"Class-memaccess" compilation error in test [C++, gcc 10.1.0, Flatbuffers 1.12.0/master]

[bakinovsky-m]

Compilation is failing on tests when using gcc 10.1.0.

flatbuffers/tests/arrays_test_generated.h: In constructor ‘MyGame::Example::ArrayStruct::ArrayStruct(float, int8_t, int32_t)’:
flatbuffers/tests/arrays_test_generated.h:156:34: error: ‘void* memset(void*, int, size_t)’ clearing an object of non-trivial type ‘struct MyGame::Example::NestedStruct’; use assignment or value-initialization instead [-Werror=class-memaccess]
  156 |     std::memset(d_, 0, sizeof(d_));
      |                                  ^
flatbuffers/tests/arrays_test_generated.h:66:40: note: ‘struct MyGame::Example::NestedStruct’ declared here
   66 | FLATBUFFERS_MANUALLY_ALIGNED_STRUCT(8) NestedStruct FLATBUFFERS_FINAL_CLASS {
      |                                        ^~~~~~~~~~~~

Master branch and v1.12.0 tag are affected, didn't test other tags

Setup

• gcc 10.1.0
• flatbuffers 1.12.0 or master

Reproducing

• cd to flatbuffers folders
• run:

$ mkdir build && cd build
$ cmake .. #FLATBUFFERS_BUILD_TESTS is enabled by default
$ cmake --build .

[aardappel]

Yes, the default constructor really shouldn't use memset, we could instead generate a constructor similar to the one with arguments, but with 0 for each field value.

Care to make a PR for this?

[bakinovsky-m]

As I understand, the problem is in constructor with parameters (not in default constructor, as it's already have static_cast<void*>(...) fix)
It appears when std::memset applies to NestedStruct array

I have some thoughts of how to fix this, so yeah, i will make the PR

[Qengineering]

I've exactly the same error with TensorFlow 2.4.0 tag on Ubuntu 20.10 and gcc 10.2.0.

Oddly, there is no problem with TF 2.4.0 on Ubuntu 20.04 + gcc 9.3.0, nor on Ubuntu 18.04 + gcc 7.5.0.
Both compile the flatbuffers flawlessly, while Ub20.10 + gcc 10.2.0 throws the above class-memaccess errors

[aardappel]

@Qengineering as you can see, it has been fixed.. so you'd need to get TensorFlow to update their copy of FlatBuffers.

As a local work-around, turn of -Werror for the files involved.

[Qengineering]

It's fixed was also what I thought after reading the comments above. Never had any problem before with the FlatBuffers.

I followed your steps, and downloaded version TF 2.4.0 again, hoping, against better, it will solve the issue.

$ wget -O tensorflow.zip https://github.com/tensorflow/tensorflow/archive/v2.4.0.zip
$ unzip tensorflow.zip
$ mv tensorflow-2.4.0 tensorflow
$ cd tensorflow
$ ./tensorflow/lite/tools/make/download_dependencies.sh
$ cd ~/tensorflow/tensorflow/lite/tools/make/downloads/flatbuffers
$ mkdir build
$ cd build
$ cmake ..
$ make -j4

Turning -Werror of is something I'd rather not do. After all, why issue an error if it has no bearing on functionality?
Using Ubuntu 20.10 with gcc 10.2.0. on a Raspberry Pi 4 with 2 GByte RAM. I'm just reporting the issue, so you know.

[vglavnyy]

Probably this issue connected with #6337.
@aardappel
Maybe it's better to disable Werror by default and enable this flag only with CI?
Another option is to add overriding flags using an environment variable for example (FLATBUFFERS_CMAKE_CXX_FLAGS).

[aardappel]

@vglavnyy yeah, that be a good idea. I like -Werror, but it keeps producing problems when compilers introduce new warnings, and we can't test them all. It shouldn't stop people from building.

And generally allowing people their own control over compiler flags is good.

[baqwas]

Hello,
I was following the tutorial by @Qengineering and ran into the exact problem that he commented on Jan 5. This issue is far above my job grade. I need to use RPi02 with Bullseye. My system is:

`
$ uname -a

Linux hostname 5.10.63-v7+ #1488 SMP Thu Nov 18 16:14:44 GMT 2021 armv7l GNU/Linux
`

Basic processor info:
`
processor : 3
model name : ARMv7 Processor rev 4 (v7l)
BogoMIPS : 38.40
Features : half thumb fastmult vfp edsp neon vfpv3 tls vfpv4 idiva idivt vfpd32 lpae evtstrm crc32
CPU implementer : 0x41
CPU architecture: 7
CPU variant : 0x0
CPU part : 0xd03
CPU revision : 4

Hardware : BCM2835
Revision : 902120
Model : Raspberry Pi Zero 2 Rev 1.0
`

cmake does not report any error. Unfortunately, make reports the following error that suspends all subsequent install and ldconfig operations:

/home/pi/projects/tensorflow/tensorflow/lite/tools/make/downloads/flatbuffers/tests/arrays_test_generated.h:156:34: error: ‘void* memset(void*, int, size_t)’ clearing an object of non-trivial type ‘struct MyGame::Example::NestedStruct’; use assignment or value-initialization instead [-Werror=class-memaccess] 156 | std::memset(d_, 0, sizeof(d_));

Would appreciate indulgence for a newbie question from anyone who has a basic workaround. Thanks.

Kind regards.

[aardappel]

@baqwas again, this has been fixed in FlatBuffers, so ask about it on the tf/tflite repo.. they need to upgrade. Or force disable -Werror.

[baqwas]

Thanks @aardappel for the prompt response. Will do as advised. Truly appreciate your indulgence.

Kind regards.