[avanish-garg]

Issue resolved by this Pull Request:
Resolves #4442

Splits each paragraph into (text, formatting, hyperlink) tuples per run, mirroring
msword_backend.py's existing _iter_paragraph_content/_get_format_from_run/
_get_hyperlink_target pattern (python-pptx's public Font API only exposes bold/italic/underline,
unlike python-docx's richer run API, so strikethrough/sub/superscript aren't in scope here):

• Adjacent runs sharing identical formatting/hyperlink are coalesced back into one unit first,
  since PowerPoint sometimes splits a single word across runs with no actual formatting difference
  (e.g. spell-check markup) — without this, wrapping each such run in its own item would let the
  markdown serializer's inter-sibling space (InlineGroup always joins children with " ") split
  real words apart.
• A paragraph with more than one run is wrapped in an inline group, mirroring the DOCX backend's
  own _create_or_reuse_parent, so a sentence with a bold word or a link in the middle stays one
  flowing unit on export instead of splintering into separate markdown blocks.
• Title/section-header paragraphs are kept as a single item exactly as before — a heading is one
  outline-level unit, not a sequence of independently formatted spans.
• Each item's own provenance is now built from its own text rather than the whole paragraph's, so
  charspan stays correct once a paragraph can produce more than one item (python-pptx exposes no
  finer bbox than the shape's, so the bbox itself stays shape-level for every item in it, same as
  before).

Changes: docling/backend/mspowerpoint_backend.py gets three new helpers
(_get_run_format/_get_run_hyperlink_target/_iter_paragraph_runs) plus a
_add_list_item_with_formatting helper for the list case (mirroring msword_backend.py's
_add_formatted_list_item), and _handle_text_elements is rewired to use them.
tests/test_backend_pptx.py gets a new regression test,
test_pptx_preserves_run_level_formatting_and_hyperlinks, covering a bold run, a plain run, and a
hyperlinked run in one paragraph — verified to fail on unmodified main (all three collapse into
one plain-text item) and pass with the fix.

Golden fixtures regenerated (DOCLING_GEN_TEST_DATA=1). Every diff was reviewed by hand and is
exactly the fixed behavior: a real markdown hyperlink where the URL used to be silently dropped
(powerpoint_comments.pptx.md), a real bold marker where formatting used to be silently dropped
(powerpoint_issue_2663.pptx.md, **Key Benefits:**), and an explicit
"formatting": {"bold": false, ...} block now present on plain-text items where the field was
previously just absent — no text content changed anywhere, no unrelated diffs. (One golden JSON's
embedded base64 image bytes also changed on first regeneration; confirmed via a clean regen on
unmodified main that this is a pre-existing Pillow/PNG-encoder version difference in this
environment, unrelated to this fix, and reverted just that field back to its original value before
committing.)

Full tests/test_backend_pptx.py suite: 30 passed, 2 skipped (pre-existing, LibreOffice-gated), no
regressions. ruff check/ruff format --check: clean. ty check: same 7 pre-existing warnings as
unmodified main (confirmed via diff), no new ones, none on the changed lines.

Checklist:

• Documentation has been updated, if necessary. — n/a, no user-facing docs describe PPTX
  run-formatting handling to update
• Examples have been added, if necessary. — n/a
• Tests have been added, if necessary.

[github-actions]

✅ DCO Check Passed

Thanks @avanish-garg, all your commits are properly signed off. 🎉

[mergify]

Merge Protections

🔴 1 of 2 protections blocking · waiting on 👀 reviews

	Protection	Waiting on
🔴	Require two reviewer for test updates	👀 reviews
🟢	Enforce conventional commit	—

🔴 Require two reviewer for test updates

Waiting for

• #approved-reviews-by >= 2

This rule is failing.

When test data is updated, we require two reviewers

• #approved-reviews-by >= 2

---

Show 1 satisfied protection

🟢 Enforce conventional commit

Make sure that we follow https://www.conventionalcommits.org/en/v1.0.0/

• title ~= ^(fix|feat|docs|style|refactor|perf|test|build|ci|chore|revert)(?:\(.+\))?(!)?:

[codecov]

Codecov Report

❌ Patch coverage is 90.00000% with 6 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
docling/backend/mspowerpoint_backend.py	90.00%	4 Missing and 2 partials ⚠️

📢 Thoughts on this report? Let us know!

[ceberam]

Thanks @avanish-garg for this PR, since indeed the PPTX backend was not yet keeping the formatting unlike the other Office backends.
The new code and tests are technically sound and functionally correct.

My only recommendation would be to clean up the excessive inline commentary and verbose private docstrings.
There are multiple lengthy inline comment blocks that explain why previous behavior was buggy, detail developer rationale, or repeat method-level documentation. The inline comments in Docling should also limit explanations on how third-party applications like PowerPoint work, since developers and maintainers can refer to the documentation of those applications directly, if necessary.

[avanish-garg]

Thanks for the review @ceberam. Done with the changes!

[ceberam]

Thanks @avanish-garg for addressing my comment.
All good from my side.

[dolfim-ibm]

@avanish-garg do you mind having a last round fixing the conflict with main?

[github-actions]

AI triage (advisory)

Checked commit 450cdaccee. The analysis read the diff only. It did not run code from this pull request.

This pull request preserves run-level character formatting and hyperlinks in the PowerPoint backend. It wraps multi-run paragraphs into inline groups.

CI safety: no concern found. A maintainer must still approve CI.

1 related pull request(s) in the same code

PR	State	Signal	Assessment
#4787	merged	also references issue #4442	related: PR #4787 handles hyperlinks on entire shapes, while this pull request handles formatting and hyperlinks on individual text runs.

Reference data changes

Change	Files
structure	1
text	8

• structure tests/data/pptx/groundtruth/powerpoint_comments.pptx.json: paragraph: 2 → 4
• text tests/data/pptx/groundtruth/powerpoint_comments.pptx.itxt: +10 / -7 lines
• text tests/data/pptx/groundtruth/powerpoint_comments.pptx.md: +1 / -1 lines
• text tests/data/pptx/groundtruth/powerpoint_issue_2663.pptx.json: item attributes changed
• text tests/data/pptx/groundtruth/powerpoint_issue_2663.pptx.md: +2 / -2 lines
• text tests/data/pptx/groundtruth/powerpoint_malformed_pictures.pptx.json: item attributes changed
• text tests/data/pptx/groundtruth/powerpoint_sample.pptx.json: item attributes changed
• text tests/data/pptx/groundtruth/powerpoint_unrecognized_shape.pptx.json: item attributes changed
• text tests/data/pptx/groundtruth/powerpoint_with_image.pptx.json: item attributes changed

Assessment: expected. The reference data changes match the new paragraph run splitting and inline group structure in PowerPoint documents.

_{The ai:* labels are advisory. They do not approve CI and they do not add tests:full. A new push runs this check again.}

[ceberam]

@avanish-garg I've just seen a blocking issue and added a couple of suggestions worth addressing.

[ceberam]

_get_run_hyperlink_target does not guard against KeyError from a dangling relationship. run.hyperlink.address calls part.target_ref(rId) internally, which raises KeyError if the relationship ID no longer exists in the file, a condition that occurs with sanitized or re-saved PPTX files. The DOCX backend's counterpart explicitly catches this. Without the guard, a single malformed hyperlink crashes the entire document conversion.

Fix:

Suggested change

	address = run.hyperlink.address
	try:
	address = run.hyperlink.address
	except KeyError:
	return None

[ceberam]

• Minor performance comment: _iter_paragraph_runs is called twice per shape: once in the early-return guard (result discarded) and once inside _handle_text_elements. The old _paragraph_fragments was cheap; the new method does font inspection and hyperlink resolution per run.
• Add type annotation to paragraph

[PeterStaar-IBM]

@mittalpk please fix the linting

[mittalpk]

@PeterStaar-IBM I think this one is for @avanish-garg, since #4443 is his branch; my #4787 only references it. The lint failure is the max-lines hook: tests/test_backend_pptx.py is at 1547 lines. Moving the new tests into their own file fixes it, as I did for the same failure in #4733.

[avanish-garg]

Done with lint changes
THanks @PeterStaar-IBM @mittalpk

[avanish-garg]

Any update on this @ceberam ?

[ceberam]

Thanks @avanish-garg ! Approved.
Just a non-blocking message though below if you can check it out.

[ceberam]

What are these inline comments referring to? Or they should just be removed?