PoseClustering_test fails on on Arch Linux using Clang 3.7

[dstoeckel]

PoseClustering_test fails in the the serializeWardClusterTree test with the following error message:

checking serializeWardClusterTree(std::ostream& out)... 
   TEST_FILE: line mismatch:
    got:      '22 serialization::archive 14 0 0 15 14 0 0 1 0 0 1 0.000000000e+00 1 0 1 1 0.000000000e+00 1 0 2 1 0.000000000e+00 1 0 3 1 0.000000000e+00 1 0 4 1 0.000000000e+00 1 0 5 1 0.000000000e+00 1 0 6 1 0.000000000e+00 1 0 7 1 0.000000000e+00 0 0 2 2.930124538e-05 0 0 3 1.017511487e+00 0 0 4 2.349177122e+00 0 0 2 9.453108311e-01 0 0 2 1.095688367e-05 0 0 4 5.062472916e+01 0 0 8 5.839497375e+01 8 1 0 0 8 0 9 3 9 8 10 2 10 9 11 7 11 6 12 4 12 5 13 12 13 11 14 13 14 10 14'
    expected: '22 serialization::archive 10 0 0 15 14 0 0 1 0 0 1 0 1 0 1 1 0 1 0 2 1 0 1 0 3 1 0 1 0 4 1 0 1 0 5 1 0 1 0 6 1 0 1 0 7 1 0 0 0 2 2.9301245e-05 0 0 3 1.0175115 0 0 4 2.3491771 0 0 2 0.94531083 0 0 2 1.0956884e-05 0 0 4 50.624729 0 0 8 58.394974 8 1 0 0 8 0 9 3 9 8 10 2 10 9 11 7 11 6 12 4 12 5 13 12 13 11 14 13 14 10 14'
    (line 1382: TEST_FILE(filename.c_str(), BALL_TEST_DATA_PATH(PoseClustering_wardtree.dat)): false - 
    FAILED

BALL is configured in Debug mode. I am using the following software versions:

• Clang 3.7.1
• Boost 1.60
• Eigen 3.2.8
• TBB 4.4

[tkemmer]

Ditto for Gentoo:

checking serializeWardClusterTree(std::ostream& out)... 
   TEST_FILE: line mismatch:
    got:      '22 serialization::archive 14 0 0 15 14 0 0 1 0 0 1 0.000000000e+00 1 0 1 1 0.000000000e+00 1 0 2 1 0.000000000e+00 1 0 3 1 0.000000000e+00 1 0 4 1 0.000000000e+00 1 0 5 1 0.000000000e+00 1 0 6 1 0.000000000e+00 1 0 7 1 0.000000000e+00 0 0 2 2.930124538e-05 0 0 3 1.017511487e+00 0 0 4 2.349177122e+00 0 0 2 9.453108311e-01 0 0 2 1.095688367e-05 0 0 4 5.062472916e+01 0 0 8 5.839497375e+01 8 1 0 0 8 0 9 3 9 8 10 2 10 9 11 7 11 6 12 4 12 5 13 12 13 11 14 13 14 10 14'
    expected: '22 serialization::archive 13 0 0 15 14 0 0 1 0 0 1 0.000000000e+00 1 0 1 1 0.000000000e+00 1 0 2 1 0.000000000e+00 1 0 3 1 0.000000000e+00 1 0 4 1 0.000000000e+00 1 0 5 1 0.000000000e+00 1 0 6 1 0.000000000e+00 1 0 7 1 0.000000000e+00 0 0 2 2.930124538e-05 0 0 3 1.017511487e+00 0 0 4 2.349177122e+00 0 0 2 9.453108311e-01 0 0 2 1.095688367e-05 0 0 4 5.062472916e+01 0 0 8 5.839497375e+01 8 1 0 0 8 0 9 3 9 8 10 2 10 9 11 7 11 6 12 4 12 5 13 12 13 11 14 13 14 10 14'
    (line 1382: TEST_FILE(filename.c_str(), BALL_TEST_DATA_PATH(PoseClustering_wardtree.dat)): false - 
    FAILED

• gcc 5.3.0
• Boost 1.60 (contrib)
• Eigen 3.1.3
• TBB 4.3

[philthiel]

This problem also occurs on all Linux slaves (different gcc versions, current ball_contrib).
However, what I'm curious about is your failure Daniel. The expected result is read from file 'PoseClustering_wardtree.dat' but the printed line does't match this file. Nevertheless, the produced result also doesn't match the expected one ; )

[philthiel]

The problem, most likely, is only a boost archive format version mismatch.
I have to double check that but the heading fields '22 serialization::archive 14' seems to be boost-internal and may change with archive format versions.

[anhi]

Yeah, that’s exactly the problem. Isn’t it possible to use regex matching in the tests when comparing to a file? We could then just replace the version string with a wildcard.

Von: philthiel <notifications@github.commailto:notifications@github.com>
Antworten an: BALL-Project/ball <reply@reply.github.commailto:reply@reply.github.com>
Datum: Mittwoch, 27. April 2016 um 09:32
An: BALL-Project/ball <ball@noreply.github.commailto:ball@noreply.github.com>
Betreff: Re: [BALL-Project/ball] PoseClustering_test fails on on Arch Linux using Clang 3.7 (#578)

The problem, most likely, is only a boost archive format version mismatch.
I have to double check that but the heading fields '22 serialization::archive 14' seems to be boost-internal and may change with archive format versions.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHubhttps://github.com//issues/578#issuecomment-214994113

[philthiel]

Under construction ...

[dstoeckel]

This problem also occurs on all Linux slaves (different gcc versions, current ball_contrib).
However, what I'm curious about is your failure Daniel. The expected result is read from file 'PoseClustering_wardtree.dat' but the printed line does't match this file. Nevertheless, the produced result also doesn't match the expected one ; )

I suppose I was switching between branches when I ran the test and somehow got an older version of the file lying around ...

Anyhow, the regular expression fix seems appropriate.

[philthiel]

Should be fixed with commit 6eb56bc

[dstoeckel]

Interestingly the PoseClustering_test still fails on Ubuntu 14.04 with boost 1.55, albeit for another reason. It seems that floating point numbers are encoded differently in newer versions. Oliver fixed the test for new versions in 201cd01 but broke it for older versions.

Probably the test should serialise and deserialise the PoseClustering object and just compare the object to the expected value instead of the string ...

[philthiel]

I'm not sure if I got your point but I think we have similar in mind: skip the file comparison, deserialize directly in the test and compare the clustertrees? If so, we simply could merge the serialize- and deserializeWardClusterTree subtests.

[dstoeckel]

Yes, that is what I meant. I would leave the deserialize test as it is (in the hope that this does not break based on the boost version) as it should make it easier to distinguish between broken serialisation and broken deserialisation.

[philthiel]

Solution commited 0604808 .

[dstoeckel]

Awesome! That fixes it for me.

[tkemmer]

I would leave the deserialize test as it is (in the hope that this does not break based on the boost version) as it should make it easier to distinguish between broken serialisation and broken deserialisation.

The deserialization using boost 1.55 is indeed broken (cf. #600, confirmed), but due to Philipp's serializeWardClusterTree fix and boost's backwards-compatibility, the old version of the PoseClustering_wardtree.dat file (before Oliver's changes) should now work again for us and all boost versions since 1.55.