PublishOp fails with empty value channel from collect() on empty process output

[pinin4fjords]

Bug report

Related to #6458 but a different trigger. That issue covers errorStrategy 'ignore' creating empty value channels. This issue is about collect() on empty process output leading to the same error.

Expected behavior

When a process never runs (empty input), and its output is passed through collect(), the workflow should complete without error if the output block has enabled true.

Actual behavior

ERROR ~ Cannot access first() element from an empty List

Stack trace points to PublishOp.onComplete(PublishOp.groovy:199).

Minimal reproduction

process UPSTREAM {
    input: tuple val(meta), path(reads)
    output: tuple val(meta), path("out.txt"), emit: results
    script: "echo done > out.txt"
}

process DOWNSTREAM {
    input: tuple val(meta), path(data)
    output: tuple val(meta), path("final.txt"), emit: results
    script: "echo final > final.txt"
}

workflow {
    // Empty input - UPSTREAM never runs
    UPSTREAM(Channel.empty())

    // collect() on empty process output never emits
    ch_collected = UPSTREAM.out.results
        .collect { it[1] }
        .map { files -> [['id': 'all'], files] }

    // DOWNSTREAM never starts (no input from collect)
    DOWNSTREAM(ch_collected)

    publish:
    results = DOWNSTREAM.out.results
}

output {
    results {
        enabled true
        path "results"
    }
}

Run with: nextflow run repro.nf -output-dir results

Real-world example: nf-core/rnaseq

This bug was discovered while implementing workflow outputs for nf-core/rnaseq. The pattern appears in:

1. QUANTIFY_PSEUDO_ALIGNMENT subworkflow uses collect() to aggregate results before passing to TXIMETA_TXIMPORT:

   TXIMETA_TXIMPORT (
       ch_pseudo_results.collect{ meta_results -> meta_results[1] }.map { results -> [ ['id': 'all_samples'], results ] },
       ...
   )

2. Output block (on workflow-outputs branch) publishes TXIMPORT outputs:

   pseudo_counts_gene {
       enabled !params.skip_pseudo_alignment && params.pseudo_aligner
       path { params.pseudo_aligner }
   }

3. Trigger condition: In BAM input mode (--skip_alignment true), SALMON_QUANT never runs because there are no FASTQs to process. The collect() on its empty output never emits, TXIMETA_TXIMPORT never starts, and publishing its output triggers the bug.

Observations

• collect() on an empty channel does not emit (verified with .view() and .ifEmpty())
• toList() on an empty channel emits []
• Direct pass-through of an empty process output to the output block works fine
• The error occurs in PublishOp.onComplete when it encounters a channel from a process that never started

Environment

• Nextflow version: 25.10.2
• Java version: OpenJDK 21
• OS: macOS

[bentsherman]

Thanks Jon. It's the same issue of something creating an empty value channel, which is invalid. First the ignore strategy, now we see also it's the collect operator, which returns nothing when the source channel is empty, when really it should return an empty list like toList

For now the remedy is to do collect(...).ifEmpty([]). It should ultimately be addressed by v2 operators, which is currently in the works

[pinin4fjords]

Thanks Jon. It's the same issue of something creating an empty value channel, which is invalid. First the ignore strategy, now we see also it's the collect operator, which returns nothing when the source channel is empty, when really it should return an empty list like toList

For now the remedy is to do collect(...).ifEmpty([]). It should ultimately be addressed by v2 operators, which is currently in the works

I still had one lingering error like this even when I thought I'd nuked all the collect()s, so I went nuclear and added the ifEmtpy() on very publish item: https://github.com/nf-core/rnaseq/blob/e844249c9ace178d26541879db5f51246255a830/main.nf#L324

But I might put that PR on hold anyway after our other dicussions.