The/Large files

A gigabyte file is not a “please wait”.

An ordinary editor reads the whole file into memory and hopes it fits. The memory-maps the original and states plainly which mode it is in and what that mode switches off.

Download and try it See the modes

server.log— ingest-pipeline

server.logmetrics.csv

2026-07-01T00:00:00.004Z INFO boot seq=1

2026-07-01T00:00:00.041Z INFO mount /data ro=false

…

— jumping to a line is instant: the index is built in the background —

18,442,188

2026-07-28T21:14:02.377Z INFO flush ok

18,442,189

2026-07-28T21:14:02.402Z INFO checkpoint offset=2,214,883,072

18,442,190

2026-07-28T21:14:02.418Z INFO idle

main↑2✓0⚠02.4 GB · resident: a small fraction

LargeTextLn 18,442,188, Col 1UTF-8LFLOG

mechanics

Three decisions that add up to speed.

mmap

The original is mapped, not copied

Files from 100 MB open through memory-mapping: pages are faulted in on access, so opening costs almost no RAM and requires no full read from disk.

index

Line indexing runs in the background

Jumping to a line and the cursor position are instant while the interface never blocks on counting: the index is built in parallel and its status is visible in the status bar.

piece table

Editing anywhere is equally cheap

The buffer is assembled from pieces rather than rewritten: inserting at the start of a several-hundred-megabyte file costs the same as at the end. Undo is snapshot-based.

modes

The editor says what it turned off instead of stalling.

The mode is chosen by file size and shown in the status bar. This is the key difference from “sudden slowdowns”: behaviour is predictable and the limits are named out loud.

mode 01

Normal

Ordinary files. Everything on: highlighting, folding, project-wide search, git markers, language intelligence.

threshold: < 16 MiB
opening: plain read
limits: none

mode 02

LargeText

Big logs and dumps. The file is memory-mapped and heavy whole-file passes are constrained.

threshold: 16 MiB — 1 GiB
opening: mmap + background index
limits: some whole-file passes

mode 03

HugeSlice

Gigabyte files. Work proceeds in slices: the editor deliberately disables anything that needs to walk the whole content.

threshold: ≥ 1 GiB
opening: mmap, slice reads
limits: maximal, stated explicitly

placeholder

The exact switching thresholds and the list of features disabled in each mode must be confirmed against the 0.4.3 source before publication. The values above are a guide taken from project documentation, not a measurement.

data

A million-line log is easier to read as a table.

CSV and TSV table mode

Rainbow columns — the eye does not lose a column while scrolling.
The header stays pinned, line numbers are large.
Works on multi-million-row tables: the same mmap and background index.
Project-wide search on RE2 — results grouped by file.

All formats →

metrics.csv— ingest-pipeline

metrics.csv

	ts	shard	op	rows	ms
994,112	21:14:02.118	07	flush	2,048	18.4
994,113	21:14:02.140	08	flush	2,048	17.1
994,114	21:14:02.166	08	queue	8,412	—
994,115	21:14:02.201	03	rebalance	2,048	57.0
994,116	21:14:02.244	03	checksum	44,907	0.4
994,117	21:14:02.288	03	retry	44,907	42.6
994,118	21:14:02.301	03	flush	2,048	19.8
994,119	21:14:02.334	04	flush	2,048	16.9

main↑2✓0⚠01,412,908 rows · 6 columns

Ln 994,116, Col 3UTF-8LFCSV

one more thing

A big file is especially easy to ruin.

So saving is atomic: a crash or a power cut during a write will not leave a torn file. An open file is locked across processes, and if it changed on disk the editor tells you instead of silently overwriting it.

Download The Why it works this way