Linux support

[dromer]

Duh.

[DamRsn]

Contributions are welcome 🙂

The building process should be very similar to OSx.

[gnac]

I've made some steps to build this (and the /libonnxruntime-neuralnote dependency) in Linux. At the moment, I've got NeuralNote compiling but it fails in the linker stage. I'll push what I have to my NeuralNote and onnxruntime lib forks. But in short, I'm currently stuck on the following error:

[100%] Built target NeuralNote_Standalone
/usr/lib64/gcc/x86_64-suse-linux/12/../../../../x86_64-suse-linux/bin/ld: libBasicPitchCNN.a(BasicPitchCNN.cpp.o): relocation R_X86_64_32 against `.rodata.str1.1' can not be used when making a shared object; recompile with -fPIC
/usr/lib64/gcc/x86_64-suse-linux/12/../../../../x86_64-suse-linux/bin/ld: failed to set dynamic section sizes: bad value
collect2: error: ld returned 1 exit status

[gnac]

I've created a (pull request](https://github.com/DamRsn/NeuralNote/pull/44) that is dependent on a prebuilt onnxruntime library. It isn't building (linking errors) yet, but I figured its a place to start a discussion.
FWIW, I also have a pending pull request on the libonnx project which adds support for building the linux archive. You will probably need to pull this project to build an archive to test the linux build of this pull request.

[RustoMCSpit]

just tried running this on wine with yabridge on bitwig, it kinda works but the gui has a weird rendering issues which causes huge delay / latency

[dromer]

@RustoMCSpit but we have the source-code, so no need for wine.

Thnx for the work @gnac not sure if I'll be able to get your stuff tested, but looking forward to see it working!

[RustoMCSpit]

@RustoMCSpit but we have the source-code, so no need for wine.

Thnx for the work @gnac not sure if I'll be able to get your stuff tested, but looking forward to see it working!

no need for wine until it actually is supported you mean right?

[dromer]

@RustoMCSpit sure, if it works for you and you can play with the software that's great.

[RustoMCSpit]

@RustoMCSpit sure, if it works for you and you can play with the software that's great.

ehhh, 'works' but barely, very bugged

[RustoMCSpit]

yeah after a few days the app is pretty unbareable to work with through WINE, if the devs want to see it through they could collect crash reports they get and see whats going on

[dromer]

@RustoMCSpit if you want to go that route please open a separate ticket so we don't mix up topics.

I think it would make more sense to build the program for Linux then to try and debug wine issues.

[thelabcat]

Lemme know if this goes anywhere! RN it is only usable in a VM for me ;-(

[RustoMCSpit]

just to update, program still really weird on WINE after new update

[dromer]

@RustoMCSpit please open a separate topic on WINE, this request is about actual Linux builds and NOT any WINE work around.

Lets not dilute either topic by cramming it into the same ticket.

[polygon]

@gnac Did you get any further with this? I've also managed to build my own version of onnxruntime and actually got NeuralNote to compile. But right now it segfaults when starting. I've so far traced this back to this line of code (juce_XmlElement.cpp):

static const String juce_xmltextContentAttributeName ("text");

For some reason, this static variable is not initialized correctly and the pointer inside the string (to the actual text data) is a Null pointer so it crashes when it's being used (in this case, to parse the Linux fontconfig). Might just be a small issue with some JUCE settings required on Linux, I'm sifting through the things, but it's a pretty strange error.

[polygon]

To me, this looks like a case of "static initialization fiasco". In NeuralNote/Lib/Components/UIDefines.h there are pointers to fonts declared like so:

// Fonts
const juce::Typeface::Ptr MONTSERRAT_BOLD =
    juce::Typeface::createSystemTypefaceFor(BinaryData::MontserratBold_ttf, BinaryData::MontserratBold_ttfSize);

const juce::Typeface::Ptr MONTSERRAT_SEMIBOLD =
    juce::Typeface::createSystemTypefaceFor(BinaryData::MontserratSemiBold_ttf, BinaryData::MontserratSemiBold_ttfSize);

const juce::Typeface::Ptr MONTSERRAT_REGULAR =
    juce::Typeface::createSystemTypefaceFor(BinaryData::MontserratRegular_ttf, BinaryData::MontserratRegular_ttfSize);

Under Linux, this triggers (unless JUCE_FONT_PATH is set) a font lookup that involves parsing /etc/fonts/fonts.conf using the XML parser. The XML parser has the static string initialization mentioned above. In my compiled variant, this static string has not been initialized when the createSystemTypefaceFor calls are being made. I guess this works for Windows and Mac because loading fonts there does not involve the XML parser. Still, this could be considered unsound (at least on Linux). I will try to work around this.

Setting JUCE_FONT_PATH seems to solve the problem and makes the program start. Seems to work well so far. I'm not sure why JUCE insists on parsing the fontconfig since this plugin only loads fonts from its local resources. But it's probably a simple patch to fix this (or just set JUCE_FONT_PATH environment to literally anything).

So in summary, it's working in Linux for me, note detection etc. is working as expected. Unfortunately, I cannot seem to get MIDI out of the plugin. I can start the dragging, but neither dropping it into my DAW (Bitwig), nor into the file browser seems to be doing anything.

[gnac]

@polygon you've gotten further than I have. How did you resolve the linker issue?

[polygon]

I don't remember running into this linker issue to be honest. I've uploaded my current state here: https://github.com/polygon/NeuralNote/tree/linux

I re-factored the UIDefines.h code into a singleton provider to work around the initialization order issues. Setting JUCE_FONT_PATH would solve the issue for running the program standalone but not when used as a plugin. With the refactoring, I also got things working in Bitwig.

I'm not currently using build.sh but just run CMake directly like so:

cmake -B build -DCMAKE_BUILD_TYPE="Release"
cmake --build build -j

This expects to have a Linux built libonnxruntime-neuralnote in place as well, similarly unpacked as described in the build.sh. I've also got this done, but it's still quite hacky and I'm in the process of writing scripts for all of this. So you can poke around if you want, but it's still all a bit hacky and I'm working on consolidating everything.

[CodesoundR]

I don't remember running into this linker issue to be honest. I've uploaded my current state here: https://github.com/polygon/NeuralNote/tree/linux

I re-factored the UIDefines.h code into a singleton provider to work around the initialization order issues. Setting JUCE_FONT_PATH would solve the issue for running the program standalone but not when used as a plugin. With the refactoring, I also got things working in Bitwig.

I'm not currently using build.sh but just run CMake directly like so:

cmake -B build -DCMAKE_BUILD_TYPE="Release"
cmake --build build -j

This expects to have a Linux built libonnxruntime-neuralnote in place as well, similarly unpacked as described in the build.sh. I've also got this done, but it's still quite hacky and I'm in the process of writing scripts for all of this. So you can poke around if you want, but it's still all a bit hacky and I'm working on consolidating everything.

Hi, and thanks so much for porting this beautiful project in Linux (I'm under Mint). I ask you if you have any news on the script you proposed. I'm having a hard time figuring out how to install this softare... thanks again!!!

[Ken-Andre]

I don't remember running into this linker issue to be honest. I've uploaded my current state here: https://github.com/polygon/NeuralNote/tree/linux I re-factored the UIDefines.h code into a singleton provider to work around the initialization order issues. Setting JUCE_FONT_PATH would solve the issue for running the program standalone but not when used as a plugin. With the refactoring, I also got things working in Bitwig. I'm not currently using build.sh but just run CMake directly like so:

cmake -B build -DCMAKE_BUILD_TYPE="Release"
cmake --build build -j

This expects to have a Linux built libonnxruntime-neuralnote in place as well, similarly unpacked as described in the build.sh. I've also got this done, but it's still quite hacky and I'm in the process of writing scripts for all of this. So you can poke around if you want, but it's still all a bit hacky and I'm working on consolidating everything.

Hi, and thanks so much for porting this beautiful project in Linux (I'm under Mint). I ask you if you have any news on the script you proposed. I'm having a hard time figuring out how to install this softare... thanks again!!!

Yes it will be really great. Think that they had already started working on it.

[CodesoundR]

Yes it will be really great. Think that they had already started working on it.

Thanks ..... I'm waiting.... ;-)

[polygon]

I drove this a bit further and opened Pull Requests in both project components:

https://github.com/DamRsn/NeuralNote/pull/85 https://github.com/tiborvass/libonnxruntime-neuralnote/pull/2

However, communication quickly died down (or didn't happen), so I assumed that the maintainers were not interested in this. I finished the packaging for the Linux distribution of my choice (NixOS) and carried on. I'll give a short rundown on how to build this yourself, though. You will need to use my forks of libonnxruntime-neuralnote and NeuralNote. For both, you want the "Linux" branch.

I will link my Nix build recipes, since they are also a good indicator on the dependencies that you will need. You can ignore a lot of the CMake download links replacement, that is a Nix/NixOS peculiarity (no internet access during builds).

libonnxruntime-neuralnote (Nix-recipe)

Once you installed everything unter buildInputs, you just run the instructions under buildPhase . This should yield you a libonnxruntime-neuralnote.tar.gz which you will need in the next step.

NeuralNote (Nix-recipe)

Again, make sure you have all the tools and dependencies installed. Now, run all the steps under postPatch in the root-folder of the repo (in line 83, reference the tar.gz file from the build above). This also patches a Drag&Drop bug in JUCE. You then just run CMake like above and this should give you the final binary / plugin.

Hopefully this gets you started. Let me know if you get stuck anywhere.

[thelabcat]

Thanks for doing this.

[CodesoundR]

I drove this a bit further and opened Pull Requests in both project components:

#85 tiborvass/libonnxruntime-neuralnote#2

However, communication quickly died down (or didn't happen), so I assumed that the maintainers were not interested in this. I finished the packaging for the Linux distribution of my choice (NixOS) and carried on. I'll give a short rundown on how to build this yourself, though. You will need to use my forks of libonnxruntime-neuralnote and NeuralNote. For both, you want the "Linux" branch.

I will link my Nix build recipes, since they are also a good indicator on the dependencies that you will need. You can ignore a lot of the CMake download links replacement, that is a Nix/NixOS peculiarity (no internet access during builds).

libonnxruntime-neuralnote (Nix-recipe)

Once you installed everything unter buildInputs, you just run the instructions under buildPhase . This should yield you a libonnxruntime-neuralnote.tar.gz which you will need in the next step.

NeuralNote (Nix-recipe)

Again, make sure you have all the tools and dependencies installed. Now, run all the steps under postPatch in the root-folder of the repo (in line 83, reference the tar.gz file from the build above). This also patches a Drag&Drop bug in JUCE. You then just run CMake like above and this should give you the final binary / plugin.

Hopefully this gets you started. Let me know if you get stuck anywhere.

hello and thank you for your interest. But I don't understand how to install it on my Mint system....

[Ken-Andre]

Ok, thanks for this contribution.

On Wed, 20 Mar 2024 at 17:57 CodesoundR @.***> wrote:

I drove this a bit further and opened Pull Requests in both project components:

#85 https://github.com/DamRsn/NeuralNote/pull/85 tiborvass/libonnxruntime-neuralnote#2 https://github.com/tiborvass/libonnxruntime-neuralnote/pull/2

However, communication quickly died down (or didn't happen), so I assumed that the maintainers were not interested in this. I finished the packaging for the Linux distribution of my choice (NixOS) and carried on. I'll give a short rundown on how to build this yourself, though. You will need to use my forks of libonnxruntime-neuralnote https://github.com/polygon/libonnxruntime-neuralnote and NeuralNote https://github.com/polygon/NeuralNote. For both, you want the "Linux" branch.

I will link my Nix build recipes, since they are also a good indicator on the dependencies that you will need. You can ignore a lot of the CMake download links replacement, that is a Nix/NixOS peculiarity (no internet access during builds). libonnxruntime-neuralnote (Nix-recipe https://github.com/polygon/audio.nix/blob/master/vst/neuralnote/libonnxruntime-neuralnote.nix )

Once you installed everything unter buildInputs https://github.com/polygon/audio.nix/blob/1217ff4ed70c562b5eaf9832151967b3a568c65c/vst/neuralnote/libonnxruntime-neuralnote.nix#L32-L41, you just run the instructions under buildPhase https://github.com/polygon/audio.nix/blob/1217ff4ed70c562b5eaf9832151967b3a568c65c/vst/neuralnote/libonnxruntime-neuralnote.nix#L52-L55. This should yield you a libonnxruntime-neuralnote.tar.gz which you will need in the next step. NeuralNote (Nix-recipe https://github.com/polygon/audio.nix/blob/master/vst/neuralnote/neuralnote.nix )

Again, make sure you have all the tools and dependencies https://github.com/polygon/audio.nix/blob/1217ff4ed70c562b5eaf9832151967b3a568c65c/vst/neuralnote/neuralnote.nix#L42-L58 installed. Now, run all the steps under postPatch https://github.com/polygon/audio.nix/blob/1217ff4ed70c562b5eaf9832151967b3a568c65c/vst/neuralnote/neuralnote.nix#L77-L86 in the root-folder of the repo (in line 83, reference the tar.gz file from the build above). This also patches a Drag&Drop bug in JUCE. You then just run CMake like above https://github.com/DamRsn/NeuralNote/issues/33#issuecomment-1784055326 and this should give you the final binary / plugin.

Hopefully this gets you started. Let me know if you get stuck anywhere.

hello and thank you for your interest. But I don't understand how to install it on my Mint system....

— Reply to this email directly, view it on GitHub https://github.com/DamRsn/NeuralNote/issues/33#issuecomment-2010055747, or unsubscribe https://github.com/notifications/unsubscribe-auth/AZIUF3B3NWRW67RNGPVVABTYZG5WDAVCNFSM6AAAAAAYBK7CWSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDAMJQGA2TKNZUG4 . You are receiving this because you commented.Message ID: @.***>

[polygon]

hello and thank you for your interest. But I don't understand how to install it on my Mint system....

These instructions are to build the project, distribution for specific Linux systems comes later. I'm very sorry, but I don't have the time to build and maintain packages for a Linux distribution I am not using myself. If you find someone who knows how to package for Ubuntu/Mint and is interested, I'm happy to assist getting things started.

[CodesoundR]

hello and thank you for your interest. But I don't understand how to install it on my Mint system....

These instructions are to build the project, distribution for specific Linux systems comes later. I'm very sorry, but I don't have the time to build and maintain packages for a Linux distribution I am not using myself. If you find someone who knows how to package for Ubuntu/Mint and is interested, I'm happy to assist getting things started.

Hi, I wanted to thank you anyway for what you did. I try to see in the various formums if there is someone who can help us for this porting.... bye

EDIT: BUT TO DATE I HAVE NOT BEEN ABLE TO FIND ANYONE :-(

[karimdjemai]

Hi, thank you so much for your work @polygon, I got it to build on archlinux but when I run it, I get the following error:

./build/NeuralNote_artefacts/Release/Standalone/NeuralNote terminate called after throwing an instance of 'Ort::Exception' what(): inference_session.cc:1020 LoadOrtModelWithLoader The ORT format model version [6] is not supported in this build 1.14.1. This build doesn't support ORT format models older than version 5. See: https://github.com/microsoft/onnxruntime/blob/rel-1.14.0/docs/ORT_Format_Update_in_1.13.md [1] 71338 IOT instruction (core dumped) build/NeuralNote_artefacts/Release/Standalone/NeuralNote

which is particularily odd, since the newst ORT model version is 5 according to https://onnxruntime.ai/docs/performance/model-optimizations/ort-format-models.html

I did run the convert-onnx-to-ort stuff before building the onnxruntime and cp-ing the tar, but my guess is now that something went wrong in that step. However I did not find a solution and have no idea on where to look further. Does anybody have an idea on how to debug this? Thanks in advance!

[DamRsn]

@karimdjemai I encountered a similar error message on Windows recently when working with onnxruntime. The issue was that my executable was linking with the onnxruntime dynamic library available in my system dynamic library directory. Maybe the same is happening for you on Linux.

So check that you successfully get a static lib for onnxruntime and that you link with it and not with some other version available on your system.

[karimdjemai]

Hi @DamRsn, thanks for your reply!

I checked the dynamic library links of the executable, and onnxruntime was not there. The onnxruntimelib.a file sits in the right location as well.

But you might still be onto something, as I could not find a log proving the linking of the static lib in the build output. This might be normal, though. I did not find any other way to verify that the static lib file was linked correctly, do you have any idea?

I am very grateful for further ideas on where to look to get it running in general, thanks in advance!

[karimdjemai]

I finally got it working, so if anyone encounters my problem, here is how I solved it: In general I followed these steps outlined by @polygon

I drove this a bit further and opened Pull Requests in both project components:

#85 tiborvass/libonnxruntime-neuralnote#2

However, communication quickly died down (or didn't happen), so I assumed that the maintainers were not interested in this. I finished the packaging for the Linux distribution of my choice (NixOS) and carried on. I'll give a short rundown on how to build this yourself, though. You will need to use my forks of libonnxruntime-neuralnote and NeuralNote. For both, you want the "Linux" branch.

I will link my Nix build recipes, since they are also a good indicator on the dependencies that you will need. You can ignore a lot of the CMake download links replacement, that is a Nix/NixOS peculiarity (no internet access during builds).

libonnxruntime-neuralnote (Nix-recipe)

Once you installed everything unter buildInputs, you just run the instructions under buildPhase . This should yield you a libonnxruntime-neuralnote.tar.gz which you will need in the next step.

NeuralNote (Nix-recipe)

Again, make sure you have all the tools and dependencies installed. Now, run all the steps under postPatch in the root-folder of the repo (in line 83, reference the tar.gz file from the build above). This also patches a Drag&Drop bug in JUCE. You then just run CMake like above and this should give you the final binary / plugin.

Hopefully this gets you started. Let me know if you get stuck anywhere.

BUT when building libonnxruntime-neuralnote, you need to install the requirements.txt into the venv. Here it is important to use the correct python package version of onnxruntime (its 1.14.1) and not the newest. For that I had to downgrade python to below 3.11, so that the onnxruntime package is available.

@DamRsn Thanks for your help, without your comment I wouldn't have suspected that my build of libonnxruntime-neuralnote was faulty, also thanks for building this incredible plugin in general. It is super cool!