GitLab

I ran into a problem with pdfinfo -struct-text, which I think is caused by a bug.

The attached file demonstrates the problem. Extracting the textual content gives this result:

$ pdfinfo -struct-text poppler-bug.pdf
Document
  P (block)
    "xxxyyy"

where the expected output would be:

Document
  P (block)
    "xxxyyyzzz"

The relevant part of the attached pdf is the page content stream:

3 0 obj
<< /Length 215 >>        
stream
1 0 0 1 48.272 46.73 cm
/P <</MCID 0>> BDC 1 0 0 1 -48.272 -46.73 cm
BT
/F1 9.96264 Tf
1 0 0 1 48.272 46.73 Tm [(xxx)]TJ
/Span << >> BDC
1 0 0 1 64.046 46.73 Tm [(yyy)]TJ
EMC
1 0 0 1 79.82 46.73 Tm [(zzz)]TJ
ET
EMC

endstream

As you see, the problem is caused by a nonstructural marked content sequence (the /Span) inside a marked content item (marked with /P). This is explicitly allowed by the specification (see §14.7.4.1, p.560), yet somehow cuts short pdfinfo’s text extraction function.

This is on debian unstable, with the following version:

$ pdfinfo -v
pdfinfo version 20.09.0
Copyright 2005-2020 The Poppler Developers - http://poppler.freedesktop.org
Copyright 1996-2011 Glyph & Cog, LLC