Tagged text extraction removes a word boundary within one marked-content section

[lin-string]

Summary

Page.extract_text_tagged() concatenates consecutive text objects within the same marked-content section without preserving a word boundary when the next object continues on a new visual line.

Versions tested

• playa-pdf 1.1.0 (v1.1.0, commit 85a9c1e22e327bf873e4282d66ed1dba89fa7575)
• Current main at commit 9496dfea5150343c9dc05544d9c004bbcccf17e6, checked on 2026-09-07

Test files

• UA1_Tpdf-G3_02.pdf, with expected behavior described in the PDF Association test documentation.

Reproduction

import playa

with playa.open("UA1_Tpdf-G3_02.pdf") as document:
    print(document.pages[0].extract_text_tagged())

Actual behavior

The sentence contains The blue backgroundis decorative: the end of one text object and the beginning of the next are joined without whitespace.

Expected behavior

The line transition should preserve the word boundary and return The blue background is decorative.

Relevant implementation

Page.extract_text_tagged() groups text objects by marked-content section and uses chars = "".join(c) inside each group. Its line-transition check only runs between different MCIDs, so it cannot insert a separator between text objects within the same marked-content section.

Disclosure

This issue was prepared by an AI coding agent on behalf of @lin-string. The reproduction and results were verified against the current default branch.