Re: Alternatives To Geo::Distance

[gray]

Speaking as the author of Geo-Distance-XS, I consider it to be actively developed. Even though Geo-Distance might be considered decommissioned by it's author, he probably will accept patches via its github repo. As will I for Geo::Distance::XS at its its repo.

A good reason to use Geo::Distance::XS:

---- [ Formula: hsin ] ------------------------------------
perl      - distance from BWI to LAX: 2324.06319758555 miles
xs        - distance from BWI to LAX: 2324.06319758555 miles
gis::fast - distance from BWI to LAX: 2324.06319758555 miles

Benchmark: running gis::fast, perl, xs for at least 1 CPU seconds...
 gis::fast:  1 wallclock secs ( 1.05 usr +  0.01 sys =  1.06 CPU) @ 23
+866.04/s (n=25298)
      perl:  1 wallclock secs ( 1.05 usr +  0.00 sys =  1.05 CPU) @ 78
+019.05/s (n=81920)
        xs:  2 wallclock secs ( 1.12 usr + -0.01 sys =  1.11 CPU) @ 12
+45236.94/s (n=1382213)
               Rate gis::fast      perl        xs
gis::fast   23866/s        --      -69%      -98%
perl        78019/s      227%        --      -94%
xs        1245237/s     5118%     1496%        --

---- [ Formula: polar ] ------------------------------------
perl      - distance from BWI to LAX: 2652.59252292455 miles
xs        - distance from BWI to LAX: 2652.59252292455 miles
gis::fast - distance from BWI to LAX: 2652.59252292455 miles

Benchmark: running gis::fast, perl, xs for at least 1 CPU seconds...
 gis::fast:  1 wallclock secs ( 1.08 usr +  0.00 sys =  1.08 CPU) @ 18
+962.96/s (n=20480)
      perl:  1 wallclock secs ( 1.07 usr +  0.00 sys =  1.07 CPU) @ 80
+736.45/s (n=86388)
        xs:  2 wallclock secs ( 1.14 usr +  0.00 sys =  1.14 CPU) @ 15
+09051.75/s (n=1720319)
               Rate gis::fast      perl        xs
gis::fast   18963/s        --      -77%      -99%
perl        80736/s      326%        --      -95%
xs        1509052/s     7858%     1769%        --

---- [ Formula: cos ] ------------------------------------
perl      - distance from BWI to LAX: 2324.06319758555 miles
xs        - distance from BWI to LAX: 2324.06319758555 miles
gis::fast - distance from BWI to LAX: 2324.06319758555 miles

Benchmark: running gis::fast, perl, xs for at least 1 CPU seconds...
 gis::fast:  1 wallclock secs ( 1.07 usr +  0.00 sys =  1.07 CPU) @ 23
+642.99/s (n=25298)
      perl:  1 wallclock secs ( 1.06 usr +  0.00 sys =  1.06 CPU) @ 77
+282.08/s (n=81919)
        xs:  0 wallclock secs ( 1.00 usr +  0.00 sys =  1.00 CPU) @ 13
+10719.00/s (n=1310719)
               Rate gis::fast      perl        xs
gis::fast   23643/s        --      -69%      -98%
perl        77282/s      227%        --      -94%
xs        1310719/s     5444%     1596%        --

---- [ Formula: gcd ] ------------------------------------
perl      - distance from BWI to LAX: 0 miles
xs        - distance from BWI to LAX: 2324.06319758555 miles
gis::fast - distance from BWI to LAX: 0 miles

Benchmark: running gis::fast, perl, xs for at least 1 CPU seconds...
 gis::fast:  1 wallclock secs ( 1.07 usr +  0.00 sys =  1.07 CPU) @ 18
+270.09/s (n=19549)
      perl:  2 wallclock secs ( 1.07 usr +  0.00 sys =  1.07 CPU) @ 73
+080.37/s (n=78196)
        xs:  1 wallclock secs ( 1.05 usr +  0.01 sys =  1.06 CPU) @ 12
+98354.72/s (n=1376256)
               Rate gis::fast      perl        xs
gis::fast   18270/s        --      -75%      -99%
perl        73080/s      300%        --      -94%
xs        1298355/s     7006%     1677%        --

---- [ Formula: mt ] ------------------------------------
perl      - distance from BWI to LAX: 2324.06319758555 miles
xs        - distance from BWI to LAX: 2324.06319758555 miles
gis::fast - distance from BWI to LAX: 2324.06319758555 miles

Benchmark: running gis::fast, perl, xs for at least 1 CPU seconds...
 gis::fast:  1 wallclock secs ( 1.12 usr +  0.01 sys =  1.13 CPU) @ 17
+300.00/s (n=19549)
      perl:  1 wallclock secs ( 1.06 usr +  0.00 sys =  1.06 CPU) @ 67
+621.70/s (n=71679)
        xs:  2 wallclock secs ( 1.07 usr +  0.01 sys =  1.08 CPU) @ 12
+74311.11/s (n=1376256)
               Rate gis::fast      perl        xs
gis::fast   17300/s        --      -74%      -99%
perl        67622/s      291%        --      -95%
xs        1274311/s     7266%     1784%        --

---- [ Formula: tv ] ------------------------------------
perl      - distance from BWI to LAX: 2328.95218785171 miles
xs        - distance from BWI to LAX: 2324.51147636132 miles
gis::fast - distance from BWI to LAX: 2328.95218785171 miles

Benchmark: running gis::fast, perl, xs for at least 1 CPU seconds...
 gis::fast:  2 wallclock secs ( 1.09 usr +  0.01 sys =  1.10 CPU) @ 21
+720.91/s (n=23893)
      perl:  1 wallclock secs ( 1.09 usr +  0.00 sys =  1.09 CPU) @ 16
+439.45/s (n=17919)
        xs:  2 wallclock secs ( 1.05 usr +  0.00 sys =  1.05 CPU) @ 77
+1011.43/s (n=809562)
              Rate      perl gis::fast        xs
perl       16439/s        --      -24%      -98%
gis::fast  21721/s       32%        --      -97%
xs        771011/s     4590%     3450%        --

Benchmarking script:

#!/usr/bin/env perl
use strict;
use warnings;

use Benchmark qw(cmpthese timethese);
use Geo::Distance::XS;
use GIS::Distance;

# When benchmarking, need to have it call import/unimport before the
# code is executed.
my $orig_timethis_sub = \&Benchmark::timethis;
{
    no warnings 'redefine';
    *Benchmark::timethis = sub {
        my $sub = ('perl' eq $_[2] ? 'un' : '') . 'import';
        Geo::Distance::XS->$sub();
        $orig_timethis_sub->(@_);
    };
}

# Lon, Lat
my @coord = (-76.668851, 39.179689, -118.408618, 33.943532);

my $geo = Geo::Distance->new;
my $gis = GIS::Distance->new;

my %gis_formula = (
    hsin  => 'Haversine',
    polar => 'Polar',
    cos   => 'Cosine',
    gcd   => 'GreatCircle',
    mt    => 'MathTrig',
    tv    => 'Vincenty',
);

sub geo {
    my $d = $geo->distance(mile => @coord);
}
sub gis {
    my $d = $gis->distance(@coord[ 1, 0, 3, 2 ]);
    return $d->mile;
}

for my $formula (qw(hsin polar cos gcd mt tv)) {
    print "---- [ Formula: $formula ] --------------------------------
+----\n";

    $geo->formula($formula);
    $gis->formula($gis_formula{$formula});

    Geo::Distance::XS->unimport;
    printf "perl      - distance from BWI to LAX: %s miles\n", geo();
    Geo::Distance::XS->import;
    printf "xs        - distance from BWI to LAX: %s miles\n", geo();
    printf "gis::fast - distance from BWI to LAX: %s miles\n", gis();
    print "\n";

    my $benchmarks = timethese - 1, {
        perl        => \&geo,
        xs          => \&geo,
        'gis::fast' => \&gis,
    };

    cmpthese $benchmarks;
    print "\n";
}

[Limbic~Region]

gray,
...he probably will accept patches via

I have emailed the author. I would like to offer a patch but I am not sure what the problem is. I was in a hurry to get my project complete so I could go to sleep.

A good reason to use Geo::Distance::XS

First, thanks! I admit that I didn't try it for a few reasons. The first is that I assumed one was just an XS implementation of the other to include the bug. The second reason is that in the environment where this needed to run, XS is not an easy option. Without going into a lot of detail - local politics have created technical hurdles I didn't want to jump through (I was tired). It is also the reason I didn't use GIS::Distance. I was pretty sure (and you confirmed) that it produces correct results but the litany of dependencies without an internet connection was unreasonable.

I am happy that there are working solutions out there. I guess after reflecting on my lamentation, what I really should have asked is: Does anyone know of a Geo::Distance::Lite that works?

Regarding your benchmark. First, I suspect you didn't "You can stick with the pure Perl version by setting the GEO_DISTANCE_PP environment variable before using this module" This is because Geo::Distance gave the correct answer which it I know it doesn't (the point of this thread). Second, it would only be fair to add tye's version which avoids the whole OO overhead.

[gray]

Regarding your benchmark. First, I suspect you didn't "You can stick with the pure Perl version by setting the GEO_DISTANCE_PP environment variable before using this module" This is because Geo::Distance gave the correct answer which it I know it doesn't (the point of this thread).

$ENV{GEO_DISTANCE_PP} just triggers Geo::Distance::XS->unimport(), which my benchmark script calls directly. I also checked the results by hiding Geo::Distance::XS with Test-Without-Module and they were consistent. Note also the results for the 'gcd' formula- both Geo::Distance and GIS::Distance::Fast failed, whereas Geo::Distance::XS produced a reasonable result. Perhaps if you posted your test script we could dig further.

[Limbic~Region]

gray,
Perhaps if you posted your test script we could dig further.

Without Geo::Distance::XS present.

#!/usr/bin/perl
use strict;
use warnings;
use Geo::Distance;

my $geo = Geo::Distance->new;
for (qw/tv hsin polar cos gcd mt/) {
    $geo->formula($_);
    print $_, " ", $geo->distance('mile', 39.175555,-76.671388 => 33.9
+42222,-118.407222), "\n";
}
__DATA__
tv 9537.71241222878
hsin 2881.23714304656
polar 3114.47526297555
cos 2881.23714304656
gcd 12438.0476860875-9076.08896733252i
mt 2881.23714304656

Geo::Distance::XS

use strict;
use warnings;
use Geo::Distance::XS;

my $geo = Geo::Distance->new;
for (qw/tv hsin polar cos gcd mt/) {
    $geo->formula($_);
    print $_, " ", $geo->distance('mile', 39.175555,-76.671388 => 33.9
+42222,-118.407222), "\n";
}
__DATA__
tv 9538.66771882495
hsin 2881.23714304656
polar 3114.47526297555
cos 2881.23714304656
gcd 2881.23714304656
mt 2881.23714304656

Update: Ok, I am a freaking idiot who should not write code while up late at night. I swapped long/lat. This was further complicated by tye's method having in the same order as I was using which produces the correct result. I need to have my JAPH card revoked. Please forgive my gigantic stupidity.