Previously a couple of different attempts at fixing this problem has been rejected, given how *crucial* this code is for the correct function of the viewer, since no one has thus far provided any evidence that the problem actually affects the default viewer[1] nor an example using the viewer components directly (without another library on top).
The fact that none of the prior patches contained even a *simple* unit-test probably contributed to the unwillingness of a reviewer to sign off on the suggested changes.
However, it turns out that it's possible to create a reduced test-case, using the default viewer, that demonstrates the error[2]. Since this utilizes a hidden `<iframe>`, please note that this error will thus affect Firefox as well.
Note that while errors are thrown when the hidden `<iframe>` loads, the default viewer doesn't break completely since rendering does start working once the `<iframe>` becomes visible (although the errors do break the initial Toolbar state).
Before making any changes here, I carefully read through not just the immediately relevant code but also the rendering code in the viewer (given it's dependence on `getVisibleElements`). After concluding that the changes should be safe in general, the default viewer was tested without any issues found. (The above being much easier with significant prior experience of working with the viewer code.)
Finally the patch also adds new unit-tests, one of which explicitly triggers the relevant code-path and will thus fail with the current `master` branch.
This patch also makes `PDFViewerApplication` slightly more robust against errors during document opening, to ensure that viewer/document initialization always completes as expected.
Please keep in mind that even though this patch prevents an error in `getVisibleElements`, it's still not possible to set the initial position/zoom level/sidebar view etc. when the viewer is hidden since rendering and scrolling is completely dependent[3] on being able to actually access the DOM elements.
---
[1] And hence the PDF Viewer that's built-in to Firefox.
[2] Copy the HTML code below and save it as `iframe.html`, and place the file in the `web/` folder. Then start the server, with `gulp server`, and navigate to http://localhost:8888/web/iframe.html
```html
<!DOCTYPE html>
<html>
<head>
<title>Iframe test</title>
<script>
window.onload = function() {
const button = document.getElementById('button1');
const frame = document.getElementById('frame1');
button.addEventListener('click', function(evt) {
frame.hidden = !frame.hidden;
});
};
</script>
</head>
<body>
<button id="button1">Toggle iframe</button>
<br>
<iframe id="frame1" width="800" height="600" src="http://localhost:8888/web/viewer.html" hidden="true"></iframe>
</body>
</html>
```
[3] This is an old, pre-exisiting, issue that's not relevant to this patch as such (and it's already being tracked elsewhere).
With modern JavaScript supporting block-scoped variables, it's no longer necessary to have large numbers of top-level variable definitions (especially when all variables are, implicitly, set to `undefined`).
Besides moving variable definintions, a number of them are also converted to `const` to help ensure that they cannot be *accidentally* modified in the code.
Finally, the `views.length` calculation is now cached *once* rather than being re-computed multiple times.
Handle more cases of corrupt PDF files with missing 'endobj' operators, where the "obj" string is immediately followed by the dictionary (PR 9288 follow-up)
This is *really* the best that we can do here, since other proposed solutions would interfere with (and break) the painstakingly implemented browsing history that's present in the default viewer.
I'm still not convinced that this is a good idea in general, but this patch implements it in a way where it is possible to toggle[1] for users that wish to have this feature. In particular, there's a couple of reasons why I'm not finding this feature necessary/great:
- It's already possible to easily obtain the current hash, by simply clicking on the `viewBookmark` button at any time.
- Hash changes requires a bit of special handling[2], i.e. extra code, to prevent issues when the browser history is traversed (see `PDFHistory._popState`). Currently this is only necessary when the user has manually changed the hash, with this patch it will always be the case (assuming the feature is active).
- It's not always possible to change the URL when updating the browser history. For example: In the Firefox built-in viewer, the URL cannot be modified for local files (i.e. those using the `file://` protocol).
This leads to inconsistent behaviour, and may in some cases even result in errors being thrown and the history thus not updating, if the browser prevents changes to the URL during `pushState`/`replaceState` calls.
---
[1] Using the `historyUpdateUrl` viewer preference.
[2] This depends, to a great extent, on browsers always firing `popstate` events *before* `hashchange` events, which may or may not actually be guaranteed.
This should hopefully be sufficient to address issue 6847, and given the limited impact of the code changes I'm not completely sure if this would need to be controlled by a preference!?
Initially my intention was to try and provide some (slightly more detailed) implementation suggestions in the issue, but having looked briefly at doing that it would essentially have amounted to actually writing the code anyway. (Especially considering that the recent questions seemed to more-or-less ignore the information already provided in the first post.)
Finally, note that since `performance.navigation.type` is marked as deprecated, a slightly different approach was choosen instead.
The error was triggered for a particular set of metadata, where an end tag was encountered without the corresponding begin tag being present in the data.
(The patch also fixes a minor oversight, from a recent PR, in the `SimpleDOMNode.nextSibling` method.)
Given that the issue, as filed, is incomplete since no PDF file was provided for debugging, this patch is really the best that we can do here. *Please note:* This patch will *not* enable the Metadata to be successfully parsed, but it should at least prevent the errors.
This method creates quite a few intermediate strings on each call and
it's called often, even for smaller documents like the Tracemonkey
document. Scrolling from top to bottom in that document resulted in
14126 strings being created in this method. With this commit applied,
this is reduced to 2018 strings.
This method creates quite a few intermediate strings on each call and
it's called often, even for smaller documents like the Tracemonkey
document. Scrolling from top to bottom in that document resulted in
12936 strings being created in this method. With this commit applied,
this is reduced to 3610 strings.
The `toString` method always creates two string objects (for the 'R'
character and for the `num` concatenation) and in the worst case
creates three string objects (one more for the `gen` concatenation).
For the Tracemonkey paper alone, this resulted in 12000 string
objects when scrolling from the top to the bottom of the document.
Since this is a hot function, it's worth minimizing the number of string
objects, especially for large documents, to reduce peak memory usage.
This commit refactors the `toString` method to always create only one
string object.
