Comments on: Overview of pdf.js guts

By: po_chuang

po_chuang — Mon, 09 Jan 2012 07:35:58 +0000

This idea is gorgeous!

By: cjones

cjones — Thu, 24 Nov 2011 00:45:14 +0000

abhilash: There’s an issue on file in the pdf.js github tracking the SVG backend. @notmasteryet made a prototype.

shobo: HTTP byte-range requests.

By: shobo

shobo — Wed, 23 Nov 2011 14:39:53 +0000

How can a client-side browser component only request a specific page from a server-side PDF without some server-side component to extract it individually? Can please somebody answer this question? I would really appreciate it! Thank you!

By: abhilash

abhilash — Thu, 15 Sep 2011 06:00:04 +0000

When do you think will start working on the SVG version. I would like to contribute to the code once basic text selection is possible.

By: Leif Halvard Silli

Leif Halvard Silli — Sat, 16 Jul 2011 07:29:03 +0000

ACCHSH: http://www.puzzleflow.com/company

By: cjones

cjones — Sun, 10 Jul 2011 21:16:24 +0000

Bashev: Pdf.js should support Cyrillic documents (it doesn’t really know what alphabet it’s using to render). Have you tried http://andreasgal.github.com/pdf.js/multi_page_viewer.html ? If that’s still broken, can you let us know which PDF file isn’t working for you? Pdf.js still has lots of bugs .

By: Bashev

Bashev — Sun, 10 Jul 2011 15:15:15 +0000

This version does not support Cyrillic documents.

May be you fixed this in some of next releases?

By: cjones

cjones — Mon, 27 Jun 2011 16:41:08 +0000

Baz: That is incredibly helpful, thank you. I’ve linked your notes from our wiki. Of course, the code is at https://github.com/andreasgal/pdf.js, if you ever have some free time …

By: ThomasW

ThomasW — Mon, 27 Jun 2011 15:56:33 +0000

O.K., then my comment was a bit short-sighted. I thought that this complex shaping might be handled in form of ligatures, but if it can’t, then I’m remaining silent…

By: Baz

Baz — Mon, 27 Jun 2011 12:54:14 +0000

Re the ‘big project’ for text selection: I did some work on fixing this in poppler, which had taken the approach mentioned (identifying & chaining textruns), but it works *really* badly for multicolumn text.

What I did to fix this was to use the reading-order sort algorithm used in Ocropus [T Breuel: High performance document layout analysis. Symposium on Document Image Understanding Technology, pp.209-218 (2003)]. This is a massive improvement, but poppler still has issues because its paragraph detection is primitive; implementing the gutter detection part of the same paper would help with this.

Adobe’s solution is tagged pdfs, which contain all the info to determine reading order, but trying to deal with tagged pdfs in poppler’s existing data structures is messy. It would have been better to deal with tagged PDFs first, then apply the reading-order heuristics above to infer a tagging structure for ‘normal’ pdfs.

Another problem is that poppler suffers from a lack of an accessible API from the outset. Text and selections work by looking at text within a bounding box, rather than finding a start character and end character and selecting all characters in between. (like eg ATK.Text)

It’d be nice if pdf.js, with its clean slate, can avoid these problems