Reading nested Map data structures in WDL

I’m having trouble getting WDL’s read_json to work as expected when trying to create a Map[String, Map[String, String]] data structure. If I explicitly declare the data structure it works, as shown below (the gm variable):

workflow TestWorkflow {

    String genome_choice
    String output_name = "xyz.txt"
    Map[String, Map[String,String]] gm = {
        "A": {"gtf":"A.gtf", "idx":"A.idx"}, 
        "B": {"gtf":"B.gtf", "idx":"B.idx"} 
    }

    call something {
        input:
            genome_file_map = gm[genome_choice],
            outname = output_name
    }

    output {
        File fout = something.fout
    }
}

task something {

    
    String outname
    Map[String, String] genome_file_map

    String idx = genome_file_map["idx"]
    String gtf = genome_file_map["gtf"]

    command {
        echo ${idx} >> ${outname}
        echo ${gtf} >> ${outname}
    }
    output {
        File fout = "${outname}"
    }
}

However, instead of explicitly writing gm, I would like to read that from a file. If I then create a JSON-format file:

{
  "A": {"gtf":"A.gtf", "idx":"A.idx"},
  "B": {"gtf":"B.gtf", "idx":"B.idx"}
}

and submit that as an input file

workflow TestWorkflow {
    ...
    File resource_map_jsonfile
    ...
    Map[String, Map[String,String]] gm = read_json(resource_map_jsonfile)

it fails to parse, giving me:

ERROR: gm is declared as a Map[String, Map[String, String]] but the expression evaluates to a Object:

    Map[String, Map[String,String]] gm = read_json(resource_map_jsonfile)

If I change the type declaration to Object, then all the bracket-style indexing fails. So that’s out.

Curiously, if I instead declare gm as a Map[String,String] it then fails giving:

java.lang.IllegalArgumentException: No coercion defined from wom value(s) '{"gtf": "B.gtf", "idx": "B.idx"}' of type 'Map[String, String]' to 'String'.

so clearly it is parsing the JSON file correctly, as it identifies that nested Map[String, String]. I can probably find ways around this (using other read_* functions), but that’s less than ideal.