From: Stefano Zacchiroli <zack@upsilon.cc>
Date: Wed, 4 Feb 2009 18:14:23 +0000 (+0100)
Subject: sort + fosdem post
X-Git-Url: http://git.upsilon.cc/?p=homepage.git;a=commitdiff_plain;h=4ab4b10e2e12d19e43be89c890111f4057096152

sort + fosdem post
---

diff --git a/blog/archives/2009/02.mdwn b/blog/archives/2009/02.mdwn
new file mode 100644
index 00000000..238bda07
--- /dev/null
+++ b/blog/archives/2009/02.mdwn
@@ -0,0 +1 @@
+[[template id=archive_month year="2009" month="02"]]
diff --git a/blog/posts/2009/02.mdwn b/blog/posts/2009/02.mdwn
new file mode 100644
index 00000000..ff29ca95
--- /dev/null
+++ b/blog/posts/2009/02.mdwn
@@ -0,0 +1 @@
+[[meta redir=archives/2009/02]]
diff --git a/blog/posts/2009/02/sort_gotcha.mdwn b/blog/posts/2009/02/sort_gotcha.mdwn
new file mode 100644
index 00000000..b959f068
--- /dev/null
+++ b/blog/posts/2009/02/sort_gotcha.mdwn
@@ -0,0 +1,42 @@
+# sorting text records on the first column only
+
+It turns out that [`join`](http://man.cx/join) is a strange
+beast. Let's say you have two files, which are textual records using
+TAB as separator, containing different columns beside the first one,
+which is the record key. Using join you can do cool stuff like `join
+file1 file2` to obtain a join in the relations sense, by default on
+the first column.
+
+The [join manpage](http://man.cx/join) states that the two fields must
+be sorted on the join field. This is of course to avoid complexity
+explosion. The annoying thing is that if you forget to do that, you
+will miss tuples, without any warning, which is not good (and can
+actually be at least *spotted* avoiding complexity explosion ...).
+<small>So yeah, if I'm here it's because after having been hit by
+this, I finally read the fine manpage :-) </small>
+
+But how to sort on the join field? Turns out that dear old `sort -u
+file | [sponge](http://packages.debian.org/sid/moreutils) file` is not
+enough. Why? Because by default it will use *all* keys to sort and
+then you will end up sorting also on data values; needless to say that
+such sorting methods is not stable among the two different files you
+want to sort in turn.
+
+OK then, let's try with `sort -u -k 1 file | sponge file`. Not even,
+but close. ... because [`sort`](http://man.cx/sort) with this syntax
+will *start* sorting at key 1 (which is the default), but then
+continue sorting with the other keys, in the hope of doing you a
+favour.
+
+The magic line is then `sort -u -k 1,1 file | sponge file`, which
+tells `sort` to (damn) sort using only the first (damned) key and
+stopping there. Thank you `sort`. Have a nice day.
+
+<a href="http://www.fosdem.org"><img
+src="http://www.fosdem.org/promo/going-to" alt="I’m going to FOSDEM,
+the Free and Open Source Software Developers’ European Meeting" /></a>
+**PS** I've been sick this week, but I'm definitely getting better, so
+I can finally announce that I'm going to FOSDEM 2009, yay, see you
+there!
+
+[[!tag lang/english planet/debian]]
diff --git a/blog/posts/2009/02/sort_gotcha/going-to-fosdem.png b/blog/posts/2009/02/sort_gotcha/going-to-fosdem.png
new file mode 100644
index 00000000..11fad7b6
Binary files /dev/null and b/blog/posts/2009/02/sort_gotcha/going-to-fosdem.png differ