Introduction
Base64 is an encoding for transferring binary data in 7-bit text. Originally used in email, it is also used for binary encoding data in HTML files. Another common use for Base64 is in HTTP Basic Access Authentication where it is used to transfer login details (which might not be printable characters).
The key library for handling Base64 on the Mac is normally libcrypto (the OpenSSL library), so it's a little disappointing that libcrypto isn't available on the iPhone.
Using OpenSSL
Via the command line
On the Mac, you can handle simple encoding tasks like base64 encoding with OpenSSL on the command line:
echo "Base64 encode this text." | openssl enc -base64
gives the encoding result:
QmFzZTY0IGVuY29kZSB0aGlzIHRleHQuCg==
The reverse is handled in the following manner:
echo "QmFzZTY0IGVuY29kZSB0aGlzIHRleHQuCg==" | openssl enc -d -base64
giving
Base64 encode this text.
In code
As you'd expect, doing the same work in code takes a little more typing. First, we're using a library, so we need to include it (in your Project's Build Settings under Other Linker Flags add the flag -lcrypto). Once that's done, you should be able to use the following method in a category on NSData:
#include <openssl/bio.h>
#include <openssl/evp.h>
- (NSString *)base64EncodedString
{
// Construct an OpenSSL context
BIO *context = BIO_new(BIO_s_mem());
// Tell the context to encode base64
BIO *command = BIO_new(BIO_f_base64());
context = BIO_push(command, context);
// Encode all the data
BIO_write(context, [self bytes], [self length]);
BIO_flush(context);
// Get the data out of the context
char *outputBuffer;
long outputLength = BIO_get_mem_data(context, &outputBuffer);
NSString *encodedString = [NSString
stringWithCString:outputBuffer
length:outputLength];
BIO_free_all(context);
return encodedString;
}
To handle a Base64 encode.
By default, encodedString will have newlines every 64 characters. If needed, you can disable the inclusion of newlines by adding the following line before the BIO_write:
BIO_set_flags(context, BIO_FLAGS_BASE64_NO_NL);
Tthe "BIO" system (I think it stands for buffered I/O) is not very symmetric so the code for decoding is quite different:
+ (NSData *)dataByDase64DecodingString:(NSString *)decode
{
decode = [decode stringByAppendingString:@"\n"];
NSData *data = [decode dataUsingEncoding:NSASCIIStringEncoding];
// Construct an OpenSSL context
BIO *command = BIO_new(BIO_f_base64());
BIO *context = BIO_new_mem_buf((void *)[data bytes], [data length]);
// Tell the context to encode base64
context = BIO_push(command, context);
// Encode all the data
NSMutableData *outputData = [NSMutableData data];
#define BUFFSIZE 256
int len;
char inbuf[BUFFSIZE];
while ((len = BIO_read(context, inbuf, BUFFSIZE)) > 0)
{
[outputData appendBytes:inbuf length:len];
}
BIO_free_all(context);
return outputData;
}
An interesting point to note at the top of this function: I add an extra newline to the start of the string. This is because if you have not disabled newlines and the string does not contain at least 1 newline, BIO_read will fail.
Handling Base64 on the iPhone
Using libcrypto isn't possible on the iPhone — the library isn't there. The other functions that libcrypto handles can be found elsewhere:
- md5 — use the CommonCrypto implementation
CC_MD5
- sha — use the CommonCrypto implementation
CC_SHA
- Public/Private Key Encryption/Decryption — use the
SecKeyEncrypt/SecKeyDecrypt functions in the Security framework
You can find the documentation for the Security Framework by performing a standard Xcode API lookup. For some reason though, the CommonCrypto functions only appear in a full-text search.
The Base64 functionality of OpenSSL doesn't have an accessible equivalent on the iPhone, even though NSURLConnection, CFHTTPMessageRef and WebKit must all have access to an implementation — whatever they use is not accessible.
Encoding Base64
Fortunately, Base64 is a fairly simple encoding. At its heart, it looks like this:
static unsigned char base64EncodeLookup[65] =
"ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";
//
// Inner loop: turn 3 bytes into 4 base64 characters
//
outputBuffer[j++] = base64EncodeLookup[(inputBuffer[i] & 0xFC) >> 2];
outputBuffer[j++] = base64EncodeLookup[((inputBuffer[i] & 0x03) << 4)
| ((inputBuffer[i + 1] & 0xF0) >> 4)];
outputBuffer[j++] = base64EncodeLookup[((inputBuffer[i + 1] & 0x0F) << 2)
| ((inputBuffer[i + 2] & 0xC0) >> 6)];
outputBuffer[j++] = base64EncodeLookup[inputBuffer[i + 2] & 0x3F];
This might be a little ugly to look at if you're not use to seeing bitmasks and bitshifts but it is only a couple lines. It does little more than the comment states: it turns 3 bytes into 4 chars, with the specific chars specified by the base64EncodeLookup mapping.
Of course, while this code handles the center of the main loop, there's almost a hundred lines total in the complete implementation that I wrote.
As part of keeping the function optimal, I wanted to keep the conditionals out of the inner loop (making vectorizing easier). I succeeded and there are no conditionals in the inner loop but this means that there are a few tail conditions to handle in the epilogue.
I also wanted to calculate the exact size that would be required for the output buffer, so it can be allocated once with no waste, but this too occupies a few lines worth of space.
Decoding Base64
Decoding works similarly to encoding, except that in decoding we are reducing 4 characters down to 3 bytes instead of vice versa:
//
// Store the 6 bits from each of the 4 characters as 3 bytes
//
outputBuffer[j] = (accumulated[0] << 2) | (accumulated[1] >> 4);
outputBuffer[j + 1] = (accumulated[1] << 4) | (accumulated[2] >> 2);
outputBuffer[j + 2] = (accumulated[2] << 6) | accumulated[3];
More interesting than the code in this case is the lookup table that each of these accumulated bytes passes through before being used here:
//
// Definition for "masked-out" areas of the base64DecodeLookup mapping
//
#define xx 65
//
// Mapping from ASCII character to 6 bit pattern.
//
static unsigned char base64DecodeLookup[256] =
{
xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx,
xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx,
xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, 62, xx, xx, xx, 63,
52, 53, 54, 55, 56, 57, 58, 59, 60, 61, xx, xx, xx, xx, xx, xx,
xx, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14,
15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, xx, xx, xx, xx, xx,
xx, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40,
41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, xx, xx, xx, xx, xx,
xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx,
xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx,
xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx,
xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx,
xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx,
xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx,
xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx,
xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx,
};
The "xx"s in this table are just a #define of 65 (i.e. outside the valid range of Base64) but they provide an interesting visual representation of the 6-bits that each Base64 character can occupy within the 8-bit byte.
I was unable to remove all of the conditionals from the inner loop of the decode side and keep the "skip over invalid characters" requirement.
This "skip over invalid characters" stage (where characters are accumulated until 4 valid characters are found) is handled by the following loop (which immediately preceeds the previous "store the 6 bits from each of the 4 characters as 3 bytes" code):
//
// Accumulate 4 valid characters (ignore everything else)
//
unsigned char accumulated[BASE64_UNIT_SIZE];
size_t accumulateIndex = 0;
while (i < length)
{
unsigned char decode = base64DecodeLookup[inputBuffer[i++]];
if (decode != xx)
{
accumulated[accumulateIndex] = decode;
accumulateIndex++;
if (accumulateIndex == BASE64_UNIT_SIZE)
{
break;
}
}
}
This is the only part which makes the decode stage sub-optimal. If you had Base64 input data with no newlines and no other characters requiring skipping, I think you could remove this section entirely so that the inner loop of the decode function could be vectorizable.
Conclusion
Download the NSData+Base64 class and header (4kB).
In this post, I've shown you how to use the default command-line and library options for Base64 handling on Mac OS X. I've also shown you the approach I use for Base64 encoding and decoding on the iPhone.
The libcrypto libraries (when available) are not as tight and simple as custom code for the task but do have the advantage that the pipeline for feeding data into them is more configurable.
I'm certainly not the only one to present C libraries for Base64 encoding that will work on the iPhone but the approach I've used should be efficient (especially the internal implementations in the C-functions) and it should drop into a Cocoa project on the iPhone very easily.
mustered. Because from the outside looking in the Mets’ training-medical situation looks awful.”
In the simplest case, we want to put the most selective column first in the index, so that the number of possible matching rows is the smallest, i.e. we find the rows as quickly as possible. Assuming that all the columns have an even distribution of values, we can just count the number of matching rows for each criterion.
Silicon Valley businessmen fancy themselves unflinching hard-core capitalists. Yet they hate to compete for workers — and the New York Times found an email to prove it.
Wow, 
In response to my 
“I’m sure when Sonia Sotomayor was six years old, being raised by a single Puerto Rican mother in the Bronx in the 70’s, she knew she could kick back and riiiiiide it out, because she was on fucking Easy Street, baby.” Sonia Sotomayor was nine when her father passed away. Other than that, 
Many New Yorkers have been quick to jump aboard the Stumptown bandwagon, eagerly anticipating the Portland coffee chain's roaster that's set to open in Red Hook, happily accepting it at some of the city's hippest restaurants (Momofuku, Frankies Spuntino, Freemans) while they await its big opening in the Ace Hotel this summer. It's safe to assume there are still glowing trend pieces about the growing coffee brand that have yet to be written. 
Dear Answer Lady,
Central Park: Now that Sir Danny Meyer has signed on the dotted line, reps at Union Square Hospitality Group can confirm that their fearless leader will 
I’m happy to announce that I’m joining the advisory board at 
This 
New York City's flagship Apple Store, located on 
![juliannemoore_vogue_july09cover[1].jpg](http://fashionista.com/juliannemoore_vogue_july09cover%5B1%5D.jpg)
This morning, 
The New York Times' Amanda Hesser made 
Valve is turning a sequel around in record Valve time, announcing today that 
Yesterday, Putz walked one batter, allowed two hits and one run and got just one out before being removed from the game.
General Motors (
Out on DVD is Philippe Garrel X 2 (
One of the relatively unsung features of the upcoming 
“In the coming week there will be much debate over what it means for the United States to be the majority shareholder of a major car company. Much of that debate will be serious and interesting. There will also be a lot of talk about how the United States is a socialist country in which the government has nationalized half of what was formerly known as private industry. 
General Motors is bankrupt. Whoops. It was probably going to happen no matter what, but lots of people hoped that bankruptcy would remain a threat that would encourage everyone to band together to save the company. Who is to blame for the death of the American auto industry?
Hey, we read 
This weekend I saw Up, a frightening animated picture from the folks at Pixar about a creepy old man who uses his bright and shiny balloons to spirit a confused and lonely boy away from his mother to a barely inhabited spot in South America where extradition is nearly impossible.
As seen on 
Since my previous release (v0.1) of 
Bookmark this on Delicious
