Syntax#

• sed -I extension - FreeBSD sed (continuous line-counter)
• sed -I[extension] - NetBSD and Illumos sed (continuous line-counter)
• sed -i extension - FreeBSD sed
• sed -i[extension] - NetBSD, OpenBSD, Illumos, BusyBox and GNU sed
• sed —in-place[=extension] - Illumos, BusyBox, and GNU sed

Parameters#

From a BSD sed manual

(a section like this appears in all current BSD sed manuals, and those of their derivatives)

It is not recommended to give a zero length extension when in place editing files, as it risks corruption or partial content in situations where disk space is exhausted, etc.

Don’t forget the mighty ed

There is definitely a use for sed and for in-place editing features of sed, but when the UNIX standard is extended, we should always ask why the old UNIX standard did not include that feature. Though UNIX is not perfect, the orthogonality and completeness of the tools has been developed to be quite near to perfection, at least for purposes that where visible around 1970: Text editing and automated text editing was surely visible around that time.

Actually, the idea of sed is not to edit a file in place, but to edit a stream. That’s why the name sed is a short form of stream editor. Take away the s, and you get the tool that was actually designed for file editing: ed:

printf 'g/what to replace/s//with what to replace/g\nw\nq\n' | ed file

or cat file_edit_commands | ed file.

Replacing strings in a file in-place

sed -i s/"what to replace"/"with what to replace"/g $file

We use -i to select in-place editing on the $file file. In some systems it is required to add suffix after -i flag which will be used to create backup of original file. You can add empty string like -i '' to omit the backup creation. Look at Remarks in this topic about -i option.

The g terminator means do a global find/replace in each line.

$ cat example 
one
two
three
total
$ sed -i s/"t"/"g"/g example 
$ cat example 
one
gwo
ghree
gogal

Portable Use

In-place editing, while common, is a non-standard feature. A viable alternative would be to use an intermediate file to either store the original, or the output.

sed 'sed commands' > file.out && mv file.out file
# or
mv file file.orig && sed 'sed commands' file.orig > file

To use the -i option with both the GNU and FreeBSD syntax an extension must be specified and appended to the -i option. The following will be accepted by both, and produce two files, the original version at file.orig and the edited version at file:

sed -i.orig 'sed commands' file

See a basic example given a file file:

$ cat file
one
two
three
$ sed -i.orig 's/one/XX/' file
$ cat file                       # the original file has changed its content
XX
two
three
$ cat file.orig                  # the original content is now in file.orig
one
two
three

A more complex example, replacing each line with line number:

$ printf 'one\ntwo\n' | tee file1 | tr a-z A-Z > file2
$ sed -ni.orig = file1 file2
$ cat file1.orig file2.orig
one
two
ONE
TWO
$ cat file1 file2
1
2
1
2

Why a backup file is required

In order to use in-place editing without a backup file, -i must be given a zero-length argument and FreeBSD sed requires an argument to -i, either appended or separate, while the GNU optional argument extension requires the argument be appended to -i. Both support appending the argument to -i, but without it being required -i'' command is indistinguishable from -i extension, and so a zero-length argument can not be appended to -i.

In-place editing without specifying a backup file overrides read-only permissions

sed -i -e cmd file will modify file even if its permissions are set to read-only.

This command behaves similarly to

sed -e cmd file > tmp; mv -f tmp file

rather than

sed -e cmd file > tmp; cat tmp > file; rm tmp

The following example uses gnu sed:

$ echo 'Extremely important data' > input
$ chmod 400 input  # Protect that data by removing write access
$ echo 'data destroyed' > input
-bash: input: Permission denied
$ cat input  
Extremely important data (#phew! Data is intact)
$ sed -i s/important/destroyed/ input
$ cat input
Extremely destroyed data (#see, data changed)

This can be mitigated by creating a backup by specifying a SUFFIX with the i option:

$ sed -i.bak s/important/destroyed/ input
$ cat input
Extremely destroyed data
$ cat input.bak
Extremely important data