Is it possible to force the generation of an OpenApiReference client?

[paulomorgado]

_GenerateOpenApiCode is skipped if the output is not outdated.

Is it possible to force it to run?

Any hints, @rainersigwald?

[danmoseley]

I'm not aware of any way to tell MSBuild to not skip a target that has no inputs newer than outputs. That is typically done in the authoring by eg deleting the output or by conditionally adding a bogus missing output.

[rainersigwald]

Yeah I have occasionally thought of having an engine setting to bypass inputs/outputs and run everything but it doesn't exist today.

I don't understand that target though--if it's basically only calling the MSBuild task, why not drop the target-level incrementality?

[paulomorgado]

I added this to my build:

  <Target Name="__My__GenerateOpenApiCode" BeforeTargets="_GenerateOpenApiCode"
    Condition="$(OpenApiGenerateCodeAtDesignTime) OR ('$(DesignTimeBuild)' != 'true' AND '$(BuildingProject)' == 'true')">
    <Error Condition="!Exists('%(OpenApiReference.FullPath)')" Text="Input OpenAPI file %(Identity) does not exist." />

    <MSBuild Projects="$(MSBuildProjectFullPath)"
        BuildInParallel="$(BuildInParallel)"
        Properties="GeneratorTargetPath=%(OpenApiReference.OutputPath);GeneratorTarget=Generate%(CodeGenerator);GeneratorMetadata=%(SerializedMetadata)"
        Targets="_InnerGenerateOpenApiCode" />
  </Target>

and it does the trick. __My__GenerateOpenApiCode is always executed (on those conditions) and _GenerateOpenApiCode ends up never being executed, because the conditions are the same and, if __My__GenerateOpenApiCode ran, outputs are up to date.

[peteraritchie]

I ran into this today. Rebuild didn't help. I had to do a Clean, then a Build before it would work. A Clean then a Rebuild, did not. :/

[paulomorgado]

That's odd. A Rebuild should be a Clean followed by a Build.

You could collect a binlog to see what's missing.