How to cancel a PDF operation safely

Updated 18 August 2026

Press Stop. The label tells you what stopping means for that operation — within the current item, after the current page, or not during the final rebuild — and the message afterwards reads “Stopped. Nothing was saved and your original file is unchanged.” Nothing resumes: all ${RESTART_ALL_N} tools are recorded as restart-required.

What Stop actually does in a browser PDF tool: which tools stop where, what the message means, and why nothing in the product claims to resume.

1. Start the operation and read the hint next to the progress row — it says where this particular run can stop.
2. Press Stop when you want it to end. On most tools it stops within the current item.
3. Read the confirmation: “Stopped. Nothing was saved and your original file is unchanged.”
4. Check that the tool is usable again — the configuration stage returns and its action button is enabled.
5. Start again from the beginning if you still want the result. There is no resume anywhere in the product.

What Stop says before you press it

A single vague “Cancel” is a promise a tool cannot keep uniformly, so each operation carries the label that is true for it. Twenty-six tools read “You can stop this — it stops within the current item.” Seven read “You can stop this — it stops after the current page finishes” — recognition work, where a page cannot be abandoned half-way. Four read “This cannot be stopped during the final rebuild”, the phase where the new document is assembled, and one more is item-cancellable right up to its rebuild and then not. Four — the image tools — read “Stopping discards the work — there is no resume; you would start again”, because each runs as one pass with nothing part-finished to keep. One tool is recorded as not wired for cancellation at all, in the published registry, rather than showing a button that does nothing.

Those six counts add up to 43, which is the whole product. They are enumerated from capabilities.json, the same machine-readable record the build refuses to emit if a row contradicts the shipped code.

What Stop does when you press it

The run ends, and everything it allocated is released in reverse order of acquisition — canvases have their backing store zeroed, image bitmaps are closed, object URLs are revoked, vendor threads are terminated, large arrays are dropped. Each disposal is independent, so failing to clean one resource can never overwrite the run’s real outcome. Then the message: “Stopped. Nothing was saved and your original file is unchanged.” Both halves of that sentence are literal. No output document is handed over — a cancelled run that still produces bytes has written something you asked it not to — and your input was never written to in the first place, because Merge PDF and every other tool build a new document rather than editing yours in place.

That was checked after the fact as well as by design. Across the 124 measured runs of 15 August 2026, 44 of the 45 corpus files were untouched by modification time; the single exception had been rewritten by an unrelated test script before the matrix started and was not one of its fixtures.

Nothing resumes, and that is deliberate

All 43 tools are recorded as restart-required. The registry’s resumable value is defined so that it can never be assigned: the underlying PDF libraries need a full restart to continue work, so a resume claim would be a lie the code could not honour, and a field that cannot hold a lie is better than a policy someone has to remember. If you stop a 100-page recognition at page 90, you start again at page 1. That is worth knowing before you press Stop on something long — OCR on a large scan is the operation where it costs the most.

How we know the button is real

A unit test can prove a runtime exposes a cancellation contract and prove nothing about whether a tool honours it: a loop that never checks would pass every assertion and run happily to completion after you pressed Stop. So cancellation is driven through four real tools in a real browser on every gate run — compress, recognition, page images and merge, chosen because they cancel through four different structures — asserting that a Stop control exists while the run is in flight, that pressing it ends the run within a bounded time, that no result is produced, that the configuration stage returns with its action button enabled, that the console stays clean, and that a fresh run afterwards produces a real result. That last assertion is the one that catches the expensive failure: a cancel that leaves the page’s single-run lock latched or a half-consumed document behind.

Stopping is not the same as failing

Two other things can end a run, and neither is a stop. A document that cannot be opened produces a named refusal — on 15 August 2026 nine encrypted-file runs across three engines were refused in 0.6 seconds or less, every one of them actionable, though only three of the nine named the tool that fixes it. And a run that the pre-flight check will not start says so before anything begins: “we are stopping before it starts. Your file has not been changed.” Across 27 runs on damaged, encrypted, rotated and structurally deep documents there were zero silent hangs, zero crashes and zero “something went wrong”. What a long run that is merely slow looks like is in why browser PDF tools freeze on large files.

The full cancellation record per tool, the cleanup ledger and the suites behind all of it are in how UnboundPDF handles large PDFs locally. Practical large-document routes are in processing a large PDF without uploading, compressing a 100MB PDF locally and desktop versus mobile; the privacy claim you can verify yourself is in the PDF Privacy Lab, and dated changes are in the changelog.

Related

Compress PDF · OCR PDF · All 43 tools

Frequently asked questions

Does cancelling a PDF operation damage the file?

No. The file you opened is read into the browser and never written to — the tool produces a new document rather than editing yours in place. After 124 measured runs on 15 August 2026, 44 of the 45 corpus files were untouched by mtime and the single exception had been rewritten by an unrelated test script beforehand. Every refusal case left the page still holding the user's file.

Can I resume a cancelled PDF operation?

No, and no page on this site says otherwise. All ${RESTART_ALL_N} tools are recorded as restart-required, and the registry's resumable value is defined so that it can never be assigned to anything — the underlying libraries need a full restart, so a resume claim would be one the code could not honour. Stopping at page 90 of 100 means starting from page 1.

Why can't some operations be stopped immediately?

Because of where the work sits. Twenty-six tools stop within the current item. Seven — recognition-driven ones — stop after the current page finishes, because a page cannot be abandoned half-recognised. Four cannot be stopped during the final rebuild, when the new document is being assembled, and one is item-cancellable up to that rebuild. Four — the image tools — run in a single pass, so stopping discards the work rather than ending it part-way. One tool is recorded as not wired for cancellation, and says so in the registry rather than pretending.

Is Stop actually connected to anything?

It is tested on every gate run rather than assumed. A suite drives cancellation through four real tools — compress, OCR, page images and merge, chosen because they cancel through four different structures — in a real browser, and asserts that a Stop control exists mid-run, that pressing it ends the run, that no result is handed over, that the tool returns to a usable state with a clean console, and that a re-run afterwards produces a real result.