pdftotext

pdftotext

extract text from PDF files

curl -fsSL https://unflab.app/get | sh -s -- pdftotext

Installs to ~/.local/bin. Add --prefix /usr/local/bin to
put it somewhere else, or --uninstall to remove it.

Why it’s here

Escapes a dependency tree: Homebrew’s build pulls in libraries this tool never touches at run time.

Details

Built from that exact tarball, with its SHA-256 pinned in the
recipe. The binary links against nothing outside /usr/lib
and /System/, checked in CI before release.

pdftotext extracts the text from a PDF. Not OCR — it reads the text
that is already in the file, which is most PDFs that came from a word
processor, a browser, LaTeX or a report generator.

# Text to a file (in.pdf -> in.txt)
pdftotext report.pdf

# ...or to stdout, to pipe somewhere
pdftotext report.pdf - | grep -i invoice

# Preserve the page's column layout instead of reflowing
pdftotext -layout report.pdf -

# Just pages 3 to 7
pdftotext -f 3 -l 7 report.pdf -

# What is this file, and does it have text at all?
pdfinfo report.pdf
pdffonts report.pdf

If pdftotext gives you nothing, the PDF is probably a scan — page
images with no text layer. No amount of extraction will help; that needs
OCR, which this does not do.

Seven tools ship, all from one build:

• pdftotext — the text.
• pdfinfo — page count, size, producer, dates, encryption.
• pdffonts — which fonts a PDF uses and whether they are embedded.
• pdftops — convert to PostScript.
• pdfdetach — list and extract file attachments.
• pdfseparate — split into one file per page.
• pdfunite — join files back together.

Install

./install.sh                      # into ~/.local/bin
./install.sh --prefix /usr/local/bin
./install.sh --uninstall

About this build

This is the collection’s clearest case. brew install poppler pulls a
dependency closure of 50 formulae — cairo, glib, harfbuzz, nss,
nspr, gpgme, GnuPG, five X11 libraries and a full TLS stack among them —
and installs 13 binaries, in order to get text out of a PDF. You can
check that yourself with brew deps --tree poppler.

This package needs none of it. Poppler’s Qt, GLib, cairo, crypto and
image-codec backends are all optional and simply default to on; the
font backend can be set to generic, which drops fontconfig and
harfbuzz with it. What is left needs FreeType, and Brotli’s decoder
for PDFs that compress streams with it — both compiled from source and
linked statically here — plus zlib and iconv, which macOS provides.

otool -L on any of the seven binaries shows only libz, libc++ and
libSystem, all from /usr/lib.

Dropping fontconfig costs nothing here. Font matching — finding a
system font to stand in for one the PDF does not embed — matters for
rendering a page, not for reading its text. A two-column academic paper
with embedded Type 1 subsets extracts 6,152 words correctly, accents,
CJK, ligatures and -layout column positions included.

What is deliberately not here

Poppler also builds pdfimages, pdftoppm and cairo-backed
pdftohtml. Those are raster tools, and without the image codecs this
build leaves out they do not merely lose features — pdftoppm -png
produces no file at all. A binary that silently does nothing is worse
than one that is absent, so they are not shipped.

If you want those, you want a full poppler with its image stack, and
Homebrew is the right way to get it. This package is for the text.