Sed Tester

How to use it

1. Write the script in the top-left editor — s/old/new/g to start, or ; and newlines to chain commands such as 2,3p or /error/d.
2. Paste the text underneath, or drop a .txt or .log file on the input pane; the file is read by the browser and never sent anywhere.
3. Tick -n when only explicit prints should reach the output, and -E when you want ( ) + ? | as operators instead of backslashed BRE forms.
4. Switch the right pane to Trace to see each cycle read a line, run every command against it, and print — with the hold space beside it.
5. Copy or download the output, hand the whole session over as a state file, or share a link — links carry options only, never your text.

Write a sed script, paste some text, and watch it run one cycle at a time: every address decision, every substitution, and both registers — pattern space and hold space — after each command. Execution stays on this device, in a background worker under a kill timer. The dialect is a pinned subset of BSD sed with POSIX as the floor, and anything outside it is refused by name rather than quietly imitated.

How it works

sed is a stream editor: it never sees the whole document at once. Each cycle it reads one line, strips the trailing newline, and places the line in the pattern space. Every command in the script then runs in order — but only if its address selects the current line. An address can be a line number, $ for the last line, a /regex/, a first,last range, or any of those negated with !. When the script ends, the pattern space is printed with its newline put back — unless -n is set — and the next cycle begins. That auto-print is why p appears to double lines and why -n and p travel together.

The hold space is the second register. It survives across cycles, which is what makes multi-line tricks possible: h and H copy or append the pattern space into it, g and G bring it back, and x swaps the two. The classic reverse-a-file one-liner 1!G;h;$p is nothing but those moves — G glues the accumulated reversal onto every line after the first, h stores the growing result, and $p prints it once at the last line. The trace view exists to make exactly this kind of script legible: it snapshots both registers and everything emitted after every single command.

Regular expressions here are POSIX BRE by default — \( \) groups, \{n,m\} intervals, backreferences \1 through \9 — and ERE with -E, where the brackets drop their backslashes. Patterns are translated and executed by the browser's own JavaScript engine, and where the two regex families genuinely disagree the run says so in a note instead of hiding it. Runs are budgeted: past 200,000 cycles or 2,000,000 command steps the tool refuses with both the budget and the actual count, because a silently truncated answer would look exactly like a correct one.

sed sits in a small family of text tools on this site, and each one keeps its own job: when the work is per-column arithmetic or report shaping rather than per-line editing, hand it to the awk runner; authoring and stress-testing a regular expression on its own is the regex tester's territory; and a one-off rewrite over a pasted body of text with no script syntax at all is what the regex replace workbench is for. This page is where a stream of lines meets a sed script, one cycle at a time.

Examples

alpha one
beta two
gamma three
delta four

one-alpha
two-beta
three-gamma
four-delta

first line
second line
third line
fourth line

second line
third line

deploy started
cache warmed
ready to serve

ready to serve
cache warmed
deploy started

yyy

-y-y-y-

Every example on this page is executed against the same engine that powers the tool, on every build.

Common errors

Line 1, column 8: an occurrence number and g cannot combine on BSD sed — pick the N-th match or every match

GNU sed reads 2g as from the second match onward; BSD refuses the combination outright, and so does this page, because the two dialects would silently give different answers. Use an occurrence number alone, or g alone.

Line 1, column 1: v is not a sed command this tool knows — the supported set is s y d p P D q n N = h H g G x

Usually a quoting slip: a stray character ended up where a command belongs, or a command from another tool's syntax leaked in. The supported subset is printed in full on this page, and everything else refuses with its name.

Line 1, column 5: \U case conversion is GNU-only — BSD sed prints a literal U instead of converting

Case conversion in the replacement is a GNU extension. On a Mac the same script does not error — it quietly writes a literal U into your text, which is worse. A y/// transliteration covers the alphabetic cases portably.

Line 1, column 1: \2 is not defined in the pattern — it opens 1 group

A backreference can only point at a group the pattern actually opens before it. Count the \( pairs — nested groups number by their opening parenthesis, left to right.

Input is larger than the 1 MB limit for this tool — this input is 3.0 MB. For bigger files use the CLI: sed 's/old/new/g' input.txt

A refusal, not a trim: a partial pass would report an answer for text never read. The command line handles any size — and the s/// you already wrote here pastes straight into it.

Stopped after 10 s — this script and input exceeded the time budget on this device.

Runs execute in a background worker under a kill timer: when the timer trips, the worker is destroyed and a fresh one takes the next run, and the page itself stays responsive. Narrow the addresses, or trim the input, and run again.

Keyboard shortcuts

⌘/Ctrl + Enter

Run the script now, skipping the debounce.

⌘/Ctrl + S

Download the output as .txt (overrides the browser's save dialog).

⌘/Ctrl + C

Copy the output pane while nothing is selected.

⌘/Ctrl + Shift + C

Copy the output pane even with a selection active.

t

Cycle to the next theme.

? or ⌘/Ctrl + /

Open the shortcut help overlay.

Esc

Dismiss the topmost overlay.