* quickjs-eval.js has been generated using https://github.com/mozilla/pdf.js.quickjs/
* lazy load of sandbox code
* Rewrite tests to use the sandbox
* Add a task `watch-sandbox` which update bundle pdf.sandbox.js on change in the sandbox code
Given that it's generally faster to call *one* function and have it loop through an object, rather than looping through an object and calling a function for every iteration, this patch will reduce the total time spent in `PDFViewerApplication._readPreferences` ever so slightly.
Also, over time we've been adding more and more preferences, rather than removing them, so using the new `AppOptions.setAll` method should be generally beneficial as well.
While the effect of these changes is quite small, it does reduces the time it takes for the preferences to be fully initialized. Given the amount of asynchronous code during viewer initialization, every bit of time that we can save should thus help.
Especially considering the recently added `viewerCssTheme` preference, which needs to be read very early to reduce the risk of the viewer UI "flashing" visibly as the theme changes, I figured that a couple of small patches reducing the time spend reading preferences cannot hurt.
Given that only two debugging hash parameters (i.e. `disableWorker` and `pdfBug`) will make this method asynchronous, we can avoid what's most of the time is an unnecessary `Promise.all` invocation.
While this does work pretty well in my quick testing, it's *very much* a hack since as far as I can tell there's no support in the CSS specification for using e.g. a CSS variable to override a `@media (prefers-color-scheme: dark) {...}` block.
The solution implemented here is thus to *edit* the viewer CSS, by either removing the entire `@media ...` block in light-mode or by ensuring that its rules become *unconditionally* applied in dark-mode.
To simplify the overall implementation, since all of this does seem like somewhat of an edge-case, the `viewerCssTheme` preference will *only* be read during viewer initialization. (Similar to many other existing preferences, a reload is thus required when changing it.)
Originally the default preferences were defined in a JSON-file checked into the repository, which was loaded using SystemJS in development mode.
Over the years a number of changes have been made to this code, most notably:
- The preferences JSON-file is now generated automatically, during building, from the `AppOptions` abstraction.
- All SystemJS usage has been removed from the development viewer.
Hence the default preferences are now available *synchronously* even in the development viewer, and it's thus no longer necessary to defer to the microtask queue (since `getDefaultPreferences` is async) just to get the default preferences.
While the effect of these changes is quite small, it *does* reduces the time it takes for the preferences to be fully initialized. Given the amount of asynchronous code during viewer initialization, every bit of time that we can save should thus help.
- Add support for logical assignment operators, i.e. `&&=`, `||=`, and `??=`, with a Babel-plugin. Given that these required incrementing the ECMAScript version in the ESLint and Acorn configurations, and that platform/browser support is still fairly limited, always transpiling them seems appropriate for now.
- Cache the `hasJSActions` promise in the API, similar to the existing `getAnnotations` caching. With this implemented, the lookup should now be cheap enough that it can be called unconditionally in the viewer.
- Slightly improve cleanup of resources when destroying the `WorkerTransport`.
- Remove the `annotationStorage`-property from the `PDFPageView` constructor, since it's not necessary and also brings it more inline with the `BaseViewer`.
- Update the `BaseViewer.createAnnotationLayerBuilder` method to actaually agree with the `IPDFAnnotationLayerFactory` interface.[1]
- Slightly tweak a couple of JSDoc comments.
---
[1] We probably ought to re-factor both the `IPDFTextLayerFactory` and `IPDFAnnotationLayerFactory` interfaces to take parameter objects instead, since especially the `IPDFAnnotationLayerFactory` one is becoming quite unwieldy. Given that that would likely be a breaking change for any custom viewer-components implementation, this probably requires careful deprecation.
*Note that I wasn't able to reproduce the issue in Firefox, but only in Chromium-browsers.*
The bug, and it's feels almost trivial once you've found it, is that we're not passing the `transform` parameter as intended to `PDFPageProxy.render` when drawing thumbnails on HiDPI displays. Instead the canvas context is, for reasons that I don't even pretent to understand, *manually* scaled in `PDFThumbnailView._getPageDrawContext`, which thus doesn't guarantee that the `baseTransform` property on the `CanvasGraphics`-instances becomes correct.
The solution is really simple though, just handle the `transform` the same way in `PDFThumbnailView.draw` as in `PDFPageView.paintOnCanvas` and things should just work.
*This is a pre-existing issue that I noticed while working on PR 12613, and fixing this also brings the thumbnail code inline with the page code.*
Given the intermittent nature of all of this, it's somewhat difficult to reproduce it consistently; however the following steps should at least provide an outline:
1. Open the sidebar, and the thumbnailView, and start scrolling around.
2. *Quickly* close the sidebar, so that all thumbnails won't have time to finish rendering.
3. Either wait for the cleanup-timeout to occur, or simply run `PDFViewerApplication.cleanup()` in the console.
What *intermittently* happens here is that `WorkerTransport.startCleanup` rejects, and consequently that cleanup doesn't complete as intended, since some of the thumbnails are left in a *pending* renderingState[1].
Fixing this is simple though, and only requires updating `PDFThumbnailViewer.cleanup` along the lines of `BaseViewer.cleanup`.
---
[1] Keep in mind that thumbnails will *only* render when the thumbnailView is visible, to reduce resource usage.
This patch will help reduce memory usage, especially for longer documents, when the user scrolls around in the thumbnailView (in the sidebar).
Note how the `PDFPageProxy.cleanup` method will, assuming it's safe to do so, release main-thread resources associated with the page. These include things such as e.g. image data (which can be arbitrarily large), and also the operatorList (which can also be quite large).
Hence when pages are evicted from the `PDFPageViewBuffer`, on the `BaseViewer`-instance, the `PDFPageView.destroy` method is invoked which will (among other things) call `PDFPageProxy.cleanup` in the API.
However, looking at the `PDFThumbnailViewer`/`PDFThumbnailView` classes you'll notice that there's no attempt to ever call `PDFPageProxy.cleanup`, which implies that in certain circumstances we'll essentially keep all resources allocated permanently on the `PDFPageProxy`-instances in the API.
In particular, this happens when the users opens the sidebar and starts scrolling around in the thumbnails. Generally speaking you obviously need to keep all thumbnail *images* around, since otherwise the thumbnailView is useless, but there's still room for improvement here.
Please note that the case where a *rendered page* is used to create the thumbnail is (obviously) completely unaffected by the issues described above, and this rather only applies to thumbnails being explicitly rendered by the `PDFThumbnailView.draw` method.
For the latter case, we can fix these issues simply by calling `PDFPageProxy.cleanup` once rendering has finished. To prevent *accidentally* pulling the rug out from under `PDFPageViewBuffer` in the viewer, which expects data to be available, this required adding a couple of new methods[1] to enable checking that it's indeed safe to call `PDFPageProxy.cleanup` from the `PDFThumbnailView.draw` method.
It's really quite fascinating that no one has noticed this issue before, since it's been around since basically "forever".
---
[1] While it should be *very* rare for `PDFThumbnailView.draw` to be called for a pageView that's also in the `PDFPageViewBuffer`, given that pages are rendered before thumbnails and that the *rendered page* is used to create the thumbnail, it can still happen since rendering is asynchronous.
Furthermore, it's also possible for `PDFThumbnailView.setImage` to be disabled, in which case checking the `PDFPageViewBuffer` for active pageViews *really* matters.
* remove 1st param of _createPopup (almost useless for a method)
* prepend popup div to avoid to have them on top of some highlights (and so "disable" partially mouse events)
* add a ref test for issue #12504
* When no actions then set it to null instead of empty object
* Even if a field has no actions, it needs to listen to events from the sandbox in order to be updated if an action changes something in it.
By using optional chaining, see https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Operators/Optional_chaining, it's possible to reduce unnecessary code-repetition in many cases.
Note that these changes also reduce the size of the *built* `pdf.js` file, when `SKIP_BABEL == true` is set, and for the `MOZCENTRAL` build-target that result in a `0.1%` filesize reduction from a simple and mostly mechanical code change.
Note that a number of these cases are covered by existing unit-tests, and a few others only matter for the development/build scripts.
Furthermore, I've also tried to the best of my ability to test each case *manually* to hopefully further reduce the likelihood of this patch introducing any bugs.
Please find additional details about the ESLint rule at https://eslint.org/docs/rules/no-useless-escape
The vast majority of the time, unless a Pattern is active, the `strokeColor`-property contains a "simple" colour value represented by a String. Hence it seems somewhat ridiculous to do a `hasOwnProperty` check on a String, and it's should thus be possible to improve things a tiny bit here.
Unfortunately using a simple `instanceof` check would only work for `TilingPattern`s, but not for the `ShadingIRs` given how they are implemented; see `src/display/pattern_helper.js`. (While that file could probably do with some clean-up, given the age of some of its code, that probably shouldn't happen here.)
Finally, the `this.type = "Pattern"`-property of the various Shadings/TilingPatterns were removed, since I cannot see why it's necessary when we can simply check for a `getPattern` method instead. Note that part of this code even pre-dates the main/worker-thread split, which probably in part explains why it looks the way it does.
The `PartialEvaluator.hasBlendModes` method is necessary to determine if there's any blend modes on a page, which unfortunately requires *synchronous* parsing of the /Resources of each page before its rendering can start (see the "StartRenderPage"-message).
In practice it's not uncommon for certain /Resources-entries to be found on more than one page (referenced via the XRef-table), which thus leads to unnecessary re-fetching/re-parsing of data in `PartialEvaluator.hasBlendModes`.
To improve performance, especially in pathological cases, we can cache /Resources-entries when it's absolutely clear that they do not contain *any* blend modes at all[1]. This way, subsequent `PartialEvaluator.hasBlendModes` calls can be made significantly more efficient.
This patch was tested using the PDF file from issue 6961, i.e. https://github.com/mozilla/pdf.js/files/121712/test.pdf:
```
[
{ "id": "issue6961",
"file": "../web/pdfs/issue6961.pdf",
"md5": "a80e4357a8fda758d96c2c76f2980b03",
"rounds": 100,
"type": "eq"
}
]
```
which gave the following results when comparing this patch against the `master` branch:
```
-- Grouped By browser, page, stat --
browser | page | stat | Count | Baseline(ms) | Current(ms) | +/- | % | Result(P<.05)
------- | ---- | ------------ | ----- | ------------ | ----------- | ---- | ------ | -------------
firefox | 0 | Overall | 100 | 1034 | 555 | -480 | -46.39 | faster
firefox | 0 | Page Request | 100 | 489 | 7 | -482 | -98.67 | faster
firefox | 0 | Rendering | 100 | 545 | 548 | 2 | 0.45 |
firefox | 1 | Overall | 100 | 912 | 428 | -484 | -53.06 | faster
firefox | 1 | Page Request | 100 | 487 | 1 | -486 | -99.77 | faster
firefox | 1 | Rendering | 100 | 425 | 427 | 2 | 0.51 |
```
---
[1] In the case where blend modes *are* found, it becomes a lot more difficult to know if it's generally safe to skip /Resources-entries. Hence we don't cache anything in that case, however note that most document/pages do not utilize blend modes anyway.
Given the number of parameters, and the fact that many of them are booleans, the call-sites are no longer particularly easy to read and understand. Furthermore, this slightly improves the formatting of the JSDoc-comment, since it needed updating as part of these changes anyway.
Finally, this removes an unnecessary `numViews === 0` check from `getVisibleElements`, since that should be *very* rare and more importantly that the `binarySearchFirstItem` function already has a fast-path for that particular case.
