Is there a way to force the eval to interpret it as a

string

instead of a

float

or an

integer

? Asking because I sometimes get versions like this:

"biopython": 1.85, <- Interpreted as a float. 
"pandas": "2.2.3",
"python": "3.9.21"

Made with:

tuple val("${task.process}"), val('python'), eval("python --version | sed 's/Python //g'"), topic: versions
tuple val("${task.process}"), val('pandas'), eval("pip show pandas | sed '2!d;s/Version: //g'"), topic: versions
tuple val("${task.process}"), val('biopython'), eval("pip show biopython | sed '2!d;s/Version: //g'"), topic: versions

Hmm that also happens with parameters so it's probably built-in to Nextflow. You could updated your code handling the versions to convert all versions to a string:

def string_versions = versions_map.collectEntries { tool, version ->
    [ tool, version.toString() ]
}

Where

versions_map

is the map containing the tools as key and versions as value

I tried this:

def topic_versions_string = topic_versions.versions_tuple
        .map { process, tool, version ->
            [ process[process.lastIndexOf(':')+1..-1], "  ${tool}: ${version.toString()}" ]
        }                                                                   ^^^^^^^^^^^
        .groupTuple(by:0)
        .map { process, tool_versions ->
            tool_versions.unique().sort()
            "${process}:\n${tool_versions.join('\n')}"
        }

But to no avail, don’t think you’re suggestion is any different right?

What's the error you get?

Ooh no errors, just no difference in the snapshot

Does it really matter that much in this case? It shouldn't, right?

(higlighted with

^

just to show what’s different)

There probably is 😛

Don't worry on that, it's just an inconsistency that was bugging me 🤣

I get that 🤣 It should be fixed once we have proper typing in processes

wait, you had an

eval()

that was returning a number instead of a string? that sounds weird 😆 if you can reproduce it in a small example, I would submit a nextflow issue for it.

eval()

should always return a string

Update on this, it’s not

eval()

that is causing this issue. The module snapshots all still have quotes, but it’s only in the main pipeline-wide tests where this occurs. According to seqera AI the root lies within the nf-core template of the function

processVersionsFromYAML

and the bugchain is:

1. eval("pip show biopython | sed ...") runs in the shell → emits the plain string "1.85" ✅
2. Nextflow receives that into a channel tuple — here Groovy infers the type from the value content, and "1.85" is coerced to Double(1.85) 🐛
3. In topic_versions_string, the line "  ${tool}: ${version}" does GString interpolation of Double(1.85) → produces the YAML line biopython: 1.85 (no quotes)
4. processVersionsFromYAML() in utils_nfcore_pipeline then calls new org.yaml.snakeyaml.Yaml().load(...) which reads 1.85 as Double again, and yaml.dumpAsMap() with MINIMIZE_QUOTES re-serialises it as the bare scalar 1.85
5. nf-test snapshots store this as a JSON number 1.85

And these are the suggested solutions:

/ ===========================================================================
// VERSION 1 – current nf-core implementation (processVersionsFromYAML)
//   Copied verbatim from:
//   subworkflows/nf-core/utils_nfcore_pipeline/main.nf
//
//   BUG: dumpAsMap() uses MINIMIZE_QUOTES – it only adds quotes when the value
//   would be re-parsed as a different type.  SnakeYAML's Resolver decides:
//     - "1"    → !!int   → needs quotes → '1'
//     - "1.85" → !!float → needs quotes → '1.85'   (but unquoted input stays Double!)
//     - "1.2.3"→ !!str   → no quotes   → 1.2.3     (safe, but inconsistent)
//   The result: Integer/Double values that came in unquoted are dumped UNQUOTED,
//   so nf-test's JSON serialiser stores them as numbers → snapshot type mismatch.
// ===========================================================================
def processVersionsFromYAML_ORIGINAL(String yaml_file) {
    def yaml     = new org.yaml.snakeyaml.Yaml()
    def versions = yaml.load(yaml_file).collectEntries { k, v -> [k.tokenize(':')[-1], v] }
    return yaml.dumpAsMap(versions).trim()
}

// ===========================================================================
// VERSION 2 – fixed implementation
//   Two changes:
//   1. Stringify every version value after load() so Integer/Double → String.
//   2. Use a custom Representer that always DOUBLE_QUOTES String scalars,
//      so the dump is safe regardless of what the value looks like.
//      This means "1", "1.85", "1.2.3", "1.2.3.4" are all output identically.
// ===========================================================================
def processVersionsFromYAML_FIXED(String yaml_file) {
    // --- loader (unchanged) ---
    def yamlIn = new org.yaml.snakeyaml.Yaml()

    // --- dumper: always quote strings ---
    def opts = new DumperOptions()
    opts.defaultFlowStyle = DumperOptions.FlowStyle.BLOCK

    def rep = new Representer(opts)
    rep.representers[String] = { data ->
        rep.representScalar(Tag.STR, (String) data, DumperOptions.ScalarStyle.DOUBLE_QUOTED)
    } as Represent

    def yamlOut = new Yaml(rep, opts)

    // --- load → stringify all values → dump ---
    def versions = yamlIn.load(yaml_file)
        .collectEntries { k, v -> [k.tokenize(':')[-1], v] }   // strip process prefix
        .collectEntries { proc, tools ->
            // Stringify: Integer(1) → "1", Double(1.85) → "1.85", String → unchanged
            [proc, tools instanceof Map
                ? tools.collectEntries { tool, ver -> [tool, ver.toString()] }
                : tools.toString()]
        }

    return yamlOut.dumpAsMap(versions).trim()
}

so the problem is yaml 🙃