Fix LT-22586: Normalize Word export content to NFC

Src/xWorks/ConfiguredLcmGenerator.cs

@@ -429,9 +429,14 @@ internal static IFragment GenerateContentForEntry(ICmObject entry, ConfigurableD
 					settings.ContentGenerator.EndEntry(xw);
 					xw.Flush();
 
-					// Do not normalize the string if exporting to word doc--it is not needed and will cause loss of document styles
-					if (bldr is LcmWordGenerator.DocFragment)
-						return bldr;
+					// All content should be in NFC (LT-18177). For Word export we normalize the text inside the
+					// OpenXml runs in place: normalizing bldr.ToString() and rebuilding a fragment from the string
+					// (as done below for XHTML) would discard the document's run and paragraph styles.
+					if (bldr is LcmWordGenerator.DocFragment wordFragment)
+					{
+						wordFragment.NormalizeText(FwNormalizationMode.knmNFC);
+						return wordFragment;
+					}
 
 					return settings.ContentGenerator.CreateFragment(CustomIcu.GetIcuNormalizer(FwNormalizationMode.knmNFC).Normalize(bldr.ToString())); // All content should be in NFC (LT-18177)
 				}

Src/xWorks/LcmWordGenerator.cs

@@ -10,6 +10,7 @@
 using SIL.FieldWorks.Common.FwUtils;
 using SIL.FieldWorks.Common.Widgets;
 using SIL.LCModel;
+using SIL.LCModel.Core.Text;
 using SIL.LCModel.Core.WritingSystems;
 using SIL.LCModel.DomainServices;
 using SIL.LCModel.Utils;
@@ -424,6 +425,11 @@ internal static DocFragment GenerateLetterHeaderDocFragment(string str, string s
 				// Only create paragraph, run, and text objects if string is nonempty
 				if (!string.IsNullOrEmpty(str))
 				{
+					// All exported content should be NFC (LT-18177). The header letter derives from the NFD sort
+					// word (see ConfiguredLcmGenerator.GetSortWordForLetterHead) and its upper-cased form can be
+					// decomposed, so normalize before writing it into the run.
+					str = CustomIcu.GetIcuNormalizer(FwNormalizationMode.knmNFC).Normalize(str);
+
 					// Don't add this paragraph before the first letter header. It results in an extra blank line.
 					if (!firstHeader)
 					{
@@ -540,6 +546,27 @@ public int Length()
 				return str.Length;
 			}
 
+			/// <summary>
+			/// Normalizes the text in every run of the fragment to the given Unicode normalization form, in place.
+			/// FieldWorks keeps strings as NFD in memory, but all exported content should be NFC (LT-18177), to
+			/// match the XHTML/Webonary export. We normalize the text within the OpenXml text nodes rather than
+			/// round-tripping the fragment through a string so the document structure and styles are preserved.
+			/// </summary>
+			public void NormalizeText(FwNormalizationMode mode)
+			{
+				if (IsNullOrEmpty())
+					return;
+
+				var normalizer = CustomIcu.GetIcuNormalizer(mode);
+				foreach (WP.Text txt in DocBody.Descendants<WP.Text>())
+				{
+					// Setting Text only changes the run's text value; the xml:space attribute and all
+					// run/paragraph properties (i.e. the styles) are left untouched.
+					if (!string.IsNullOrEmpty(txt.Text))
+						txt.Text = normalizer.Normalize(txt.Text);
+				}
+			}
+
 			/// <summary>
 			/// Appends one doc fragment to another.
 			/// Use this if styles have already been applied.

Src/xWorks/xWorksTests/LcmWordGeneratorTests.cs

@@ -18,6 +18,7 @@
 using SIL.LCModel;
 using SIL.LCModel.Core.KernelInterfaces;
 using SIL.LCModel.Core.Text;
+using SIL.LCModel.Core.WritingSystems;
 using SIL.LCModel.DomainServices;
 using SIL.WritingSystems;
 using SIL.TestUtilities;
@@ -599,6 +600,65 @@ public void GenerateWordDocForEntry_LineBreaksInBeforeContentWork()
 				WordNamespaceManager);
 		}
 
+		[Test]
+		public void GenerateWordDocForEntry_GeneratesNFC()
+		{
+			// FieldWorks stores strings as NFD (decomposed) in memory. All exported content should be NFC
+			// (LT-18177), matching the XHTML/Webonary export. This guards the Word export against emitting NFD.
+			var headwordNode = new ConfigurableDictionaryNode
+			{
+				FieldDescription = "MLHeadWord",
+				CSSClassNameOverride = "headword",
+				DictionaryNodeOptions = ConfiguredXHTMLGeneratorTests.GetWsOptionsForLanguages(new[] { "ko" }),
+				Style = "Dictionary-Headword"
+			};
+			var mainEntryNode = new ConfigurableDictionaryNode
+			{
+				FieldDescription = "LexEntry",
+				Children = new List<ConfigurableDictionaryNode> { headwordNode },
+				Style = MainEntryParagraphStyleName
+			};
+			CssGeneratorTests.PopulateFieldsForTesting(mainEntryNode);
+
+			Cache.LangProject.AddToCurrentVernacularWritingSystems(Cache.WritingSystemFactory.get_Engine("ko") as CoreWritingSystemDefinition);
+			var wsKo = Cache.WritingSystemFactory.GetWsFromStr("ko");
+			var entry = Cache.ServiceLocator.GetInstance<ILexEntryFactory>().Create();
+			// Decompose explicitly so the input is NFD regardless of how this source file stores the literal.
+			var nfdHeadword = CustomIcu.GetIcuNormalizer(FwNormalizationMode.knmNFD).Normalize("자ㄱㄴ시"); // Korean
+			entry.CitationForm.set_String(wsKo, TsStringUtils.MakeString(nfdHeadword, wsKo));
+			var storedHeadword = entry.CitationForm.get_String(wsKo);
+			Assert.That(storedHeadword.get_IsNormalizedForm(FwNormalizationMode.knmNFD), "Should be NFDecomposed in memory");
+			Assert.That(storedHeadword.Text.Length, Is.EqualTo(6), "NFD headword should have 6 codepoints");
+
+			//SUT
+			var result = ConfiguredLcmGenerator.GenerateContentForEntry(entry, mainEntryNode, null, DefaultSettings, 0) as DocFragment;
+			Assert.That(result, Is.Not.Null, "Results should have been generated");
+			var tsResult = TsStringUtils.MakeString(result.ToString(), wsKo);
+			Assert.That(TsStringUtils.IsNullOrEmpty(tsResult), Is.False, "Results should have been generated");
+			Assert.That(tsResult.get_IsNormalizedForm(FwNormalizationMode.knmNFC), "Resulting Word content should be NFComposed (NFC)");
+		}
+
+		[Test]
+		public void GenerateLetterHeaderDocFragment_GeneratesNFC()
+		{
+			// The letter header's letter derives from the NFD sort word (see GetSortWordForLetterHead) and its
+			// upper-cased form can be decomposed. All exported content should be NFC (LT-18177).
+			var wsString = Cache.WritingSystemFactory.GetStrFromWs(Cache.DefaultVernWs);
+			// Vietnamese "Ầ ầ" fully decomposed (A/a + combining circumflex + combining grave).
+			var nfdHeader = CustomIcu.GetIcuNormalizer(FwNormalizationMode.knmNFD).Normalize("Ầ ầ");
+			Assert.That(TsStringUtils.MakeString(nfdHeader, Cache.DefaultVernWs).get_IsNormalizedForm(FwNormalizationMode.knmNFD),
+				"Sanity check: the header input should be NFDecomposed");
+
+			//SUT
+			var headerFrag = DocFragment.GenerateLetterHeaderDocFragment(nfdHeader,
+				WordStylesGenerator.LetterHeadingDisplayName, false, wsString);
+
+			var tsResult = TsStringUtils.MakeString(headerFrag.ToString(), Cache.DefaultVernWs);
+			Assert.That(TsStringUtils.IsNullOrEmpty(tsResult), Is.False, "Header should have been generated");
+			Assert.That(tsResult.get_IsNormalizedForm(FwNormalizationMode.knmNFC),
+				"Letter header content should be NFComposed (NFC)");
+		}
+
 		[Test]
 		public void GenerateUniqueStyleName()
 		{