From what I understand, 2>&1 just means 'send stderr to the same place as stdout'. But for some reason, 1>foo 2>foo and 1>foo 2>&1 don't seem equivalent.
# Assuming file 'out' exists but file 'nosuchfile' doesn't
# 'foo' contains stdout and some partial stderr in CentOS 6
$ ls -l out nosuchfile 1>foo 2>foo
$ cat foo
-rw-r--r-- 1 user user 0 May 14 14:45 out
ctory
# 'foo' contains both stdout and stderr
$ ls -l out nosuchfile 1>foo 2>&1
$ cat foo
ls: cannot access nosuchfile: No such file or directory
-rw-r--r-- 1 user user 0 May 14 14:45 out
Can anyone explain why they behave differently?
> overwrites files; >> appends to files.
When you write 1> file and 2> file both streams will overwrite file in parallel and may therefore overwrite each other – a typical race condition.
command 1>> file 2>> file should keep all output of both streams.
Example:
$ n=1""000""000
$ (seq "$n" 1>&2 & seq "$n") 1> o 2> o
$ wc -l o
1000000
$ rm o
$ (seq "$n" 1>&2 & seq "$n") 1>> o 2>> o
wc -l o
2000000