Home/Method/PDF structure

PDF file structure: the index near the end is the whole trick

A PDF is a container of numbered objects at arbitrary offsets, with an index at the end and one pointer, startxref, telling a parser where to begin. That is why page 3 can sit before page 1 in the bytes, why an incremental update contains the same object twice, and why a truncated file can still open.

Independent guide. We describe a documented format and parse text the reader pastes locally. Nothing here strips a permission or opens an encrypted document. Encryption and digital signatures are outside this page.

Findings

From Wikipedia's PDF article, File format section, retrieved 18 Oct 2026. Rows follow the source's wording.

The sentence that explains the rest. Objects live at whatever offset the writer chose, and the index maps each object number to that offset. Logical page order is a property of the page tree, not of the byte order.

Reading order, in pseudocode

open file as bytes
read first bytes                         # "%PDF-1.7" → declared version
scan forward from the END for "startxref"
offset = integer on the next line
read the index at that offset:
    if it starts with "xref":            # classic ASCII table
        entries = rows of (offset, generation, n|f)
    else:                                # xref stream, PDF 1.5+
        decode using the /W widths
follow /Root → catalogue → /Pages → each /Page's /Contents

The pointer does not say "start reading the file". A parser starts at the end, reads one table, then jumps. If the index is damaged, a recovery scan can look for every obj keyword, which is why truncated files often still open.

press Parse structure

Try it yourself

Take this PDF and do not extract its text. Report only structure: the
declared version in the header, how many indirect objects are present,
the highest object number, whether the file uses an ASCII cross-reference
table or a cross-reference stream, whether a /Prev entry exists, and the
byte offset named by startxref plus the first characters actually found at
that offset. If the two disagree, say so — that is the finding.

What we could not confirm. Which offset width a specific file dedicates (2 bytes is an example for files under 64 KiB, not a rule), the internal layout of cross-reference stream entries beyond /W, and whether your file's generator sorted objects by page. ISO 32000 section numbers are not verified here. This page cites the article's summary, not the standard text.

Next

Sources: PDF, Wikipedia, File format / Technical details / infobox, retrieved 18 Oct 2026. Comply with that article's licence when reusing wording. The parsing harness is our own code.