Sort::HashKeys - Get a sorted-by-key list from a hash


    use Sort::HashKeys;

    my %hash;

    @sorted_1 = map { ($_, $hash{$_}) } sort keys %hash;
    @sorted_2 = Sort::HashKeys::sort(%hash);
    # Same outcome, but the second is faster


    [13:37:51]  <a3f>    Is there a better way to get a sorted list out of a hash
                         than map { ($_, $versions{$_}) } reverse sort keys @versions
                         or iterating manually?
    [13:39:06]  <a3f>    oh I could provide a compare function to sort and chunk the
                         list two by two..
    [13:40:15]  <haarg>  i'd probably go with the map{} reverse sort keys
    [13:41:04]  <a3f>    I don't like it that it repeats the lookup for all keys.
                         Of course wouldn't matter in practice but still…
    [13:43:40]  <haarg>  whatever other solution you find will be slower
    [13:49:05]  <a3f>    put it into a list, pass it to XS, run qsort(3) over it
                         with double the element size. and compare taking only the
                         first part into account. return it back?


See in this distribution. Test was run on a Haswell 2.6 GHz i5 CPU (4278U) for a minute each on a copy of a randomly generated hash of 1000 keys. Keys were alphanumeric with length between 1 and 6 and values of integers between 1 and 1000.

    # perl-5.24.0 w/ BSD libc
    map {($_,$h{$_})} sort keys %h           1589/s       --       0%    -28%    -28%    -32%    -33%
    map +($_,$h{$_}), sort keys %h           1592/s        0%     --     -28%    -28%    -32%    -33%
    %h{sort keys %h}                         2204/s       39%     38%     --       0%     -6%     -7%
    S::HK::sort(%h) w/ qsort (threaded)      2199/s       38%     38%      0%     --      -6%     -7%
    S::HK::sort(%h) w/ qsort_r (threaded)    2338/s       47%     47%      6%      6%     --      -2%
    S::HK::sort(%h) w/ qsort (non-threaded)  2375/s       49%     49%      8%      8%      2%     --

49% faster on non-threaded Perl and up to 47% faster on threaded Perl. Aristotle Pagaltzis suggested adding hash slices available since Perl v5.20.0 to the benchmark. This doesn't solve the unncessary hash lookup, but skipping the map appears to have a much bigger impact on performance. The XS solution is only 8% faster than the pure Perl sort using hash slices.

The Slowdown with qsort on threaded build is because the interpreter state is fetched from thread local storage on every comparison. On GNU and FreeBSD libcs, this is avoided by using non-standard qsort_r, which allows passing an extra parameter to the comparison function.



Sorts a hash-like list (key1 => val1, key2 => val2>) by means of your libc's qsort in ascending order according to Perl's internal sv_cmp function.

Unlike Perl's built-in sort. qsort(3) is not required to be a stable sort and providing a custom comparison function is not yet supported.

Lists with odd number of elements are padded by an undef.


Sorts in descending order.



Ahmad Fatoum <>,


Copyright (C) 2017 Ahmad Fatoum

This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself.