/// <summary>
/// Create a composite schema of both the partitioned columns and the underlying loader columns.
/// </summary>
/// <param name="ectx">The exception context.</param>
/// <param name="cols">The partitioned columns.</param>
/// <param name="subLoader">The sub loader.</param>
/// <returns>The resulting schema.</returns>
private Schema CreateSchema(IExceptionContext ectx, Column[] cols, IDataLoader subLoader)
{
Contracts.AssertValue(cols);
Contracts.AssertValue(subLoader);
var builder = new SchemaBuilder();
builder.AddColumns(cols.Select(c => new Schema.DetachedColumn(c.Name, ColumnTypeExtensions.PrimitiveTypeFromKind(c.Type.Value), null)));
var colSchema = builder.GetSchema();
var subSchema = subLoader.Schema;
if (subSchema.Count == 0)
{
return(colSchema);
}
else
{
var schemas = new Schema[]
{
subSchema,
colSchema
};
return(new ZipBinding(schemas).OutputSchema);
}
}
public Bindings(ModelLoadContext ctx, DatabaseLoader parent)
{
Contracts.AssertValue(ctx);
// *** Binary format ***
// int: number of columns
// foreach column:
// int: id of column name
// byte: DataKind
// byte: bool of whether this is a key type
// for a key type:
// ulong: count for key range
// byte: bool of whether the source index is valid
// for a valid source index:
// int: source index
int cinfo = ctx.Reader.ReadInt32();
Contracts.CheckDecode(cinfo > 0);
Infos = new ColInfo[cinfo];
for (int iinfo = 0; iinfo < cinfo; iinfo++)
{
string name = ctx.LoadNonEmptyString();
PrimitiveDataViewType itemType;
var kind = (InternalDataKind)ctx.Reader.ReadByte();
Contracts.CheckDecode(Enum.IsDefined(typeof(InternalDataKind), kind));
bool isKey = ctx.Reader.ReadBoolByte();
if (isKey)
{
ulong count;
Contracts.CheckDecode(KeyDataViewType.IsValidDataType(kind.ToType()));
count = ctx.Reader.ReadUInt64();
Contracts.CheckDecode(0 < count);
itemType = new KeyDataViewType(kind.ToType(), count);
}
else
{
itemType = ColumnTypeExtensions.PrimitiveTypeFromKind(kind);
}
int? sourceIndex = null;
bool hasSourceIndex = ctx.Reader.ReadBoolByte();
if (hasSourceIndex)
{
sourceIndex = ctx.Reader.ReadInt32();
}
Infos[iinfo] = new ColInfo(name, sourceIndex, itemType);
}
OutputSchema = ComputeOutputSchema();
}
public Bindings(DatabaseLoader parent, Column[] cols)
{
Contracts.AssertNonEmpty(cols);
using (var ch = parent._host.Start("Binding"))
{
// Make sure all columns have at least one source range.
foreach (var col in cols)
{
if (col.Source < 0)
{
throw ch.ExceptUserArg(nameof(Column.Source), "Source column index must be non-negative");
}
}
Infos = new ColInfo[cols.Length];
// This dictionary is used only for detecting duplicated column names specified by user.
var nameToInfoIndex = new Dictionary <string, int>(Infos.Length);
for (int iinfo = 0; iinfo < Infos.Length; iinfo++)
{
var col = cols[iinfo];
ch.CheckNonWhiteSpace(col.Name, nameof(col.Name));
string name = col.Name.Trim();
if (iinfo == nameToInfoIndex.Count && nameToInfoIndex.ContainsKey(name))
{
ch.Info("Duplicate name(s) specified - later columns will hide earlier ones");
}
PrimitiveDataViewType itemType;
if (col.KeyCount != null)
{
itemType = ConstructKeyType(col.Type, col.KeyCount);
}
else
{
itemType = ColumnTypeExtensions.PrimitiveTypeFromType(col.Type.ToType());
}
Infos[iinfo] = new ColInfo(name, col.Source, itemType);
nameToInfoIndex[name] = iinfo;
}
}
OutputSchema = ComputeOutputSchema();
}
public Bindings(ModelLoadContext ctx, DatabaseLoader parent)
{
Contracts.AssertValue(ctx);
// *** Binary format ***
// int: number of columns
// foreach column:
// int: id of column name
// byte: DataKind
// byte: bool of whether this is a key type
// for a key type:
// ulong: count for key range
// int: number of segments
// foreach segment:
// string id: name
// int: min
// int: lim
// byte: force vector (verWrittenCur: verIsVectorSupported)
int cinfo = ctx.Reader.ReadInt32();
Contracts.CheckDecode(cinfo > 0);
Infos = new ColInfo[cinfo];
for (int iinfo = 0; iinfo < cinfo; iinfo++)
{
string name = ctx.LoadNonEmptyString();
PrimitiveDataViewType itemType;
var kind = (InternalDataKind)ctx.Reader.ReadByte();
Contracts.CheckDecode(Enum.IsDefined(typeof(InternalDataKind), kind));
bool isKey = ctx.Reader.ReadBoolByte();
if (isKey)
{
ulong count;
Contracts.CheckDecode(KeyDataViewType.IsValidDataType(kind.ToType()));
count = ctx.Reader.ReadUInt64();
Contracts.CheckDecode(0 < count);
itemType = new KeyDataViewType(kind.ToType(), count);
}
else
{
itemType = ColumnTypeExtensions.PrimitiveTypeFromKind(kind);
}
int cseg = ctx.Reader.ReadInt32();
Segment[] segs;
if (cseg == 0)
{
segs = null;
}
else
{
Contracts.CheckDecode(cseg > 0);
segs = new Segment[cseg];
for (int iseg = 0; iseg < cseg; iseg++)
{
string columnName = ctx.LoadStringOrNull();
int min = ctx.Reader.ReadInt32();
int lim = ctx.Reader.ReadInt32();
Contracts.CheckDecode(0 <= min && min < lim);
bool forceVector = ctx.Reader.ReadBoolByte();
segs[iseg] = (columnName is null) ? new Segment(min, lim, forceVector) : new Segment(columnName, forceVector);
}
}
// Note that this will throw if the segments are ill-structured, including the case
// of multiple variable segments (since those segments will overlap and overlapping
// segments are illegal).
Infos[iinfo] = ColInfo.Create(name, itemType, segs, false);
}
OutputSchema = ComputeOutputSchema();
}
public Bindings(DatabaseLoader parent, Column[] cols)
{
Contracts.AssertNonEmpty(cols);
using (var ch = parent._host.Start("Binding"))
{
Infos = new ColInfo[cols.Length];
// This dictionary is used only for detecting duplicated column names specified by user.
var nameToInfoIndex = new Dictionary <string, int>(Infos.Length);
for (int iinfo = 0; iinfo < Infos.Length; iinfo++)
{
var col = cols[iinfo];
ch.CheckNonWhiteSpace(col.Name, nameof(col.Name));
string name = col.Name.Trim();
if (iinfo == nameToInfoIndex.Count && nameToInfoIndex.ContainsKey(name))
{
ch.Info("Duplicate name(s) specified - later columns will hide earlier ones");
}
PrimitiveDataViewType itemType;
if (col.KeyCount != null)
{
itemType = ConstructKeyType(col.Type, col.KeyCount);
}
else
{
ch.CheckUserArg(Enum.IsDefined(typeof(DbType), col.Type), nameof(Column.Type), "Bad item type");
itemType = ColumnTypeExtensions.PrimitiveTypeFromType(col.Type.ToType());
}
Segment[] segs = null;
if (col.Source != null)
{
segs = new Segment[col.Source.Length];
for (int i = 0; i < segs.Length; i++)
{
var range = col.Source[i];
Segment seg;
if (range.Name is null)
{
int min = range.Min;
ch.CheckUserArg(0 <= min, nameof(range.Min));
int max = range.Max;
ch.CheckUserArg(min <= max, nameof(range.Max));
seg = new Segment(min, max + 1, range.ForceVector);
}
else
{
string columnName = range.Name;
ch.CheckUserArg(columnName != null, nameof(range.Name));
seg = new Segment(columnName, range.ForceVector);
}
segs[i] = seg;
}
}
Infos[iinfo] = ColInfo.Create(name, itemType, segs, true);
nameToInfoIndex[name] = iinfo;
}
}
OutputSchema = ComputeOutputSchema();
}
public static InternalSchemaDefinition Create(Type userType, SchemaDefinition userSchemaDefinition)
{
Contracts.AssertValue(userType);
Contracts.AssertValueOrNull(userSchemaDefinition);
if (userSchemaDefinition == null)
{
userSchemaDefinition = SchemaDefinition.Create(userType);
}
Column[] dstCols = new Column[userSchemaDefinition.Count];
for (int i = 0; i < userSchemaDefinition.Count; ++i)
{
var col = userSchemaDefinition[i];
if (col.MemberName == null)
{
throw Contracts.ExceptParam(nameof(userSchemaDefinition), "Null field name detected in schema definition");
}
bool isVector;
Type dataItemType;
MemberInfo memberInfo = null;
if (col.Generator == null)
{
memberInfo = userType.GetField(col.MemberName);
if (memberInfo == null)
{
memberInfo = userType.GetProperty(col.MemberName);
}
if (memberInfo == null)
{
throw Contracts.ExceptParam(nameof(userSchemaDefinition), "No field or property with name '{0}' found in type '{1}'",
col.MemberName,
userType.FullName);
}
//Clause to handle the field that may be used to expose the cursor channel.
//This field does not need a column.
if ((memberInfo is FieldInfo && (memberInfo as FieldInfo).FieldType == typeof(IChannel)) ||
(memberInfo is PropertyInfo && (memberInfo as PropertyInfo).PropertyType == typeof(IChannel)))
{
continue;
}
GetVectorAndItemType(memberInfo, out isVector, out dataItemType);
}
else
{
var parameterType = col.ReturnType;
if (parameterType == null)
{
throw Contracts.ExceptParam(nameof(userSchemaDefinition), "No return parameter found in computed column.");
}
GetVectorAndItemType("returnType", parameterType, null, out isVector, out dataItemType);
}
// Infer the column name.
var colName = string.IsNullOrEmpty(col.ColumnName) ? col.MemberName : col.ColumnName;
// REVIEW: Because order is defined, we allow duplicate column names, since producing an IDataView
// with duplicate column names is completely legal. Possible objection is that we should make it less
// convenient to produce "hidden" columns, since this may not be of practical use to users.
DataViewType colType;
if (col.ColumnType == null)
{
// Infer a type as best we can.
PrimitiveDataViewType itemType = ColumnTypeExtensions.PrimitiveTypeFromType(dataItemType);
colType = isVector ? new VectorDataViewType(itemType) : (DataViewType)itemType;
}
else
{
// Make sure that the types are compatible with the declared type, including
// whether it is a vector type.
VectorDataViewType columnVectorType = col.ColumnType as VectorDataViewType;
if (isVector != (columnVectorType != null))
{
throw Contracts.ExceptParam(nameof(userSchemaDefinition), "Column '{0}' is supposed to be {1}, but type of associated field '{2}' is {3}",
colName, columnVectorType != null ? "vector" : "scalar", col.MemberName, isVector ? "vector" : "scalar");
}
DataViewType itemType = columnVectorType?.ItemType ?? col.ColumnType;
if (itemType.RawType != dataItemType)
{
throw Contracts.ExceptParam(nameof(userSchemaDefinition), "Column '{0}' is supposed to have item type {1}, but associated field has type {2}",
colName, itemType.RawType, dataItemType);
}
colType = col.ColumnType;
}
dstCols[i] = col.Generator != null ?
new Column(colName, colType, col.Generator, col.AnnotationInfos)
: new Column(colName, colType, memberInfo, col.AnnotationInfos);
}
return(new InternalSchemaDefinition(dstCols));
}
