| 1 |
Puff -- A Simple Inflate |
| 2 |
3 Mar 2003 |
| 3 |
Mark Adler |
| 4 |
madler@alumni.caltech.edu |
| 5 |
|
| 6 |
What this is -- |
| 7 |
|
| 8 |
puff.c provides the routine puff() to decompress the deflate data format. It |
| 9 |
does so more slowly than zlib, but the code is about one-fifth the size of the |
| 10 |
inflate code in zlib, and written to be very easy to read. |
| 11 |
|
| 12 |
Why I wrote this -- |
| 13 |
|
| 14 |
puff.c was written to document the deflate format unambiguously, by virtue of |
| 15 |
being working C code. It is meant to supplement RFC 1951, which formally |
| 16 |
describes the deflate format. I have received many questions on details of the |
| 17 |
deflate format, and I hope that reading this code will answer those questions. |
| 18 |
puff.c is heavily commented with details of the deflate format, especially |
| 19 |
those little nooks and cranies of the format that might not be obvious from a |
| 20 |
specification. |
| 21 |
|
| 22 |
puff.c may also be useful in applications where code size or memory usage is a |
| 23 |
very limited resource, and speed is not as important. |
| 24 |
|
| 25 |
How to use it -- |
| 26 |
|
| 27 |
Well, most likely you should just be reading puff.c and using zlib for actual |
| 28 |
applications, but if you must ... |
| 29 |
|
| 30 |
Include puff.h in your code, which provides this prototype: |
| 31 |
|
| 32 |
int puff(unsigned char *dest, /* pointer to destination pointer */ |
| 33 |
unsigned long *destlen, /* amount of output space */ |
| 34 |
unsigned char *source, /* pointer to source data pointer */ |
| 35 |
unsigned long *sourcelen); /* amount of input available */ |
| 36 |
|
| 37 |
Then you can call puff() to decompress a deflate stream that is in memory in |
| 38 |
its entirety at source, to a sufficiently sized block of memory for the |
| 39 |
decompressed data at dest. puff() is the only external symbol in puff.c The |
| 40 |
only C library functions that puff.c needs are setjmp() and longjmp(), which |
| 41 |
are used to simplify error checking in the code to improve readabilty. puff.c |
| 42 |
does no memory allocation, and uses less than 2K bytes off of the stack. |
| 43 |
|
| 44 |
If destlen is not enough space for the uncompressed data, then inflate will |
| 45 |
return an error without writing more than destlen bytes. Note that this means |
| 46 |
that in order to decompress the deflate data successfully, you need to know |
| 47 |
the size of the uncompressed data ahead of time. |
| 48 |
|
| 49 |
If needed, puff() can determine the size of the uncompressed data with no |
| 50 |
output space. This is done by passing dest equal to (unsigned char *)0. Then |
| 51 |
the initial value of *destlen is ignored and *destlen is set to the length of |
| 52 |
the uncompressed data. So if the size of the uncompressed data is not known, |
| 53 |
then two passes of puff() can be used--first to determine the size, and second |
| 54 |
to do the actual inflation after allocating the appropriate memory. Not |
| 55 |
pretty, but it works. (This is one of the reasons you should be using zlib.) |
| 56 |
|
| 57 |
The deflate format is self-terminating. If the deflate stream does not end |
| 58 |
in *sourcelen bytes, puff() will return an error without reading at or past |
| 59 |
endsource. |
| 60 |
|
| 61 |
On return, *sourcelen is updated to the amount of input data consumed, and |
| 62 |
*destlen is updated to the size of the uncompressed data. See the comments |
| 63 |
in puff.c for the possible return codes for puff(). |
