Recipes

Five jobs this tool does every day. Each was run for real in the lab (transcripts noted); each scales from a dozen files to ten thousand without changing shape.

Checksum every file in a tree

From transcript t04:

ls samples/*/*.txt | parallel sha256sum {} > checksums.txt

Any list-printing command can feed it — find . -name '*.iso' is the usual one. Add --joblog when the run matters (chapter 9), and -k when a human will read the list in order.

Compress a folder without destroying the originals

From transcript t06 (-k here is gzip's own "keep the original" flag, not parallel's):

parallel --dry-run gzip -k {} ::: samples/north/*.txt
parallel gzip -k {} ::: samples/north/*.txt

The dry run first, always — it costs a second and shows every command before one runs. The same shape works for xz, zstd, or bzip2.

Turn each file into a derived file

From transcript t07 — upper-case copies of six notes, into a separate folder, names built from {/.}:

parallel 'tr a-z A-Z < {} > upper/{/.}.upper' ::: samples/south/*.txt

The pattern to keep: input from {}, output named from {/.} plus a new ending. It is also the shape of the classic conversion job, which reads the same with a converter in the middle:

parallel convert {} {.}.jpg ::: *.png

(ImageMagick was not part of this guide's lab; the pattern is the lab's, the converter is yours to supply.)

Rename in pairs

From transcript t05b. Two lists, walked in step with :::+, never combined:

parallel mv {1} {2} ::: draft-one.txt draft-two.txt :::+ final-one.txt final-two.txt

Dry-run it first. A paired rename with the lists in the wrong order is the fastest way to scramble a folder, and --dry-run is how you see it coming.

Summarise every file, one line each

From transcript t07:

$ parallel wc -w {} ::: samples/*/*.txt
44 samples/north/notes-01.txt
45 samples/north/notes-02.txt
...

Any per-file reporter drops in here — identify for images, ffprobe for media, pdfinfo for PDFs. The output is a table you can sort, total, or grep, and because parallel keeps each job's output together (chapter 8), the lines never tear.

The recipe behind the recipes

Every one of these is: a list, a template with the pieces cut by replacement strings, a slot count, and a dry run first. When a new job appears, assemble those four and you have it.