Edit PDF

Why editing a PDF is harder than editing a document

A PDF has no paragraphs. A page is a drawing program, and text in it is a series of instructions to place particular glyphs at particular coordinates — there is no record that those glyphs form a sentence, a line or a column. Everything this editor does begins by reconstructing that lost structure: the content stream is parsed back into its drawing operations, the fonts the page uses are read out of its resources, and the glyph codes are turned back into readable characters through each font's own character map. Modern writers embed two-byte subsets whose codes are raw glyph numbers with no relationship to Unicode at all, so without that map a page's text is genuinely unreadable rather than merely awkward.

Re-setting a line then needs the font's metrics, not just its name — the advance width of every glyph, read from the flat width array a simple font carries or from the run-length form a large font uses — because the new words have to be measured in the document's own typeface before they can be placed. This is why a rewritten line keeps the original's font, size and colour instead of being covered with a white patch and retyped, and it is also the source of the honest limit: text does not re-flow. A longer sentence takes up more of its line and does not push the next line down, because there is no next line to push — only more instructions at fixed coordinates. Acrobat behaves the same way, for the same reason.

Embedded fonts are usually subsets, carrying only the glyphs the document already uses. Type a character the subset does not contain and there is no glyph to draw, so the editor says so rather than writing a blank or silently substituting a different typeface. Newly added text sidesteps the problem by embedding a full Unicode face covering Latin, Cyrillic and Greek — which is why the Add text tool can type Polish, Turkish or Russian into a document that has never contained them. Scripts needing contextual shaping are refused outright: Arabic letters change form by position and Bengali conjuncts are single glyphs, and placing their code points one after another produces text that is not merely ugly but wrong.

Coordinates flip exactly once. The interface works from the top of the page down because that is how people read; a PDF measures from the bottom up. Every mark is stored in the first space and converted to the second at the moment it is written into the file, in one place, which is the only discipline that reliably keeps a signature from landing upside down or a highlight from sitting an inch below its sentence.

When you want something else

A scanned page has no text to edit — it is a photograph of text, and there is nothing in the content stream but an image. Run it through OCR first so there are words to work with, or accept that Cover up and Add text is the only honest option on that page.

When the edit is really a rewrite — reflowing a paragraph, moving a table, restyling a document — no PDF editor is the right tool, including the expensive ones, because the structure you want to change was discarded when the PDF was made. Go back to the source document if it exists, or convert to Word and edit there. For scripted changes across many files, pikepdf and mutool clean -d expose the content stream directly and neither needs a page to be rendered first.