS (Misaligned address error)

is interesting and I cannot reproduce it. I suggest disabling this test for now as the capability that is tested here (relocation of a memory segment by copy) is not used currently.
I suspect it has something to do with the fact that the memory segment copied by memcpy contains structures which are therefore indirectly copied as well. However, this memcpy does not automatically satisfy the alignment restrictions of the objects which leads to the error on use.
This would be interesting insofar as it shows that arbitrary relocation (which we are not performing in iceoryx) is not possible in general. It is probably sufficient to copy in a way that the first structure stored in the memory segment is correctly aligned which will ensure that all others are correctly aligned as well (since they were before the copy).

I will take a look at this issue in more detail but until then disabling the test is recommended.

I think I have a solution for this as well