is this true even if you set save_bias_corrected = True?

I'll try that now

You are right, no error when save_bias_corrected = True. However why call the bias-corrected image the "modulated version of input image" in the code? I didn't think the so called "modulated" output was actually the bias-corrected image, that's why i proposed this commit.

Then how to deal with this optional output?

it's called modulated only from spm's terminology - we could call it bias_corrected_image. in terms of dealing with it:

regarding the outputs, i would recommend removing exists=True from all the outputspec metadata and then determine which files are generated based on the inputs *_output_type and the save_bias_corrected flag.

if isdefined(self.inputs.save_bias_corrected) and self.inputs.save_bias_corrected:
    outputs['bias_corrected_image'] = fname_presuffix(f, prefix='m')

and so on.

Great! Should i do this modification and add it here? And change "modulated version of input image" to "bias-corrected image"?

if you have the time, that would be great!

regarding the bias corrected image, let's add another field and deprecate the old field in a couple of releases. for now we will have two outputs with the same content.

check the example under deprecated and new_name at the end of:

http://nipy.org/nipype/devel/interface_specs.html#common

The output list was already filtered according to options for the WM, GM and CSF probability maps so i only added the condition you suggested in the same part of code (cf 583 to 587), i hope that's all right. I tested it and it works fine whatever boolean the save_bias_corrected variable variable is set to.

I only saw your second message now so i'll have to add another commit for the old and new field.

In which release should it be deprecated? Did you mean in a couple of (sub)releases, say 0.9.3?

i would say 0.10

Thanks, i did this change in SegmentOutputSpec:
modulated_input_image = File(exists=True, deprecated='1.0', new_name='bias_corrected_image', desc='bias-corrected version of input image')
bias_corrected_image = File(exists=True, desc='bias-corrected version of input image')

I'm not familiar with this deprecation ability, so what to do for the new line of code in Segment? Should i just keep it with the old variable like below:
if isdefined(self.inputs.save_bias_corrected) and self.inputs.save_bias_corrected:
outputs['modulated_input_image'] = fname_presuffix(f, prefix='m')

Note that as indicated in my post above i placed this line in the Segment class as code was already implemented there to modify the effective output.

And many thanks for your help. I know you spend as much time or even more explaining me things in this discussion that you would modifying directly the code. But hopefully all what you explain me now will be useful for all the future contributions i may have (and i plan to have some!)

yup. the deprecation will kick in if you try to run an interface with an input that was deprecated using a nipype release greater than equal to the version indicated in the metadata. you can test it by saying deprecated= '0.8' - also deprecated='0.10'

finally, i think for this interface we need to remove exists=True for any of the outputs that are predicated on input settings.

(ps. no worries regarding time - i would rather teach somebody how to do something - it spreads the knowledge quicker).

Thanks for that.

I did several tests and, contrarily to what mentioned in my previous post, it seems that i need to use the new variable in the new line of code and not the old variable. So in the third post above the new line of code should actually be:

if isdefined(self.inputs.save_bias_corrected) and self.inputs.save_bias_corrected:
outputs['bias_corrected_image'] = fname_presuffix(f, prefix='m')

Do you feel this PR is fine? Also i'd be happy to help working on the augmented fslmaths Interface you proposed

testing this now - should be good to go.