Re: Optimise file line by line parsing, substitute SPLIT

[Tux]

If your data is easy to split on TAB's, perl's split is exceptionally fast. Building on BrowserUk's example - which is a good example - look at this test:

$ perl -E'say join"\t",qw(1 aap 2 noot 3 mies 4 wim 5 zus 6) for 1..2_
+000_000' > test.tsv
$ wc -l test.tsv
2000000 test.tsv
$ cat test.pl
use 5.016;
use warnings;

use Benchmark qw( cmpthese );
use Text::CSV_XS;

my $csv = Text::CSV_XS->new ({ sep_char => "\t", auto_diag => 1 });

cmpthese (3, {
    none        => sub {
        open my $fh, "<", "test.tsv";
        while (<$fh>) {
            }
        },
    splt        => sub {
        open my $fh, "<", "test.tsv";
        while (<$fh>) {
            my @f = split m/\t/ => $_, -1;
            }
        },
    tcsv        => sub {
        open my $fh, "<", "test.tsv";
        while (my $f = $csv->getline ($fh)) {
            # use $f->[1] instead of $f[1];
            }
        },
    tcsvbc      => sub {
        open my $fh, "<", "test.tsv";
        my @f = ("") x 11;
        $csv->bind_columns (\(@f));
        while ($csv->getline ($fh)) {
            }
        },
    });
$ perl test.pl
       s/iter   tcsv tcsvbc   splt   none
tcsv     5.26     --   -32%   -36%   -96%
tcsvbc   3.57    48%     --    -5%   -93%
splt     3.38    56%     6%     --   -93%
none    0.233  2156%  1429%  1347%     --
$

where vsespb is missing the point, is that when determining if a record is to be split or not, one still has to look at the record. That is exaclty what BrowserUk was pointing at. Looking at the first character of a record is a cheap operation, not needing split at all, but if you want to decide processing on the content of the 11th field, you have to split before you can decide.

All of the above will start changing when the fields can be quoted and have (valid) TAB characters inside the field, or if fields are allowed to have embedded newlines or when the fields are (much) longer. Compare to e.g.:

$ perl -E'@x=("x" x 100) x 11;say join"\t",@x for 1..2_000_000' > test
+.tsv
$ perl test.pl
       s/iter   tcsv tcsvbc   splt   none
tcsv     23.3     --   -32%   -74%   -92%
tcsvbc   15.8    47%     --   -61%   -89%
splt     6.13   280%   158%     --   -71%
none     1.78  1206%   788%   244%     --

[sundialsvc4]

... and maybe simply looking for the string-of-interest anywhere in the original-string might be enough of a filter.   Decide whether the string exists anywhere, before you spent the time needed to decide if it’s the 11th field, instead of identifying the 11th field before you consider what it contains.   If split is deemed to be pricey, split only what you (guess) you (might) need to.