Given that the `external/` folder contains various imported code/resources, all of which could affect functionality and/or rendering, it seems safest to simply run browser/font/integration/unit tests whenever any part of that folder is touched.
Looking at the coverage data there are cases where we attempt to insert inferred link-annotations *before* the annotationLayer has rendered, see [here](2348365874/blob/web/annotation_layer_builder.js (L246)), which shouldn't happen and why that's treated as an Error.
This is most likely caused by the asynchronicity of all the relevant code, since the `Autolinker` functionality can only be invoked after both the annotationLayer *and* the textLayer have finished rendering.
Given that those operations are asynchronous, by the time that they complete it's possible that the annotationLayer (and also the textLayer) has been replaced by a new instance. In that case we might thus attempt to inject inferred link-annotations before the "new" annotationLayer has rendered.
To avoid this intermittent issue, we now ensure that the annotationLayer and textLayer haven't changed between those layers rendering and the `Autolinker` functionality being invoked. (If they did change, then a future `render` call will trigger the inferred link-annotations handling).
By replacing the early return with optional chaining, a pattern that we already use in lots of places, the code becomes a tiny bit shorter and more importantly the code coverage for this file becomes 100 percent.
Given that the cursor tools are managed via the `PDFCursorTools` class, of which the `GrabToPan` instance is essentially a (semi) private implementation detail, the `GrabToPan.prototype.toggle` method is completely unused and can thus be removed.
This format is obviously not very efficient however it's been supported since "forever" and there's even examples using, hence it seems like a good idea to actually test this.
Given that any incoming data is already being ignored after loading has been aborted, it seems reasonable to reject the stream-capability to avoid it remaining in a pending state indefinitely.
*Note:* This is something that I noticed while looking at the coverage data, since the `ChunkedStreamManager.prototype.onError` method is not used and from a brief look at the history of the code it never appears to have been used either.
Obviously it's not yet possible to just migrate `gulp browsertest` to GitHub Actions, however it's already possible to at least run the browser tests there which allows collection of more code coverage data.
This should thus give us more realistic coverage numbers, since currently there's many `src/` files that have very low code coverage.
By taking advantage of the fact that the GitHub Actions runners provide multiple cores, these tests are also fairly fast:
- The ubuntu-latest/firefox job complete in ~9 minutes.
Normally entire PDFs are encrypted (or not).
But it is also possible to only encrypt attachments.
It is then also possible to *only* prompt for a password when the user opens
them.
In the existing flow, prompting for passwords happens because things are decrypted.
A specific error is thrown, caught, and the user is prompted.
To keep this flow working, this PR changes to decrypting attachments on demand,
instead of eagerly.
This sounds logical: to not read attachments on startup.
I’ve extensively tested this, not only with regular attachments, but also with outline items
and attachments in annotations.
This PR builds on GH-21234.
It’s an alternative to the naïve GH-20732.
Closes GH-20049.
Prior to PR 20321 the annotationLayer was hidden when there was no regular annotations on the page, which meant that if there were any inferred links (from the textLayer) the annotationLayer needed to be made visible but in such a way that it wouldn't override an explicit `hide`-call from the `PDFPageView` class.
With the changes in the aforementioned PR the annotationLayer is now always "visible", and this code can thus be simplified a little bit.
So that Ctrl+A, Ctrl+Z, etc. still fire on non-US keyboard layouts where
the physical "A" key produces a non-Latin character (Cyrillic, Greek,
some AZERTY combinations, ...). KeyboardManager now tries event.key first
and falls back to a US-layout translation of event.code (KeyA => a,
Digit1 => 1, Numpad1 => 1) when no shortcut is bound on event.key.
Also refactors KeyboardManager to store modifiers as a bitmask instead
of a serialized string, and treats a shortcut array without any
"mac+"-prefixed entry as applying on all platforms, letting us drop the
redundant "mac+X" duplicates of bare "X" entries across the editor code.
Identical embedded fonts and images across the merged documents are now
written once and shared, instead of being copied per source file.
And avoid to compress already compressed stream with Brotli.
