Linux

Rule of thumb is never edit your originals. If you destroy your data, you are screwed.

I agree that you should never edit the originals. How ever you can inplace edit a file with sed:


sed -i 's/==*/==/g' file.txt (or what ever the file is called).

I use this a lot in post kickstart scripts (always with a copy of the original file created first.) It is really useful.

Even that doesn't edit "in place". It deletes the old file and creates a new one.

This is important since this has side-effects if you're not the owner of the file -- it changes the owner.

i was under the impression ed does it right.

Code:

mute@goflex:~$ ls -li input && cat input
7345 -rw-r--r-- 1 mute mute 38 Jul 27 20:32 input
line1 =
line2 ==
line3 ===
line4 ====
mute@goflex:~$ printf '%s\n' 'g/===*/s//==/g' w q | ed -s input
mute@goflex:~$ ls -li input && cat input
7345 -rw-r--r-- 1 mute mute 35 Jul 27 20:32 input
line1 =
line2 ==
line3 ==
line4 ==

posix, but not wasn't installed on debian squeeze minimal..

ed does edit the file in place, but it slurps it into memory. That's a lot of memory if the file is 38 GiB. Further, some (most?) ed implementations keep a copy of the entire buffer in a tmp file.

If a humongous file (larger than unused storage and available memory) needs to be edited in place, it can be, but you may have to craft a custom solution rather than use a general purpose editor.

Regards,
Alister

Hi.
And the inode, which may have consequences for some programs: inode - Wikipedia, the free encyclopedia ... cheers, drl

Further to @alister. How on earth did you end up in the situation of trying to amend a 38Gb flat file with unix Shell tools?
Do you have a database engine?