Convert RTF to PDF

Very few people write an RTF on purpose. They are produced by systems: a transcription service returning a dictated letter, a legal or medical practice system assembling a document from a template, a mail-merge run that emitted three hundred personalised letters, a decade-old database whose report writer only knew one output format.

That origin explains the shape of the problem. The file is not yours, you did not choose the format, and what you actually need is a document you can send to a client, print for a file, or attach to a case — none of which an RTF is good at. Converting is not an upgrade of the content; it is settling the container. It also removes a small awkwardness in the exchange: sending a client an RTF invites the question of what the file is, while a PDF invites no question at all and can be opened on a phone in a car park.

Rich text describes formatting, not pagination. Where the pages break is worked out at the moment the file is opened, by whichever program opens it, using whichever fonts that machine happens to have. WordPad, Word, LibreOffice Writer and TextEdit can all produce a different page count from the same file, and all of them are behaving correctly.

For a letter that will be printed and signed, or a report that refers to a page number, that is not acceptable. A PDF resolves the layout once, on one machine, with one set of fonts, and then every subsequent reader sees the same thing. That single property is the reason this conversion exists. It also means the page count stops being a variable: a letter that is two pages in the PDF is two pages when it is printed, filed, scanned back in and counted by somebody billing for it.

This direction surprises people, because converting to PDF is normally associated with growth. RTF has no compression and writes embedded pictures as hexadecimal text, which costs about two characters for every byte of the original image. A document with a scanned letterhead and three photographs is therefore enormous as rich text.

A PDF stores the same picture as binary data inside a compressed stream, and compresses the page content as well. The saving on a text-only document is modest; on an illustrated one it is often a large multiple. If an RTF has been bouncing off an attachment limit, converting it is frequently the entire fix. The exception is a document whose pictures are already small and whose text is long, where the PDF has to carry font information the RTF did not. That case can go the other way by a modest margin, and it is worth a glance at the file size rather than an assumption in either direction.

An RTF carries a font table listing the typefaces it wants and cannot embed them. Whatever renders the file substitutes for anything it does not hold, and a substitute with different metrics rewraps every paragraph — which is how a two-page letter arrives as three with a widow on the last line.

The machine that performs this conversion carries the Liberation family, Carlito and Caladea, which are metrically identical stand-ins for Arial, Times New Roman, Courier New, Calibri and Cambria, alongside DejaVu and Noto for wider coverage including CJK text. Documents written in those common faces convert without moving. A file set in something unusual is the one to check. Once the PDF exists the question is settled for good, because the fonts used to render it are described inside the document, and no future reader has to have them installed to see the same lines break in the same places.

Tables, headers, footers, page numbering, footnotes, lists and hyperlinks all have direct representations in rich text and render as you would expect. The things that move are the ones that were already ambiguous: a table wider than the page, a picture anchored to a paragraph near a break, a header that assumed a particular margin.

System-generated RTFs are usually simple and convert without incident, because the program that wrote them used a narrow subset of the format. Hand-edited ones are less predictable, especially if they have been through several programs. Look at the first and last pages of the PDF rather than only the first.

Rich text is a seven-bit format. Anything outside plain ASCII is written as an escape interpreted against a character set the file declares in its header, which works perfectly when the declaration is right and produces the familiar garbled surname when it is not. Systems that emit RTF have been getting this wrong for thirty years.

The conversion renders what the file says rather than repairing it, so a name that was already wrong stays wrong and is now fixed permanently in a page. This is the one thing genuinely worth checking before converting a batch of correspondence: look at a letter addressed to somebody with an accent in their name, and if it is broken, fix it at the system that produced the file rather than downstream.

The PDF is the document as it would print. Comments and annotations are not drawn, revision marks are not shown, hidden text stays hidden, and any field that was still a field resolves to whatever value it currently holds rather than remaining live.

That is the correct behaviour for a document being filed or sent, and it is worth one deliberate check before you convert. If a merge field in a generated letter never received its value, the RTF shows you an empty space and the PDF preserves that empty space permanently. The conversion cannot know that the blank was a mistake.

The characteristic volume case for this pair is a directory of near-identical letters produced in a single run: statements, appointment confirmations, renewal notices. Drop the whole folder rather than working through it, and each file is converted in turn, keeps its filename with the extension replaced, and the set returns as one ZIP.

Check three of them rather than one — the first, one from the middle, and the one with the longest recipient name, because that is where a template overflows a line and pushes something onto a second page. A merge run that is wrong is usually wrong in a pattern, and one sample will not show you the pattern. It is also worth counting the output files against the number of records that went into the merge. A run that silently produced two hundred and ninety-eight letters from three hundred rows is a problem the conversion cannot see and the recipients certainly will.

Most conversions here run locally and transmit nothing. This one cannot: rendering a word-processing document to a page requires a full office engine, so the file is sent over an encrypted connection to a container that runs LibreOffice headless and holds nothing else. Saying so is more useful than an unqualified privacy claim that would not be true.

The container has no outbound internet route. Each job gets its own scratch directory and its own LibreOffice profile; the directory is deleted when the job ends, whether it succeeded or failed, and a job still running after sixty seconds is killed rather than left to hang on a malformed file. The free ceiling is 25 MB per document.

If the document might still be amended — a draft letter awaiting approval, a report that will get another paragraph — keep the rich text file. The PDF is a rendition of a moment, and going back the other way means recovering an editable document from a fixed one, which is guesswork rather than conversion.

The pattern that works in a case file is to keep both: the RTF as the working copy, the PDF as the version that was actually sent, named so that the two are obviously the same document. That costs one conversion and removes the argument about which copy is authoritative.