Sed 101

What the hell is sed and why should you care?

sed is the stream editor — it’s a command-line tool that reads text one line at a time, applies transformations based on patterns you give it, and spits out the result. Think of it as a scalpel for text files: surgical precision, no fluff.

Here’s the thing: if you’re doing anything with configuration files, logs, or bulk text changes, sed will save you hours of manual editing. It’s been around since the 1970s for a reason. Your future self at 2 AM will appreciate not having to open a file in vim and search-replace 500 times.

The basic syntax (that 90% of your usage falls into)

The bread and butter of sed is the s command (substitute). Here’s the shape of it:

sed 's/find/replace/' filename.txt

That’s it. Replace the first occurrence of “find” with “replace” on each line. But there’s more power hiding in those flags.

Flags that actually matter

The g flag — replace ALL occurrences on a line

sed 's/dog/cat/' pets.txt

If your line is “I have a dog and my dog has fleas,” only the first “dog” becomes “cat.”

sed 's/dog/cat/g' pets.txt

Now both “dog”s become “cat”s. Use g by default — you almost always want it.

The i flag — case-insensitive

sed 's/Docker/podman/i' config.yaml

Matches “docker”, “Docker”, “DOCKER”, and replaces with lowercase “podman” (the pattern is literal, only the match is case-insensitive).

The n flag — silent mode (print only matches)

By default, sed prints every line. The n flag suppresses automatic output, and p prints only lines that match. Useful for filtering:

sed -n 's/error/ERROR/p' app.log

Only prints lines that had the substitution.

Multiple expressions with -e

Want to chain multiple replacements?

sed -e 's/cats/sloths/g' -e 's/dogs/ducks/g' animals.txt

Each -e is applied in order. This is cleaner than piping sed to sed (which works, but is inelegant).

Line addressing — replace only where you want

Specific line numbers

sed '5s/old/new/g' file.txt

Replace only on line 5.

Line ranges

sed '2,7s/old/new/g' file.txt

Replace on lines 2 through 7.

Pattern matching

sed '/^DEBUG/s/old/new/g' app.log

Replace only on lines starting with “DEBUG”.

sed '/start/,/end/s/old/new/g' file.txt

Replace between the first line matching “start” and the first line matching “end” (inclusive).

Deleting lines (the d command)

Sometimes you want to remove lines entirely:

sed '/^#/d' config.conf

Delete all comment lines (starting with #).

sed '10,20d' file.txt

Delete lines 10–20.

Printing specific lines (the p command with -n)

sed -n '5,10p' file.txt

Print only lines 5–10 (like head -n 10 | tail -n 6, but shorter).

In-place editing with -i (the dangerous one)

By default, sed outputs to stdout. To actually overwrite the file:

sed -i 's/old/new/g' file.txt

But here’s the gotcha: macOS’s BSD sed is different from GNU sed. On macOS, you must provide a backup extension:

sed -i '' 's/old/new/g' file.txt

Without the empty string, it’ll create a .bak file. Forget this once and you’ll be searching StackOverflow forever. Write it into your muscle memory now.

To be safe across systems, always use:

sed -i.bak 's/old/new/g' file.txt

Creates a .bak backup on all platforms, which you can delete later (or keep, your call).

Dealing with special characters (the / problem)

If your find/replace contains slashes (like file paths), change the delimiter:

sed 's;/home/kingpin;/home/notkingpin;g' paths.txt

Or use any other character that’s unlikely to appear in your data (;, |, #, @).

Real-world use cases

Strip comments from a config:

sed '/^#/d' nginx.conf | sed '/^$/d'

Delete comment lines AND blank lines. Pipe it to tee to overwrite in place:

sed -i '/^#/d; /^$/d' nginx.conf

Fix Windows line endings (CRLF to LF):

sed -i 's/\r$//' filename.txt

Extract and clean up log lines:

sed -n '/ERROR/p' app.log | sed 's/ERROR:/[!]/g'

Bulk rename variables in code:

sed -i 's/\bOLD_VAR\b/NEW_VAR/g' *.py

The \b is a word boundary — matches “OLD_VAR” but not “OLD_VAR_SUFFIX”.

Quick cheatsheet

That’s it. You now know enough sed to handle 95% of real-world situations. The man page is a rabbit hole of obscure features — you don’t need them yet. Start with s/find/replace/g, and level up only when you hit a wall.

Your config files (and your 2 AM self) will thank you.