fix(comments): order replyLst before txBody per [MS-PPTX] CT_Comment — replies now render (issue #25)

Follow-up to ca4da3f. After the four parent-binding fixes, PowerPoint
rendered parent comment threads but SILENTLY DROPPED every reply
(`interceptor macos find "Merging"` → 0; only empty Reply boxes). The
advisor (Rule 2) correctly flagged reply-rendering as a logically
independent concern none of the four parent fixes touched.

Root cause is authoritative, not inference: the [MS-PPTX] CT_Comment
schema (learn.microsoft.com/openspecs MS-PPTX, 2018/8/main) defines

  <xsd:complexType name="CT_Comment"><xsd:sequence>
    <xsd:group ref="EG_CommentAnchor"/>                <!-- pc:sldMkLst -->
    <xsd:element name="pos" minOccurs="0"/>
    <xsd:element name="replyLst" type="CT_CommentReplyList" minOccurs="0"/>
    <xsd:group ref="EG_CommentProperties"/>            <!-- txBody, extLst -->
  </xsd:sequence></xsd:complexType>

replyLst MUST precede txBody. Our CT_ThreadedComment ZeroOrOne
successors emitted txBody first, replyLst after — out of sequence, so
PowerPoint validated the parent and discarded the misplaced replyLst.
The reply element itself (CT_CommentReply = EG_CommentProperties only)
was already correct; this was purely a child-ordering defect.

Fix: reorder the successor tuples in src/pptx/oxml/comments.py so the
emitted order is pc:sldMkLst, [pos], p188:replyLst, p188:txBody,
[p188:extLst] — matching the schema sequence exactly. No API change.

Two new regression assertions lock it
(it_places_replyLst_before_txBody_per_ms_pptx_sequence,
it_keeps_reply_txBody_with_bodyPr_after_reorder); red->green proven by
stashing oxml/comments.py (the order test fails pre-fix).

Verification: pytest 3761 passed; ruff check All checks passed; ruff
format clean; behave 1062 scenarios / 3215 steps 0 failed. Regenerated
UAT deck: replyLst-before-txBody True, reply texts present. Opened in
real PowerPoint via Interceptor — AX tree now exposes the reply node
"Comment from Alex Reviewer. Merging now." plus a "View 1 more reply"
control (PowerPoint parsed the full reply list); pre-fix that count was
zero. uat/FIXED_replies_rendering.png.

Issue #25 now fully resolves: parent comments AND replies render in
PowerPoint's Comments pane. UAT script runs clean — pending maintainer
visual signoff in PowerPoint/Keynote. No push/PR per repo §6a. Refs #25.

Author: MHoroszowski

src/pptx/oxml/comments.py

@@ -334,11 +334,23 @@ class CT_ThreadedComment(BaseOxmlElement):
     empty-Comments-pane root cause). Replies do not carry their own marker.
     """
 
-    sldMkLst = ZeroOrOne("pc:sldMkLst", successors=("p188:txBody", "p188:replyLst", "p188:extLst"))
-    txBody = ZeroOrOne("p188:txBody", successors=("p188:replyLst", "p188:extLst"))
+    # Child order is FIXED by [MS-PPTX] CT_Comment (authoritative spec, the
+    # contract PowerPoint implements):
+    #   <xsd:sequence>
+    #     EG_CommentAnchor (pc:sldMkLst)   minOccurs=1
+    #     pos      (a:CT_Point2D)          minOccurs=0
+    #     replyLst (CT_CommentReplyList)   minOccurs=0
+    #     EG_CommentProperties (txBody, extLst) minOccurs=1
+    #   </xsd:sequence>
+    # i.e. replyLst MUST precede txBody. Emitting <p188:replyLst> AFTER
+    # <p188:txBody> (the pre-2026-05-17 order) is out-of-sequence: PowerPoint
+    # renders the parent comment but SILENTLY DROPS every reply. The successor
+    # tuples below encode this exact order (pos omitted — optional, unused).
+    sldMkLst = ZeroOrOne("pc:sldMkLst", successors=("p188:replyLst", "p188:txBody", "p188:extLst"))
     replyLst: "CT_ThreadedCommentReplyList | None" = ZeroOrOne(  # pyright: ignore
-        "p188:replyLst", successors=("p188:extLst",)
+        "p188:replyLst", successors=("p188:txBody", "p188:extLst")
     )
+    txBody = ZeroOrOne("p188:txBody", successors=("p188:extLst",))
 
     id: str = RequiredAttribute(  # pyright: ignore[reportAssignmentType]
         "id", XsdString

tests/test_issue25_comments.py

@@ -758,6 +758,49 @@ def it_keeps_exactly_one_sldMkLst_when_set_twice(self):
             "set_slide_marker must be idempotent (exactly one <pc:sldMkLst>)"
         )
 
+    def it_places_replyLst_before_txBody_per_ms_pptx_sequence(self):
+        # [MS-PPTX] CT_Comment sequence: EG_CommentAnchor, pos?, replyLst?,
+        # EG_CommentProperties(txBody, extLst). replyLst MUST precede txBody.
+        # Emitting replyLst AFTER txBody makes PowerPoint render the parent
+        # comment but SILENTLY DROP every reply (issue #25, the reply defect).
+        prs = Presentation()
+        s = prs.slides.add_slide(prs.slide_layouts[1])
+        c = s.comments.add("parent body", author="Rev")
+        c.replies.add("a reply body", author="Rev2")
+        buf = io.BytesIO()
+        prs.save(buf)
+        buf.seek(0)
+        import zipfile
+
+        z = zipfile.ZipFile(buf)
+        name = next(n for n in z.namelist() if "modernComment" in n)
+        xml = z.read(name).decode()
+        assert "replyLst" in xml, "expected a <p188:replyLst> for the reply"
+        assert "a reply body" in xml, "reply text must serialize"
+        assert xml.index("replyLst") < xml.index("txBody"), (
+            "<p188:replyLst> must precede <p188:txBody> per [MS-PPTX] "
+            "CT_Comment sequence, or PowerPoint drops every reply"
+        )
+
+    def it_keeps_reply_txBody_with_bodyPr_after_reorder(self):
+        # Guard the reorder didn't regress the reply body shape.
+        prs = Presentation()
+        s = prs.slides.add_slide(prs.slide_layouts[1])
+        c = s.comments.add("p", author="Rev")
+        c.replies.add("r", author="Rev2")
+        buf = io.BytesIO()
+        prs.save(buf)
+        buf.seek(0)
+        import re
+        import zipfile
+
+        z = zipfile.ZipFile(buf)
+        name = next(n for n in z.namelist() if "modernComment" in n)
+        xml = z.read(name).decode()
+        reply = re.search(r"<p188:reply\b.*?</p188:reply>", xml, re.S).group(0)
+        assert "bodyPr" in reply
+        assert reply.index("bodyPr") < reply.index("<a:p")
+
 
 if __name__ == "__main__":
     raise SystemExit(pytest.main([__file__, "-q"]))