Thought I should mention I couldn't get this to build shaders on my minspec GPUs (an Intel HD4000 and a Radeon M9 370X, the Macbook 2015 one) even though they ostensibly support OpenGL>4.0, but in reality have poor driver support. I wasn't able to get any relevant error messages.
I also struggled to downgrade my shaders to SM3, repeatedly running into various unimplemented features in SPIRV - for example, the % operator wouldn't build because it insisted on translating them to the unimplemented Frem function instead of just falling back to Fmod. I got the impression that SPIRV is a bit lacking in supporting legacy GL/SM, but there's been a lot of work done since then so it might be time to try again.

Apologies for the unconstructive reply but I imagine most folks will be focusing on testing the newer shader models and GL4 features which SPIRV appears to handle well, but I think many monogame devs want to support older hardware too. I wasn't able to do so without reverting back to mojoshader for those devices, which is quite reliable when it comes to building legacy shaders. I'll give this another try with the latest ShaderConductor and SPIRV builds soon and hopefully come up with some solutions.

Thought I should mention I couldn't get this to build shaders on my minspec GPUs (an Intel HD4000 and a Radeon M9 370X, the Macbook 2015 one)

It fails at runtime, not during content building, right? I assume you get an exception in Shader.GetShaderHandle() in Shader.OpenGL.cs. You really get nothing useful from GL.GetShaderInfoLog()?

I got the impression that SPIRV is a bit lacking in supporting legacy GL/SM

You got the right impression. With ShaderConductor you really want to use shader model 4 or higher. Shader model 2 and 3 will map to OpenGL 2, which is just not that well supported.
I managed to get the default shaders working for OpenGL 2, so I guess it's not completely useless, but you definitely don't want to use it for extensive shader development.

I think many monogame devs want to support older hardware too.

All you really need is OpenGL 3.3 (released in 2010), you really think there's a lot of demand for even older hardware?

A little update, as I finally have some time for MonoGame stuff again. I addressed a few issues mentioned above:

• Updated to newest MonoGame version.
• Improved error and warning output.
• Merged @eylvisaker fix for matrix array parameters (sorry, I totally missed this PR when you first made it).
• Updated MGCB editor template effects to work with ShaderConductor.
• Looked into the float4x4 issue with vertex shader inputs:

The root of the float4x4 problem is this:
The C# vertex structure contains multiple separate vectors to hold the entire matrix. If you do the same in the vertex shader (use multiple vectors instead of a single matrix) everything is fine, because now there is match between vertex elements and shader inputs.
However, if you use a single matrix parameter for the vertex shader, it won't work, because now vertex elements and shader inputs don't match. This worked with MojoShader, because it would split the matrix parameter into vectors automatically. ShaderConductor doesn't do that, it will create a single matrix parameter:

layout(location = 3) in mat4 in_var_BLENDWEIGHT0;

The easiest solution would be if ShaderConductor could flatten the matrix, just like MojoShader does. I made a request on the SPIRV-Cross GitHub.

I just noticed that ShaderConductor actually has a flag to switch between row-major and column-major matrices, this allowed me to remove the manual matrix transposing for OpenGL. I could slap myself for not seeing this earlier.

Matrices should work now as vertex shader inputs. I had to store the matrix size in the effects file for this, so effects need to be recompiled.

Unfortunately I found a problem with the row-major/column-major switch I made recently (7fc7d9b)
It only works with square matrices currently (KhronosGroup/SPIRV-Cross#1702). I may have to revert this.
SkinnedEffect.fx from the default shaders fails to compile because of this, it uses a 4x3 matrix.

Apparently the square matrix limitation only applies to OpenGL 2 (shader models 2 and 3).
Support for those shader models isn't that great anyway right now, so I'm not planning to look for a solution to this. I added the limitation to the migration document, and changed the SkinnedEffect shader, so it doesn't use non-square matrices anymore.

Thank you again for all this work! It's a big and important chunk which will also have effects on the console implementations.

3.8.1 should be releasing soon, so one of the next goals will be reviewing this PR.

i can not wait anymore :( it looks so amazing!

Closing this PR, as it's contained in #7533