/// <summary>
/// Return whether the string is already in the specified normal form.
/// Note that a string may be considered to be in NFC
/// even though its text (the plain character sequence) is not.
/// This is because we don't collapse otherwise collapsible pairs if they
/// have different style properties.
/// </summary>
public bool get_IsNormalizedForm(FwNormalizationMode nm)
{
if (IsAlreadyNormalized(nm))
{
return(true);
}
if (string.IsNullOrEmpty(Text))
{
NoteAlreadyNormalized(nm);
return(true);
}
var normalizer = CustomIcu.GetIcuNormalizer(nm);
if (normalizer.IsNormalized(Text))
{
// Don't do this work a second time
if (nm == FwNormalizationMode.knmNFSC)
{
NoteAlreadyNormalized(FwNormalizationMode.knmNFC); // NFC includes NFSC
}
else
{
NoteAlreadyNormalized(nm);
}
return(true);
}
if (nm == FwNormalizationMode.knmNFSC)
{
// NFSC is a special case, where we just have to normalize and compare.
if (Equals(get_NormalizedForm(nm)))
{
NoteAlreadyNormalized(nm);
return(true);
}
}
return(false);
}
// Implementation of both get_NormalizedForm and NfdAndFixOffsets
private ITsString get_NormalizedFormAndFixOffsets(FwNormalizationMode nm, ArrayPtr oldOffsetsToFix, int numOffsetsToFix)
{
// Can we skip unnecessary work?
if (IsAlreadyNormalized(nm))
{
return(this);
}
if (string.IsNullOrEmpty(Text))
{
NoteAlreadyNormalized(nm);
return(this);
}
if (nm == FwNormalizationMode.knmLim)
{
throw new ArgumentException("Normalization mode may not be knmLim", "nm");
}
// NFSC needs to be decomposed first, then recomposed as NFC.
if (nm == FwNormalizationMode.knmNFSC && !get_IsNormalizedForm(FwNormalizationMode.knmNFD))
{
var nfd = (TsString)get_NormalizedForm(FwNormalizationMode.knmNFD);
// Line below is *not* a typo; this call will not recurse infinitely.
return(nfd.get_NormalizedFormAndFixOffsets(FwNormalizationMode.knmNFSC, oldOffsetsToFix, numOffsetsToFix));
}
bool willFixOffsets = numOffsetsToFix > 0 && oldOffsetsToFix != null && oldOffsetsToFix.IntPtr != IntPtr.Zero;
// Keys = offsets into original string, values = offsets into normalized string
var stringOffsetMapping = willFixOffsets ? new Dictionary <int, int>() : null; // Don't allocate an object if we'll never use it
var icuNormalizer = CustomIcu.GetIcuNormalizer(nm);
TsStrBldr resultBuilder = new TsStrBldr();
int segmentMin = 0;
foreach (int segmentLim in EnumerateSegmentLimits(icuNormalizer))
{
string segment = GetChars(segmentMin, segmentLim);
string normalizedSegment = icuNormalizer.Normalize(segment);
int curRun = get_RunAt(segmentMin);
int curRunLim = get_LimOfRun(curRun);
ITsTextProps curTextProps = get_Properties(curRun);
if (curRunLim >= segmentLim)
{
// The segment is contained entirely in the current run, so our job is simple
int outputLenSoFar = resultBuilder.Length;
resultBuilder.Replace(outputLenSoFar, outputLenSoFar, normalizedSegment, curTextProps);
// Calculate the orig -> norm index mappings if (and only if) they're needed, since this calculation is expensive
if (willFixOffsets)
{
foreach (RearrangedIndexMapping mapping in MatchUpIndexesAfterNormalization(segment, normalizedSegment, icuNormalizer))
{
// Note that our local mapping is from the start of this segment, but we want to keep track of indexes from the start
// of the *string*. (Both the original string and the output, normalized string). So we adjust the indexes here.
if (mapping.isFirstCharOfDecomposition)
{
stringOffsetMapping[segmentMin + mapping.origIdx] = outputLenSoFar + mapping.normIdx;
}
}
}
}
else
{
// The segment straddles two runs, so our job is harder. We have to either deal with decomposition
// rearranging things (and make sure the right characters maintain the right text properties), or
// else we have to deal with composition possibly trying to "compress" some diacritics that straddle
// a run border (which can happen, for example, if they have different text properties).
if (nm == FwNormalizationMode.knmNFD || nm == FwNormalizationMode.knmNFKD)
{
// Decomposition: we have to deal with rearranging. Some characters from after the first run's
// endpoint may have ended up "inside" the first run after rearranging, so their text properties
// will be incorrect at first. We'll fix them up after calculating the orig -> norm index mappings.
int outputLenSoFar = resultBuilder.Length; // This will be the start index from which
resultBuilder.Replace(outputLenSoFar, outputLenSoFar, normalizedSegment, curTextProps);
// Now correct the text properties, one index at a time.
IEnumerable <RearrangedIndexMapping> indexMappings = MatchUpIndexesAfterNormalization(segment, normalizedSegment, icuNormalizer);
foreach (RearrangedIndexMapping mapping in indexMappings)
{
ITsTextProps origProperties = get_PropertiesAt(segmentMin + mapping.origIdx);
int outputIdx = outputLenSoFar + mapping.normIdx;
int size = Char.IsSurrogate(normalizedSegment, mapping.normIdx) ? 2 : 1;
resultBuilder.SetProperties(outputIdx, outputIdx + size, origProperties);
// And if we also need to fix up offsets at the end, we keep track of the ones we'll need
if (willFixOffsets && mapping.isFirstCharOfDecomposition)
{
stringOffsetMapping[segmentMin + mapping.origIdx] = outputLenSoFar + mapping.normIdx;
}
}
}
else if (nm == FwNormalizationMode.knmNFSC)
{
// Composition that preserves styles. By this point, our input is NFD so we at least know there will be no rearranging.
// If there is more than one character remaining in the current run, then we might be able to compose those, at least.
if (curRunLim - segmentMin > 1)
{
// Unicode canonical ordering is such that any subsequence of a composed character can itself be composed, so this is safe.
string remainderOfFirstRun = GetChars(segmentMin, curRunLim);
string normalizedRemainder = icuNormalizer.Normalize(remainderOfFirstRun);
resultBuilder.Replace(resultBuilder.Length, resultBuilder.Length, normalizedRemainder, curTextProps);
// Now the start of the un-composable part is just the limit of the first run (which is the start of the second run).
segmentMin = curRunLim;
}
// Now there could be any NUMBER of runs between currentInputIdx and segmentLim. Maybe there are TEN composing
// characters, each with different text properties (and thus different runs). However, since the base character
// was in the first run, none of the characters from the second or subsequent runs are composable any longer. So we
// can copy them to the output as-is as one big TsString, which will carry text, runs and all.
ITsString uncomposablePartOfSegment = GetSubstring(segmentMin, segmentLim);
resultBuilder.ReplaceTsString(resultBuilder.Length, resultBuilder.Length, uncomposablePartOfSegment);
}
else
{
// For NFC and NFKC, we do not try to preserve styles or offset mappings, so this branch is quite simple
int outputLenSoFar = resultBuilder.Length;
resultBuilder.Replace(outputLenSoFar, outputLenSoFar, normalizedSegment, curTextProps);
}
}
segmentMin = segmentLim; // Next segment will start where the current segment ended
}
if (willFixOffsets)
{
stringOffsetMapping[segmentMin] = resultBuilder.Length;
int ptrSize = Marshal.SizeOf(typeof(IntPtr));
for (int i = 0; i < numOffsetsToFix; i++)
{
IntPtr offsetPtr = Marshal.ReadIntPtr(oldOffsetsToFix.IntPtr, i * ptrSize);
int oldOffset = Marshal.ReadInt32(offsetPtr);
int newOffset;
if (stringOffsetMapping.TryGetValue(oldOffset, out newOffset))
{
Marshal.WriteInt32(offsetPtr, newOffset);
}
else
{
// The only likely way for one of the offsets we've been asked to fix up to NOT
// be found in the offset mapping dictionary is if it happened to be an offset
// to the second half of a surrogate pair. In which case we want to fix it up to
// point to wherever the first half of that pair ended up, so searching downwards
// through the offset mapping dictionary will find the best match.
bool found = false;
while (!found && oldOffset > 0)
{
oldOffset--;
found = stringOffsetMapping.TryGetValue(oldOffset, out newOffset);
}
// Any offset that could not be matched at all will be pointed at the beginning
// of the TsString, since that's safe with strings of all sizes (including empty).
Marshal.WriteInt32(offsetPtr, found ? newOffset : 0);
}
}
}
var result = (TsString)resultBuilder.GetString();
result.NoteAlreadyNormalized(nm); // So we won't have to do all this work a second time
return(result);
}