Repair PDF

What is usually broken, and the two ways back

Most PDFs that will not open are not damaged in any meaningful sense. A PDF ends with a cross-reference table — an index recording the byte offset of every object in the file — and readers use it to find anything at all. Corrupt that index and a perfectly intact document becomes unopenable, because nothing can be located even though everything is present. Interrupted downloads, a save that was cut short, a crashed application, a file edited by something that did not update the offsets, a transfer that mangled line endings: all of them break the index and none of them touch the content.

The first pass exploits exactly that. The file is parsed leniently, ignoring the stated offsets where they disagree with what is actually there, and then written out fresh with a correct index, correct stream lengths and a clean trailer. This is what recovers the large majority of files, it is quick, and nothing about the pages is altered — the result is your document with a working table of contents bolted back on.

When the file will not parse at all, the second pass does what a desktop recovery tool does: it scans the raw bytes from beginning to end looking for object headers, collects everything that still looks like a PDF object, and rebuilds the index from what is genuinely present rather than from what the file claims. The page tells you which of the two ran, because it matters — a rebuild means the original structure was unreadable, and the result deserves a look before you trust it.

Neither pass can invent data that is not there, and no tool anywhere can. If a download stopped at sixty per cent, the last forty per cent of the pages do not exist in the file; recovery finds the pages that arrived and stops. If the damage is inside a compressed stream, that stream cannot be decompressed and the page it drew comes back blank while its neighbours are fine. Recovering some of a document is the normal good outcome here, and getting the original file again is always better than repairing a truncated copy.

When you want something else

qpdf is the reference tool for this and does the same first pass more thoroughly: qpdf --replace-input damaged.pdf rewrites the file with a correct index, and it handles encrypted documents, object streams and linearised files that a lenient parse can stumble over. mutool clean -ggg in.pdf out.pdf goes further still, rebuilding the page tree and discarding anything unreferenced. Both are free, both run on files far larger than a browser tab can hold, and both tell you what they had to fix.

For a file damaged badly enough that object scanning is the only option, Ghostscript is often the most stubborn: gs -o repaired.pdf -sDEVICE=pdfwrite broken.pdf interprets the page content rather than merely re-indexing it, which sometimes reconstructs pages the other tools give up on — at the cost of rewriting everything, so the output is no longer byte-identical to what survived.