CSVParser.h
下面的代码向您展示了一个使用Matt Gallagher 类并假设实体在 Core Data 中加载 csv 文件的示例MyEntity
// CSV File input.csv and not input.txt separate by space @" "
CSVParser *csvParser = [[CSVParser alloc] initWithContentOfFile:[NSBundle mainBundle] pathForResource:@"input" ofType:@"csv" separator:@" "];
// Array with all your data from CSV
NSArray *data = [csvParser parseFile];
// Your entity from Core Data
MyEntity *myEntity = nil;
NSFetchRequest *fetchRequest = [[NSFetchRequest alloc] init];
for (NSDictionary *dic in data)
{
fetchRequest.entity = [NSEntityDescription entityForName:@"MyEntity" inManagedObjectContext:context];
if (!entity)
entity = [NSEntityDescription insertNewObjectForEntityForName:@"MyEntity"
inManagedObjectContext:context];
[entity setValue:number forKey:@"number"];
[entity setValue:string1 forKey:@"string1"];
[entity setValue:string2 forKey:@"string2"];
}
// Save the context
[managedObjectContext save:nil];
CVSParser.h 使用 ARC :
//
// CSVParser.h
// CSVImporter
//
// Created by Matt Gallagher on 2009/11/30.
// Copyright 2009 Matt Gallagher. All rights reserved.
//
// Permission is given to use this source code file, free of charge, in any
// project, commercial or otherwise, entirely at your risk, with the condition
// that any redistribution (in part or whole) of source code must retain
// this copyright and permission notice. Attribution in compiled projects is
// appreciated but not required.
//
// Source : http://cocoawithlove.com/2009/11/writing-parser-using-nsscanner-csv.html
@interface CSVParser : NSObject
{
NSString *csvString;
NSString *separator;
NSScanner *scanner;
BOOL hasHeader;
NSMutableArray *fieldNames;
id receiver;
SEL receiverSelector;
NSCharacterSet *endTextCharacterSet;
BOOL separatorIsSingleChar;
}
- (id)initWithString:(NSString *)aCSVString
separator:(NSString *)aSeparatorString;
- (id)initWithContentOfFile:(NSString *)aPath
separator:(NSString *)aSeparatorString;
- (id)initWithString:(NSString *)aCSVString
separator:(NSString *)aSeparatorString
hasHeader:(BOOL)header
fieldNames:(NSArray *)names;
- (id)initWithContentOfFile:(NSString *)aPath
separator:(NSString *)aSeparatorString
hasHeader:(BOOL)header
fieldNames:(NSArray *)names;
- (NSArray *)arrayOfParsedRows;
- (void)parseRowsForReceiver:(id)aReceiver selector:(SEL)aSelector;
- (NSArray *)parseFile;
- (NSMutableArray *)parseHeader;
- (NSDictionary *)parseRecord;
- (NSString *)parseName;
- (NSString *)parseField;
- (NSString *)parseEscaped;
- (NSString *)parseNonEscaped;
- (NSString *)parseDoubleQuote;
- (NSString *)parseSeparator;
- (NSString *)parseLineSeparator;
- (NSString *)parseTwoDoubleQuotes;
- (NSString *)parseTextData;
@end
CVSParser.m 使用 ARC :
//
// CSVParser.m
// CSVImporter
//
// Created by Matt Gallagher on 2009/11/30.
// Copyright 2009 Matt Gallagher. All rights reserved.
//
// Permission is given to use this source code file, free of charge, in any
// project, commercial or otherwise, entirely at your risk, with the condition
// that any redistribution (in part or whole) of source code must retain
// this copyright and permission notice. Attribution in compiled projects is
// appreciated but not required.
//
#import "CSVParser.h"
@implementation CSVParser
//
// initWithString:separator:hasHeader:fieldNames:
//
// Parameters:
// aCSVString - the string that will be parsed
// aSeparatorString - the separator (normally "," or "\t")
// header - if YES, treats the first row as a list of field names
// names - a list of field names (will have no effect if header is YES)
//
// returns the initialized object (nil on failure)
//
- (id)initWithString:(NSString *)aCSVString
separator:(NSString *)aSeparatorString
hasHeader:(BOOL)header
fieldNames:(NSArray *)names
{
self = [super init];
if (self)
{
csvString = [aCSVString retain];
separator = [aSeparatorString retain];
NSAssert([separator length] > 0 &&
[separator rangeOfString:@"\""].location == NSNotFound &&
[separator rangeOfCharacterFromSet:[NSCharacterSet newlineCharacterSet]].location == NSNotFound,
@"CSV separator string must not be empty and must not contain the double quote character or newline characters.");
NSMutableCharacterSet *endTextMutableCharacterSet =
[[NSCharacterSet newlineCharacterSet] mutableCopy];
[endTextMutableCharacterSet addCharactersInString:@"\""];
[endTextMutableCharacterSet addCharactersInString:[separator substringToIndex:1]];
endTextCharacterSet = endTextMutableCharacterSet;
if ([separator length] == 1)
{
separatorIsSingleChar = YES;
}
hasHeader = header;
fieldNames = [names mutableCopy];
}
return self;
}
- (id)initWithString:(NSString *)aCSVString
separator:(NSString *)aSeparatorString
{
return [self initWithString:aCSVString
separator:aSeparatorString
hasHeader:YES
fieldNames:nil];
}
- (id)initWithContentOfFile:(NSString *)aPath
separator:(NSString *)aSeparatorString
{
return [self initWithString:[NSString stringWithContentsOfFile:aPath
encoding:NSUTF8StringEncoding
error:nil]
separator:aSeparatorString];
}
- (id)initWithContentOfFile:(NSString *)aPath
separator:(NSString *)aSeparatorString
hasHeader:(BOOL)header
fieldNames:(NSArray *)names
{
return [self initWithString:[NSString stringWithContentsOfFile:aPath
encoding:NSUTF8StringEncoding
error:nil]
separator:aSeparatorString
hasHeader:header
fieldNames:names];
}
//
// dealloc
//
// Releases instance memory.
//
- (void)dealloc
{
[csvString release];
[separator release];
[fieldNames release];
[endTextCharacterSet release];
[super dealloc];
}
//
// arrayOfParsedRows
//
// Performs a parsing of the csvString, returning the entire result.
//
// returns the array of all parsed row records
//
- (NSArray *)arrayOfParsedRows
{
scanner = [[NSScanner alloc] initWithString:csvString];
[scanner setCharactersToBeSkipped:[[[NSCharacterSet alloc] init] autorelease]];
NSArray *result = [self parseFile];
[scanner release];
scanner = nil;
return result;
}
//
// parseRowsForReceiver:selector:
//
// Performs a parsing of the csvString, sending the entries, 1 row at a time,
// to the receiver.
//
// Parameters:
// aReceiver - the target that will receive each row as it is parsed
// aSelector - the selector that will receive each row as it is parsed
// (should be a method that takes a single NSDictionary argument)
//
- (void)parseRowsForReceiver:(id)aReceiver selector:(SEL)aSelector
{
scanner = [[NSScanner alloc] initWithString:csvString];
[scanner setCharactersToBeSkipped:[[[NSCharacterSet alloc] init] autorelease]];
receiver = [aReceiver retain];
receiverSelector = aSelector;
[self parseFile];
[scanner release];
scanner = nil;
[receiver release];
receiver = nil;
}
//
// parseFile
//
// Attempts to parse a file from the current scan location.
//
// returns the parsed results if successful and receiver is nil, otherwise
// returns nil when done or on failure.
//
- (NSArray *)parseFile
{
scanner = [[NSScanner alloc] initWithString:csvString];
[scanner setCharactersToBeSkipped:[[[NSCharacterSet alloc] init] autorelease]];
if (hasHeader)
{
if (fieldNames)
{
[fieldNames release];
}
fieldNames = [[self parseHeader] retain];
if (!fieldNames || ![self parseLineSeparator])
{
return nil;
}
}
NSMutableArray *records = nil;
if (!receiver)
{
records = [NSMutableArray array];
}
NSDictionary *record = [[self parseRecord] retain];
if (!record)
{
return nil;
}
while (record)
{
NSAutoreleasePool *pool = [[NSAutoreleasePool alloc] init];
if (receiver)
{
[receiver performSelector:receiverSelector withObject:record];
}
else
{
[records addObject:record];
}
[record release];
if (![self parseLineSeparator])
{
break;
}
record = [[self parseRecord] retain];
[pool drain];
}
[scanner release];
scanner = nil;
return records;
}
//
// parseHeader
//
// Attempts to parse a header row from the current scan location.
//
// returns the array of parsed field names or nil on parse failure.
//
- (NSMutableArray *)parseHeader
{
NSString *name = [self parseName];
if (!name)
{
return nil;
}
NSMutableArray *names = [NSMutableArray array];
while (name)
{
[names addObject:name];
if (![self parseSeparator])
{
break;
}
name = [self parseName];
}
return names;
}
//
// parseRecord
//
// Attempts to parse a record from the current scan location. The record
// dictionary will use the fieldNames as keys, or FIELD_X for each column
// X-1 if no fieldName exists for a given column.
//
// returns the parsed record as a dictionary, or nil on failure.
//
- (NSDictionary *)parseRecord
{
//
// Special case: return nil if the line is blank. Without this special case,
// it would parse as a single blank field.
//
if ([self parseLineSeparator] || [scanner isAtEnd])
{
return nil;
}
NSString *field = [self parseField];
if (!field)
{
return nil;
}
NSInteger fieldNamesCount = [fieldNames count];
NSInteger fieldCount = 0;
NSMutableDictionary *record =
[NSMutableDictionary dictionaryWithCapacity:[fieldNames count]];
while (field)
{
NSString *fieldName;
if (fieldNamesCount > fieldCount)
{
fieldName = [fieldNames objectAtIndex:fieldCount];
}
else
{
fieldName = [NSString stringWithFormat:@"FIELD_%d", fieldCount + 1];
[fieldNames addObject:fieldName];
fieldNamesCount++;
}
[record setObject:field forKey:fieldName];
fieldCount++;
if (![self parseSeparator])
{
break;
}
field = [self parseField];
}
return record;
}
//
// parseName
//
// Attempts to parse a name from the current scan location.
//
// returns the name or nil.
//
- (NSString *)parseName
{
return [self parseField];
}
//
// parseField
//
// Attempts to parse a field from the current scan location.
//
// returns the field or nil
//
- (NSString *)parseField
{
NSString *escapedString = [self parseEscaped];
if (escapedString)
{
return escapedString;
}
NSString *nonEscapedString = [self parseNonEscaped];
if (nonEscapedString)
{
return nonEscapedString;
}
//
// Special case: if the current location is immediately
// followed by a separator, then the field is a valid, empty string.
//
NSInteger currentLocation = [scanner scanLocation];
if ([self parseSeparator] || [self parseLineSeparator] || [scanner isAtEnd])
{
[scanner setScanLocation:currentLocation];
return @"";
}
return nil;
}
//
// parseEscaped
//
// Attempts to parse an escaped field value from the current scan location.
//
// returns the field value or nil.
//
- (NSString *)parseEscaped
{
if (![self parseDoubleQuote])
{
return nil;
}
NSString *accumulatedData = [NSString string];
while (YES)
{
NSString *fragment = [self parseTextData];
if (!fragment)
{
fragment = [self parseSeparator];
if (!fragment)
{
fragment = [self parseLineSeparator];
if (!fragment)
{
if ([self parseTwoDoubleQuotes])
{
fragment = @"\"";
}
else
{
break;
}
}
}
}
accumulatedData = [accumulatedData stringByAppendingString:fragment];
}
if (![self parseDoubleQuote])
{
return nil;
}
return accumulatedData;
}
//
// parseNonEscaped
//
// Attempts to parse a non-escaped field value from the current scan location.
//
// returns the field value or nil.
//
- (NSString *)parseNonEscaped
{
return [self parseTextData];
}
//
// parseTwoDoubleQuotes
//
// Attempts to parse two double quotes from the current scan location.
//
// returns a string containing two double quotes or nil.
//
- (NSString *)parseTwoDoubleQuotes
{
if ([scanner scanString:@"\"\"" intoString:NULL])
{
return @"\"\"";
}
return nil;
}
//
// parseDoubleQuote
//
// Attempts to parse a double quote from the current scan location.
//
// returns @"\"" or nil.
//
- (NSString *)parseDoubleQuote
{
if ([scanner scanString:@"\"" intoString:NULL])
{
return @"\"";
}
return nil;
}
//
// parseSeparator
//
// Attempts to parse the separator string from the current scan location.
//
// returns the separator string or nil.
//
- (NSString *)parseSeparator
{
if ([scanner scanString:separator intoString:NULL])
{
return separator;
}
return nil;
}
//
// parseLineSeparator
//
// Attempts to parse newline characters from the current scan location.
//
// returns a string containing one or more newline characters or nil.
//
- (NSString *)parseLineSeparator
{
NSString *matchedNewlines = nil;
[scanner
scanCharactersFromSet:[NSCharacterSet newlineCharacterSet]
intoString:&matchedNewlines];
return matchedNewlines;
}
//
// parseTextData
//
// Attempts to parse text data from the current scan location.
//
// returns a non-zero length string or nil.
//
- (NSString *)parseTextData
{
NSString *accumulatedData = [NSString string];
while (YES)
{
NSString *fragment;
if ([scanner scanUpToCharactersFromSet:endTextCharacterSet intoString:&fragment])
{
accumulatedData = [accumulatedData stringByAppendingString:fragment];
}
//
// If the separator is just a single character (common case) then
// we know we've reached the end of parseable text
//
if (separatorIsSingleChar)
{
break;
}
//
// Otherwise, we need to consider the case where the first character
// of the separator is matched but we don't have the full separator.
//
NSUInteger location = [scanner scanLocation];
NSString *firstCharOfSeparator;
if ([scanner scanString:[separator substringToIndex:1] intoString:&firstCharOfSeparator])
{
if ([scanner scanString:[separator substringFromIndex:1] intoString:NULL])
{
[scanner setScanLocation:location];
break;
}
//
// We have the first char of the separator but not the whole
// separator, so just append the char and continue
//
accumulatedData = [accumulatedData stringByAppendingString:firstCharOfSeparator];
continue;
}
else
{
break;
}
}
if ([accumulatedData length] > 0)
{
return accumulatedData;
}
return nil;
}
@end