[
https://issues.apache.org/jira/browse/TIKA-4898?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18116416#comment-18116416
]
Tim Allison edited comment on TIKA-4898 at 9/17/26 4:48 PM:
------------------------------------------------------------
:robot: suggests that this information is tracked and then nulled out in PDFBox?
{noformat}
It looks like COSParser already has it: it walks the /Prev chain and records
every section in
XrefTrailerResolver, and getTrailerCount() is public — but the resolver is
reset and nulled when
parsing finishes, so nothing reaches COSDocument. Linearization would still
need special-casing
(the first-page xref's /Prev points at the main one), and a chain-based count
would miss
appended bytes that no xref references, which for our forensics use case is
the interesting case.
So probably both, not one instead of the other.
{noformat}
was (Author: [email protected]):
:robot: suggests that this information is tracked and then nulled out in PDFBox?
{noformat}
It looks like COSParser already has it: it walks the /Prev chain and records
every section in XrefTrailerResolver, and getTrailerCount() is public — but the
resolver is reset and nulled when parsing finishes, so nothing reaches
COSDocument. Linearization would still need special-casing (the first-page
xref's /Prev points at the main one), and a chain-based count would miss
appended bytes that no xref references, which for our forensics use case is the
interesting case. So probably both, not one instead of the other.
{noformat}
> Improve xref scanning efficiency
> --------------------------------
>
> Key: TIKA-4898
> URL: https://issues.apache.org/jira/browse/TIKA-4898
> Project: Tika
> Issue Type: Task
> Reporter: Tim Allison
> Priority: Minor
> Attachments: PDFJS-6078-screen-annotations.pdf
>
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
