cwltool truncates `float` and `double` inputs to 0 (zero) when they're smaller than 0.000001 (1e-06)

[sebitz]

test_tool.cwl:

class: CommandLineTool
inputs:
  float:
    type: float
    default: 1e-6
    inputBinding:
      position: 1
  double:
    type: double
    default: 9.999e-7
    inputBinding:
      position: 2
outputs:
  numbers: stdout
baseCommand: [ printf, '%s\n' ]
stdout: my_numbers.txt

Running cwltool --disable-color --debug test_tool.cwl >test_tool.out.json 2>test_tool.err.txt produces my_numbers.txt. However, it contains:

0.000001
0

Whereas I expected it to look like this:

1e-06
9.999e-07

The results are the same regardless of whether numbers are in scientific or decimal notation, I use type: float or type: double in the CommandLineTool, or whether I pass inputs in a json or yaml file. Curiously enough, it seems to behave correctly if I pass the numbers on the command line, e.g., cwltool test_tool.cwl --float 1e-6 --double 9.999e-7 or ...--float 0.000006 --double 0.0000009999 gives

1e-06
9.999e-07

From test_tool.err.txt I can see that the command line bindings are correct (i.e., datum: 9.999e-07 and datum: 1e-06). However, the command line that was executed is printf %s\n 0.000001 0 instead of printf %s\n 0.000001 0.0000009999.

This happens both with cwltool version 3.1.20241217163858 (see test_tool.err.txt) and 3.1.20250110105449 (latest at the time of writing).

[sebitz]

I think I have a workaround. Unfortunately it needs InlineJavascriptRequirement.

The workaround is:

requirements:
- class: InlineJavascriptRequirement
inputs:
  number:
    type: [ float, double ]
    inputBinding:
      valueFrom: $(self * 1.0)

Probably $(self + 0.0) would work just as well, but I haven't tried it.