Stricter rules for initialising params with no default value

[drpatelh]

It would be good to standardise how we are defining default / empty parameter values in the pipeline nextflow.config. Seeing as we are now validating these parameters against the JSON schema it will be nice to use a uniform approach.

As shouted on Slack:

• Default values should be specified as usual using native Groovy variable types
• BOOLEAN PARAMETERS SHOULD BE SET TO false in nextflow.config
• FILE PATHS / STRINGS / INTEGERS / FLOATS / LISTS / MAPS SHOULD BE SET TO null in nextflow.config

[ewels]

• nf-core schema lint should check for this and fail if false found for a non-boolean
• Can look into removing workarounds for empty strings / false values with no default..

[drpatelh]

nf-core schema lint should check for this and fail if false found for a non-boolean

This is a little tricky because NF doesn't accept empty strings on the command-line e.g. --foo '' and will ignore the quotes and instead set params.foo = true in the workflow. I came across this in the rnaseq pipeline the other day and the only workaround to evaluate if that parameter has been set is to use --foo false.

[ewels]

Ah and doing --foo false would then fail the parameter validation you mean? Yeah bit of an edge case but I can see that this could cause problems. What does --foo null do? Or --foo "''"?

[ggabernet]

I do have a case where I want to set an integer parameter to 0 by default (not null). It specifies the start position for e.g. a UMI barcode or a sequencing primer. So I actually need to perform some arithmetics on it later in the pipeline.

So I think having an integer default to 0 shouldn't be interpreted as false. How should then this default be specified?

[ewels]

Yes that should be totally fine - set it to 0. If it is set to an int and has the default set to 0 in the schema then it will be fine - basically the same setup as if it was any other number.

This is about being more strict when you don't want to have a default value, which in turn allows to be stricter about defaults that evaluate to false like this.

[ewels]

In practice this will basically mean deleting / making these workarounds more strict:

tools/nf_core/schema.py

Lines 504 to 506 in d0a6649

	# Assume that false and empty strings shouldn't be a default
	if p_val == "false" or p_val == "":
	del p_schema["default"]

tools/nf_core/pipeline-template/lib/NfcoreSchema.groovy

Lines 509 to 519 in d0a6649

	if (schema_value == null) {
	if (param_type == 'boolean') {
	schema_value = false
	}
	if (param_type == 'string') {
	schema_value = ''
	}
	if (param_type == 'integer') {
	schema_value = 0
	}
	} else {

tools/nf_core/pipeline-template/lib/NfcoreSchema.groovy

Lines 236 to 239 in d0a6649

	// remove anything evaluating to false
	if (!p['value']) {
	new_params.remove(p.key)
	}

tools/nf_core/pipeline-template/lib/NfcoreSchema.groovy

Line 440 in d0a6649

def defaultValue = group_params.get(param).default ? " [default: " + group_params.get(param).default.toString() + "]" : ''

..and probably more (eg. in website code too etc)