GNU awk and POSIX awk: Associative Array Iteration Order

Direct Answer

POSIX awk makes no guarantee about the order in which associative‑array elements are visited, and GNU awk documents the same; the sequence is implementation‑defined and may differ between runs, builds, or environments unless you explicitly sort.

Confirmed Facts

• POSIX.1‑2017 §4.1.3 states that the traversal order of an array is unspecified.
• gawk manual (Array Traversal) says the order of a for (i in array) loop is not defined and may vary between implementations and even between runs.
• Other POSIX‑compliant awks (mawk, nawk) use hash tables that yield effectively random order.

What Happens Without Explicit Sorting

When you write:

for (k in counts) {
    print k, counts[k]
}

The output order is whatever the underlying hash table produces. It may appear stable with the same gawk version and input order, but it can change after upgrading gawk, switching to mawk, or running with a different locale that affects hash seeding. 32‑bit and 64‑bit builds also differ because bucket counts change.

Portable Solutions

1. Use gawk’s PROCINFO sorted_in:

   PROCINFO["sorted_in"] = "@ind_str_asc"
   for (k in counts) print k, counts[k]

2. Pipe to external sort (works everywhere):

   for (k in counts) print k "\t" counts[k] | "sort"

3. Collect keys with asorti (gawk only) and iterate:

   n = asorti(counts, sorted)
   for (i = 1; i <= n; i++) {
       k = sorted[i]
       print k, counts[k]
   }

Quick Verification

Run the following script with gawk and with mawk to see the difference:

# test_order.awk
BEGIN {
    a["z"]=1; a["a"]=2; a["m"]=3
}
END {
    for (k in a) print k
}

Compare the outputs; they will likely differ, confirming the unspecified order.

One Diagnostic Detail

If you need the exact order a particular awk version produces on your system, execute the verification script above. Otherwise, assume the order is unspecified and apply explicit sorting.