Jtdcftul

I have a file called data whose contents are

id,col1,col2

0,-0.3479417882673812,0.5664382596767175

1,-0.26800930980980764,0.2952025161991604

2,-0.4159790791116641,-1.3375045524610152

3,-0.7859665489205871,-0.6428101880909471

4,-1.3922759043388822,-1.676262144826317

5,-1.2471867496427498,-0.4912119581361516

6,1.443385383041667,1.6974039491263593

7,-2.058899802821969,2.0607628464079917

8,-0.10641338441541626,0.035929568275064216

9,-0.517273684861199,-0.6184800988804992

10,-0.9934859021679552,1.0577312348984502

11,0.5923834706792905,-0.6693757541250825

12,0.8657741917554445,-0.6876271057571398

13,-1.2061097548360489,-0.7402582563022937

14,0.78768021182158,-0.38607117005262315

Sorting numerically (-n) on the first column gives

$ sort -nk1 -t"," data

0,-0.3479417882673812,0.5664382596767175

id,col1,col2

1,-0.26800930980980764,0.2952025161991604

2,-0.4159790791116641,-1.3375045524610152

3,-0.7859665489205871,-0.6428101880909471

4,-1.3922759043388822,-1.676262144826317

5,-1.2471867496427498,-0.4912119581361516

7,-2.058899802821969,2.0607628464079917

8,-0.10641338441541626,0.035929568275064216

9,-0.517273684861199,-0.6184800988804992

10,-0.9934859021679552,1.0577312348984502

13,-1.2061097548360489,-0.7402582563022937

6,1.443385383041667,1.6974039491263593

11,0.5923834706792905,-0.6693757541250825

12,0.8657741917554445,-0.6876271057571398

14,0.78768021182158,-0.38607117005262315

This absolutely bizarre to me. I read in the man page that -n is supposed to be numerical sort. Why would id be placed in-between numbers? How is it that 10 is larger than 9, but smaller than 6, all the while 11 being greater than them all?

The -g seems to work as I want (and as I think is natural), but this -n option totally escapes me. What's this about? I think it can be related to locale, but once I specify the delimiter as being ,, I don't think that would explain it.

TL;DR

Use sort -nk1,1 -t, or otherwise with -k1 you're sorting on the full line where , is discarded in numbers as it's interpreted as a thousand separator.

Details

In English language locales, , is the thousand separator, which sort ignores in the integer part of numbers.

In other words, in an English language locale, or any locale where , is a thousand separator (see the output of locale thousands_sep), when sort -n sees 11,000,000 it doesn't see the 11 number followed by some ignored garbage, but the 11000000 number. Similarly 11,0 is not 11 but 110.

Now (and that's something many people trip on), -k1 defines a key that starts with the first field, but as you didn't specify where it stops, ends at the end of the line, so the sort key is the full line, which is the default.

So sort -nk1 -t, is exactly the same as sort -n.

With , ignored as a thousand separator, on your input sort is actually sorting these numbers:

0

1

2

3

4

5

61.4433853830416671

7

8

9

10

110.5923834706792905

120.8657741917554445

13

140.78768021182158

So it's not 6 vs 10 vs 11, but 61.4433853830416671 vs 10 vs 110.5923834706792905.

Here, you want:

sort -nk1,1 -t,

To sort on the first ,-delimited field only. -k1,1 defines a sort key that starts at the start of the first field and ends at the end of the first field.

You can also use sort -n in the C locale where , is neither the decimal radix nor the thousand separator (and . is the decimal radix):

LC_ALL=C sort -n

sort -g works differently because sort then uses strtold() to interpret the key as a number and strtold() doesn't recognise thousands separators.

As far as the id header line is concerned, in a numeric comparison, that id... is interpreted as 0 as there's no number to be seen there. It sorts after the line that starts with 0 because when two records sort the same (here with -n in a numeric comparison) sort does a last resort comparison which is a lexical comparison of the full line (and 0 sorts before i).

With some sort implementations, that last resort comparison can be disabled with -s. Here LC_ALL=C sort -sn would put the id line first, but that's only because there are no negative keys in the input (id (which again numerically is 0) would still sort after -1). If you want to exclude the first line from the sorting, you can do:

(head -n1; LC_ALL=C sort -n) < file