GCC compiler warnings in byte order reversion code

Branch [archived branch] has been created by Participant.

[revision removed]


Detailed log of new commits:

Author: dln
Date: Tue Sep 9 14:54:08 2014 +0400

    0024537: GCC compiler warnings in byte order reversion code
    
    - preliminary version of the fix for test

Branch [archived branch] has been updated forcibly by Participant.

[revision removed]

Branch [archived branch] has been updated forcibly by Participant.

[revision removed]

Dear abv,

check it please

Branch [archived branch] has been created by Participant.

[revision removed]


Detailed log of new commits:

Author: dln
Date: Tue Sep 9 14:54:08 2014 +0400

    0024537: GCC compiler warnings in byte order reversion code
    
    - warnings in byte order inversion functionality is fixed with unions
    - test case (simulation of conversion to big endian) is added

Please review again.

Sorry, the new test case fails on x64 platform. It needs to be reworked.

Branch [archived branch] has been updated forcibly by Participant.

[revision removed]

The test case has been reworked. Now it is completely hard-coded inside the draw command OCC24537.

Andrey, please review the branch CR24537.

I suggest that some corrections should be made:

1. In FSD_FileHeader.hxx:
   - it can be worth moving this code to cxx or separate header (in BinTools?), to avoid defining this stuff (very special and not used in practice) globally
   - it can be worth renaming macro DO_INVERSE to something more specific, at least prefix it with "OCCT_"
   - it can be worth adding Standard_StaticAssert to check that sizeof(ShortReal) == sizeof(Integer) and sizeof(Real) == 2 * sizeof(Integer) etc.; alternatively templated version can be used (just as for Size)
   - InverseSizeSpecialized <4> can be just a call to InverseInt()
   - InverseLong() is obviously not correct for 64-bit systems where long is 8 bytes in size; it can be removed as unused, or implemented in the same way as for Size

2. In BinObjMgt_Persistent.cxx:
   - I believe this will not work as before on 32-bit systems, since before aData was double* (incremented by 8 bytes) and now &aWrapUnion.anAddrData is void** (incremented by 4 bytes):

- aPrevPtr = &aData[aLenInPiece / BP_REALSIZE];
+ aPrevPtr = (&aWrapUnion.anAddrData)[aLenInPiece / BP_REALSIZE];

   - I deem that casting char* to double* is generally bad idea, since on some architectures this may cause alignment failure. Why not just to move along myData by 8-byte sequences, copying them to buffer, then converting and placing back?

3. In QABugs_19.cxx: I suggest using more disperse set of probing values, not just 1-2-3. For instance:

   - int: -4000000, -1234, 0, 1, 1234, 4000000
   - double: -1e300, -1e-9, 0., 1.e-9, 1., 3.1415296, 1e100
   - float: -1e-9, 0., 1.e-9, 1., 3.1415

Branch [archived branch] has been updated forcibly by Participant.

[revision removed]

I have considered some remarks. It is needed to continue the fix. The current version of the branch CR24537 must be re-based on current master before continuing. It can be not compilable.

Branch [archived branch] has been created by Participant.

[revision removed]


Detailed log of new commits:

Author: rkv
Date: Tue Sep 22 12:48:04 2015 +0300

    0024537: GCC compiler warnings in byte order reversion code
    
    - warnings in byte order inversion functionality is fixed with unions
    - test case (simulation of conversion to big endian) is added

Reviewed.

Dear Commenter 2,
Branch CR24537_3 from occt git-repository (and master from products git-repository) was compiled on Linux, MacOS and Windows platforms and tested on Release mode.
[revision removed]

Number of compiler warnings:

occt component :
Linux: 13 (15 on master)
Windows: 0 (0 on master)

products component :
Linux: 39 (39 on master)
Windows: 0 (0 on master)

There is new additional compilation warning on Linux platform:
There are compilation errors in Qt samples products on windows platform:

Regressions/Differences/Improvements:
No regressions/differences

Testing cases:
http://occt-tests/CR24537-3-master-occt-64/Debian70-64/bugs/fclasses/bug24537.html
http://occt-tests/CR24537-3-master-occt-64/Windows-64-VC10/bugs/fclasses/bug24537.html
bugs fclasses bug24537: OK

Testing on Linux:
occt component :
Total MEMORY difference: 92282120 / 92105726 [+0.19%]
Total CPU difference: 18763.229999999014 / 18665.439999998966 [+0.52%]
products component :
Total MEMORY difference: 25921386 / 25980951 [-0.23%]
Total CPU difference: 7418.7299999999905 / 7208.589999999996 [+2.92%]

Testing on Windows:
occt component :
Total MEMORY difference: 56806777 / 56823447 [-0.03%]
Total CPU difference: 18049.986504299017 / 16808.499346099128 [+7.39%]
products component :
Total MEMORY difference: 16785580 / 16790438 [-0.03%]
Total CPU difference: 5934.1844394 / 5526.49142599996 [+7.38%]

There are no differences in images found by testdiff.

Branch [archived branch] has been deleted by Commenter 1.

[revision removed]

Branch [archived branch] has been deleted by Commenter 1.

[revision removed]

Branch [archived branch] has been deleted by Commenter 1.

[revision removed]