[saket0187]

Fix r.univar reporting -DBL_MAX instead of nan (JSON null) as the maximum of all-null maps and zones, wrong since 8.4.0 because #3512 changed the start value but not the merge check; adds a pytest test that fails without the fix.

Claude was used as a coding assistant.

[echoix]

There’s a kind of outage when installing Ubuntu packages it seems. Let’s wait before trying to trigger new runs (I reran a couple and still failed)

[wenzeslaus]

You could add the min test as suggested in my comment or given that both min and max can be tested and zd->bucket.n is thread-local, the lock can be acquired only later (hopefully slightly faster than the current code on main). The update test then continues to use the values in question.

// (Untested code)
if (zd->bucket.n != 0) {
#if defined(_OPENMP)
    omp_set_lock(&minmax[z]);
#endif
    // The max and min values are tested for NaN (their defaults).
    if (stats[z].max < zd->max || stats[z].max != stats[z].max)
        stats[z].max = zd->max;
    if (stats[z].min > zd->min || stats[z].min != stats[z].min)
        stats[z].min = zd->min;
#if defined(_OPENMP)
    omp_unset_lock(&minmax[z]);
#endif
}

[wenzeslaus]

Thank you! It is very important to keep improving r.univar, not only important to users, but we use that a lot in tests as well.