Error in reflowing of lines from PDFs

[Seb Bacon]

Given the attached test PDF, when running with the pdftotext engine:

    ebook-convert example.pdf output.epub --pdf-engine=pdftotext

I get the following, as expected:

    Patterned chiffon," she sang it off. "You must love to craft how to
    galvanise the right smell.”

But when I use the calibre engine:

    ebook-convert example.pdf output.epub --pdf-engine=calibre

I get:

    right Patternedchiffon," shesang it off. "You mustlove to craft howto
galvanise the smell.”

[Kovid Goyal (kovid)]

Changing the component for this bug.

 assignee
 status

[Alan Pettigrew (linux-k)]

That is a tricky one to decode.
I see:
right Patterned chiffon,” she sang it off. “You must love to craft how to galvanise the smell.”
i.e. the spaces between words are there.

We have:
<text top="822" left="615" width="71" height="25" font="0">galvanise</text>
<text top="827" left="686" width="5" height="16" font="1"> </text>
<text top="822" left="690" width="24" height="25" font="0">the</text>
<text top="843" left="108" width="37" height="25" font="0">right</text>
<text top="848" left="144" width="5" height="16" font="1"> </text>
<text top="843" left="149" width="53" height="25" font="0">smell.”</text>

So, top=822+height=25 gives the next line starting at 847.
And top=827+height=16 gives the next line starting at 843. It is part of the first line.
And top=843 is less than 847 so could be part of the first line (subscript).

It appears that the 2nd line overlaps the first, so is this 2 separate lines, or 1 oddly positioned one?

I don't know whether extending the test for 'where is the next top' to fix this will cause problems in the more general case. I will do some testing.

Isn't PDF formatting wonderful!

[Charles Haley (cbhaley)]

The superscripted spaces are "inside" their containing line. Their "top" is greater than the top of the nearby text and their "bottom" (top + height) is less than the nearby text. If a text line is defined by two borders, top and top + max(width), then it seems that any text that is inside those borders is part of that text, offset vertically.

I admit that I have no idea what else might fall between those borders.

And why are they using superscripted spaces in the first place?

[Kovid Goyal (kovid)]

Fixed in branch master. The fix will be in the next release. calibre is usually released every alternate Friday.