770409 - pdf containing vertical japanese text does not render correctly

Status: RESOLVED FIXED

(Firefox :: PDF Viewer, defect)

Description

[John Daggett (:jtd)]

Attachment: About the MVO of UTR50

Steps:

1. Open the attached pdf file

Result: file is never loaded, browser just shows the loading animation.

Original file: http://blogs.adobe.com/CCJKType/files/2012/07/TaroUTR50SortedList112.pdf

Comment 1

[John Daggett (:jtd)]

Confirmed using 2012-07-02 nightly.

Comment 2

[Alice0775 White]

Regression window(m-c)
Good(but No Japanese text):
http://hg.mozilla.org/mozilla-central/rev/305cd10b57d2
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:15.0) Gecko/15.0 Firefox/15.0a1 ID:20120601055003
Bad:
http://hg.mozilla.org/mozilla-central/rev/12ab69851e05
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:15.0) Gecko/15.0 Firefox/15.0a1 ID:20120601073103
Pushlog:

Regression window(m-i)
Good(but No Japanese text):
http://hg.mozilla.org/integration/mozilla-inbound/rev/63ed999a6fe9
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:15.0) Gecko/15.0 Firefox/15.0a1 ID:20120530222224
Bad:
http://hg.mozilla.org/integration/mozilla-inbound/rev/ff2ad97929e8
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:15.0) Gecko/15.0 Firefox/15.0a1 ID:20120531021523
Pushlog:
http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=63ed999a6fe9&tochange=ff2ad97929e8

Triggered by: Bug 748924

Comment 3

[Alex Keybl [:akeybl]]

Given bug 773397, tracking for FF16's release but not FF15. Also sending over to Artur since he landed the suspected regressing bug.

Comment 4

[Brendan Dahl [:bdahl]]

This file should be loading fine in the current nightly. However, the CJK font doesn't seem to be rendering correctly.

Comment 5

[Alex Keybl [:akeybl]]

It's not clear if this is a common problem, and PDF.js will ship in 17 or later, so untracking for release.

Comment 6

[John Daggett (:jtd)]

Mozilla/5.0 (Macintosh; Intel Mac OS X 10.8; rv:17.0) Gecko/17.0 Firefox/17.0
Build date: 2012-08-27

PDF now loads but vertical text elements display incorrectly (see attached screenshot).

Comment 7

[John Daggett (:jtd)]

Attachment: screenshot, pdf displayed with pdf.js vs. preview

Comment 8

[rshimazu]

The following urls are other examples of this bug, I think.

http://yama2so.com/hoki/jiinkyoukai.pdf
http://www.asahi-net.or.jp/~td6i-st/fuku-cathedral/new/gat.pdf
http://www.jcp.or.jp/web_download/seisaku/201204_teigen_panf.pdf#page=2
http://www.jsad.or.jp/paralympic/london/pdf/directory.pdf#page=2

Maybe if the PDF has not only vertical characters but also horizntal characters in the same document, it shows the problem.

Comment 9

[Brendan Dahl [:bdahl]]

Fixed in https://github.com/mozilla/pdf.js/pull/2686

Comment 10

[rshimazu]

I confirmed the fix with pdf.js 0.7.283 on Firefox 19(Vista).
Thank you.

Comment 11

[David Frank]

further confirm issue fixed in 0.7.355 with traditional chinese vertical text document sample:

http://www.kadokawa.com.tw/Upload/product_201212281735422.pdf

I can see japan and taiwan pdf layout use vertical text a lot (especially with long article like novel), would love to see it in stable Firefox sooner than later (Firefox 19.0.2 still have this issue.)

Comment 12

[Masatoshi Kimura [:emk]]

Fixed by bug 848641.
Unfortunately the fix is too big to backport to branches, IMO. So merging to Firefox will not be sooner than Firefox 22.
You can install the latest development version of PDF.js add-on if you can't wait: http://mozilla.github.com/pdf.js/extensions/firefox/pdf.js.xpi
Usually this is much newer than the Firefox internal version.

Comment 13

[Mitar]

This fix does not yet completely resolve the text layer positioning. Even on the test PDF above, text layer is nit position correctly:

http://blogs.adobe.com/CCJKType/files/2012/07/TaroUTR50SortedList112.pdf

Tested using http://mozilla.github.com/pdf.js/web/viewer.html at 79c3546c924501ba4c9852233cd4638414cdf5ea.

If you select all text or make text be displayed, you can see that latin text on the first page is not displayed correctly.

Comment 14

[Mitar]

Furthermore, it fails in same manner (hidden text layer) for PDFs from arXiv: http://arxiv.org/pdf/0707.3023.pdf

Comment 15

[Masatoshi Kimura [:emk]]

The latin text problem is a separate issue and tracked by <https://github.com/mozilla/pdf.js/issues/2817>.

Comment 16

[Mitza]

Verified as fixed on Firefox 23 Nightly(Build ID: 20130408030928) 
-the page loads correctly
-the text is correctly displayed	

Mozilla/5.0 (Macintosh; Intel Mac OS X 10.8; rv:23.0) Gecko/20130408 Firefox/23.0
Mozilla/5.0 (X11; Linux i686; rv:23.0) Gecko/20130408 Firefox/23.0
Mozilla/5.0 (Windows NT 6.2; rv:23.0) Gecko/20130408 Firefox/23.0