search

SciSharp / TensorFlow.NET

.NET Standard bindings for Google's TensorFlow for developing, training and deploying Machine Learning models in C# and F#.
https://scisharp.github.io/tensorflow-net-docs
Apache License 2.0
3.21k stars 516 forks source link

[Question]: FileLoadException OneOf.dll #1042

Open FrancescoRusticali opened 1 year ago

FrancescoRusticali commented 1 year ago

Description

Hi all, I'm having some issues trying to compile and run the current master branch. I get a FileLoadException while trying to do anything that needs to call OneOf.dll. This is the error:

'System.IO.FileLoadException: 'Could not load file or assembly 'OneOf, Version=1.0.0.0, Culture=neutral, PublicKeyToken=null' or one of its dependencies. A strongly-named assembly is required. (Exception from HRESULT: 0x80131044)'

For example I get it while calling keras.Input.

Anyone having the same issue?

Alternatives

No response

AsakusaRinne commented 1 year ago

Hi, did you open some settings in VS that enables strongly-name signature automatically? And what if uninstalling Oneof and installing it again?

FrancescoRusticali commented 1 year ago

Hi, I tried to uninstall and install OneOf, but I still get the error. Calling OneOf directly from my code works fine, the problems seems to arise when calling Tensorflow methods that use OneOf. I think this should be related to Tensorflow being signed, while OneOf is not. This comes from the csproj file:

    <Reference Include="OneOf, Version=1.0.0.0, Culture=neutral, processorArchitecture=MSIL">
      <HintPath>..\packages\OneOf.3.0.243\lib\net45\OneOf.dll</HintPath>
    </Reference>
    <Reference Include="Tensorflow.Binding, Version=1.0.0.0, Culture=neutral, PublicKeyToken=cc7b13ffcd2ddd51, processorArchitecture=MSIL">
      <HintPath>..\packages\TensorFlow.NET.1.0.0\lib\netstandard2.0\Tensorflow.Binding.dll</HintPath>
    </Reference>
    <Reference Include="Tensorflow.Keras, Version=1.0.0.0, Culture=neutral, PublicKeyToken=cc7b13ffcd2ddd51, processorArchitecture=MSIL">
      <HintPath>..\packages\TensorFlow.Keras.1.0.0\lib\netstandard2.0\Tensorflow.Keras.dll</HintPath>
    </Reference>
FrancescoRusticali commented 1 year ago

To give more info: this seems to be a general problem with unsigned thirt party dlls. I get similar error in version 0.100.4 related to HDF5CSharp.dll trying to save model weights in h5 format with save_weights. HDF5CSharp also is not signed:

    <Reference Include="HDF.PInvoke.1.10, Version=1.10.612.0, Culture=neutral, processorArchitecture=MSIL">
      <HintPath>..\packages\HDF.PInvoke.1.10.1.10.612\lib\netstandard2.0\HDF.PInvoke.1.10.dll</HintPath>
    </Reference>
    <Reference Include="HDF5CSharp, Version=1.17.0.0, Culture=neutral, processorArchitecture=MSIL">
      <HintPath>..\packages\HDF5-CSharp.1.17.0\lib\netstandard2.0\HDF5CSharp.dll</HintPath>
    </Reference>

I also found #915 that seems related to that. It's closed as resolved, but I don't understand how. Is there any workaround?

AsakusaRinne commented 1 year ago

Here may be a possible work around https://stackoverflow.com/questions/290980/error-a-strongly-named-assembly-is-required. BTW, do you build from source yourself because you want to use the latest master code, or you want to customize tf.net for your project? If you just want to use master branch code, maybe we can add an automatic workflow to make nightly releases on myget. Besides, does the error appeared on your PC when using v0.100.4 or other release of tf.net?

FrancescoRusticali commented 1 year ago

Thank you for the suggestion. From the stackoverflow link, a possible workaround can indeed be to build master source code without the "sign" flag.

BTW, do you build from source yourself because you want to use the latest master code, or you want to customize tf.net for your project?

No, I just would like to use latest master code. Nightly releases would be great!

Besides, does the error appeared on your PC when using v0.100.4 or other release of tf.net?

The error about OneOf appeared on my pc only with latest master code, but the OneOf dependency was not there in v0.100.4. Instead, the error about HDF5CSharp appeared on my PC also with v0.100.4, installed with nuget.

AsakusaRinne commented 1 year ago

Version v0.100.5 has come out and we also added nightly release, which is available by adding https://www.myget.org/F/scisharp/api/v3/index.json to your nuget source. Could you please have a try with v0.100.5?

Superpiffer commented 1 year ago

Version v0.100.5 has come out and we also added nightly release, which is available by adding https://www.myget.org/F/scisharp/api/v3/index.json to your nuget source. Could you please have a try with v0.100.5?

I have the same issue with v0.100.5 and only with .NET 4.8 (no problem with .NET 6):

image

AsakusaRinne commented 1 year ago

It's because tf.net signs the assembly while Oneof does not. I've submitted an issue to Oneof. I think we should give up Oneof before the next release if it won't be signed.

CAMB-dev commented 1 year ago

Maybe you can try installing the package OneOf https://github.com/mcintyre321/OneOf

Superpiffer commented 1 year ago

It's because tf.net signs the assembly while Oneof does not. I've submitted an issue to Oneof. I think we should give up Oneof before the next release if it won't be signed.

Any news about this issue? New incoming version still has OneOf as a dependency.