Fixes issue #16843.
In certain cases, the text layer was misaligned
due to a difference between the `lang` attribute
of the viewer and the canvas. This commit addresses
the problem by adding the `lang` attribute to the canvas.
The issue was caused because PDF.js uses serif/sans-serif
fonts to generate the text layer and relies on system fonts.
The difference in the `lang` attribute led to different fonts
being picked, causing the misalignment.
This also required changing the initial `charCodeToGlyphId`-data to an Object, which seems generally correct since it's consistent with existing code in the `src\core\{cff_font, type1_font}.js` files.
The `through2` dependency got introduced over four years ago in #11325 to
replace the unmaintained `gulp-transform` dependency. However, sadly the
same holds for `through2` since the last release was also four years ago.
Fortunately the `through2` dependency can trivially be replaced with the
built-in Node.js `stream.Transform` API nowadays. In fact, the `through2`
dependency mentions themselves in their README already that they are "a
tiny wrapper around Node.js streams.Transform". The `stream.Transform`
API is available in all Node.js versions we support, and in Node.js 6
already the simplified constructor approach for `stream.Transform` got
introduced to simplify creating custom stream transformers; see
https://nodejs.org/docs/latest-v6.x/api/stream.html#stream_new_stream_transform_options.
This commit therefore replaces `through2` by switching to the
`stream.Transform` API directly so we don't need any wrappers anymore.
Note that for our case the only change we have to make is to enable
object mode, see https://nodejs.org/api/stream.html#object-mode, because
we pass in `VinylFile` objects instead of e.g. regular `Buffer` objects.
I have confirmed in two ways that this is indeed a drop-in replacement:
- Running the Gulp targets that call the `transform` function and
diffing the resulting `build` folder before/after this patch, with
`diff -r build-old/ build-new/`, to ensure that there are no
unexpected changes in the output.
- Changing the Gulpfile to, instead of UTF-8, transform the files to
ASCII, and diffing the resulting `build` folder to confirm that the
transformation logic works and produces different results, such as:
```
diff build/lib/core/standard_fonts.js build-ascii/lib/core/standard_fonts.js
284c284
< t["Trinité"] = true;
---
> t["Trinit�"] = true;
```
This is very old code, and predates e.g. the introduction of JavaScript classes, which creates unnecessarily unwieldy code in the viewer.
By introducing a new `TextLayer` class in the API, similar to how e.g. the `AnnotationLayer` looks, we're able to keep most parameters on the class-instance itself. This removes the need to manually track them in the viewer, and simplifies the call-sites.
This also removes the `numTextDivs` parameter from the "textlayerrendered" event, since that's only added to support default-viewer functionality that no longer exists.
Finally we try, as far as possible, to polyfill the old `renderTextLayer` and `updateTextLayer` functions since they are exposed in the library API.
For *simple* invocations of `renderTextLayer` the behaviour should thus be the same, with only a warning printed in the console.
*Please note:* This doesn't really affect the viewer, but may affect the library API if multiple PDF documents are opened in parallel.
Since we clean-up "global" textLayer-data when destroying a PDF document, this means that other active PDFs could potentially break by invoking `cleanupTextLayer` unconditionally. Note that textLayer rendering is an asynchronous task, and we thus need to ensure those are all finished before running clean-up.
The `needle` dependency originally got introduced in #12024, almost four
years ago, to be able to use pre-built binaries for the `canvas`
dependency on macOS. However, nowadays the `needle` dependency isn't
used by `canvas` anymore, or any other package we use for that matter,
as shown by the empty NPM dependency tree:
```
$ npm ls needle
pdf.js
└── needle@3.3.1
```
Investigation showed that the `canvas` package depends on the
`node-pre-gyp` package which in turn depended on `needle` (see
https://github.com/Automattic/node-canvas/issues/1110#issuecomment-411232630),
but in version 1.0.0 of `node-pre-gyp` from three years ago the `needle`
dependency got dropped in favor of `node-fetch` (see
a74f5e367c/CHANGELOG.md (L52)).
This explains why the NPM dependency tree is empty now and proves that
we can safely get rid of this dependency now.
In Node.js 14.14.0 the `fs.rmSync` function was added that removes files
and directories. The `recursive` option is used to remove directories
and their contents, making it a drop-in replacement for the `rimraf`
dependency we use.
Given that PDF.js now requires Node.js 18+ we can be sure that this
option is available, so we can safely remove `rimraf` and reduce the
number of project dependencies.
Co-authored-by: Wojciech Maj <kontakt@wojtekmaj.pl>
I broke this accidentally in PR 18089, sorry about that!
Note that since `#processItems` is private we can no longer just "replace" the method as was done in PR 18052.
The issue from #18003 hasn't been shown to be caused by PDF.js, but it
did surface that we don't have (direct) unit test coverage for the
`BaseException` class. This made it more difficult to prove that the
`stack` property was already available on exception instances, but more
importantly it caused the CI to be green even though the suggested
change would have caused the `stack` property to disappear.
To avoid future regressions, for e.g. similar changes or a rewrite from
a closure to a proper class, this commit introduces a dedicated unit
test for `BaseException` that asserts that our exception instances
indeed expose all expected properties.
The Puppeteer update should in particular be helpful for us because it
contains improved WebDriver BiDi compatibility, a newer Chrome version
(both might help for #17962) and an official deprecation of CDP for
Firefox. Note that the latter doesn't require changes on our end because
we already use WebDriver BiDi unconditionally for Firefox since commit
4db0174. The full release notes can be found at
https://github.com/puppeteer/puppeteer/releases/tag/puppeteer-core-v22.8.0.
When seleciting on a touch screen device, whenever the finger moves to a
blank area (so over `div.textLayer` directly rather than on a `<span>`),
the selection jumps to include all the text between the beginning of the
.textLayer and the selection side that is not being moved.
The existing selection flickering fix when using the mouse cannot be
trivially re-used on mobile, because when modifying a selection on
a touchscreen device Firefox will not emit any pointer event (and
Chrome will emit them inconsistently). Instead, we have to listen to the
'selectionchange' event.
The fix is different in Firefox and Chrome:
- on Firefox, we have to make sure that, when modifying the selection,
hovering on blank areas will hover on the .endOfContent element
rather than on the .textLayer element. This is done by adjusting the
z-indexes so that .endOfContent is above .textLayer.
- on Chrome, hovering on blank areas needs to trigger hovering on an
element that is either immediately after (or immediately before,
depending on which side of the selection the user is moving) the
currently selected text. This is done by moving the .endOfContent
element around between the correct `<span>`s in the text layer.
The new anti-flickering code is also used when selecting using a mouse:
the improvement in Firefox is only observable on multi-page selection,
while in Chrome it also affects selection within a single page.
After this commit, the `z-index`es inside .textLayer are as follows:
- .endOfContent has `z-index: 0`
- everything else has `z-index: 1`
- except for .markedContent, which have `z-index: 0`
and their contents have `z-index: 1`.
`.textLayer` has an explicit `z-index: 0` to introduce a new stacking context,
so that its contents are not drawn on top of `.annotationLayer`.
- Change all possible semi-private methods into properly private ones. Note that this code is old enough to predate standard classes.
- Move the `appendText` helper function into `TextLayerRenderTask`, as a private method, to avoid having to manually pass in the scope.
- Simplify `#layoutText` by directly passing in all necessary data. This is possible after the changes PR 18052.
For images that failed to decode once we want to avoid a pointless round-trip to the main-thread, which could otherwise happen for globally cached images.
- These changes will allow a simpler way of implementing PR 17770.
- The /Lang attribute is fetched lazily, with the first `getTextContent` invocation. Given the existing worker-thread caching, this will thus only need to be done *once* per PDF document (and most PDFs don't included this data).
- This makes the /Lang attribute *directly available* in the `textLayer`, which has the following advantages:
- We don't need to block, and thus delay, overall viewer initialization on fetching it (nor pass it around throughout the viewer).
- Third-party users of the `textLayer` will automatically benefit from this, once we start actually using the /Lang attribute in PR 17770.
*Please note:* This also, importantly, means that the `text` reference-tests will then cover this code (which wouldn't otherwise have been the case).
