SheetJS
8215562b11
- AMD support (h/t @lmk123 for initially contributing to js-crc32) - added missing bitshift (h/t @florentbr for initially contributing to js-crc32) - flow annotaations - updated ci node versions |
||
---|---|---|
bits | ||
ctest | ||
misc | ||
perf | ||
test_files | ||
.flowconfig | ||
.gitignore | ||
.jscs.json | ||
.jshintrc | ||
.npmignore | ||
.travis.yml | ||
adler32.flow.js | ||
adler32.js | ||
LICENSE | ||
Makefile | ||
package.json | ||
README.md | ||
test.js |
adler32
Signed ADLER-32 algorithm implementation in JS (for the browser and nodejs). Emphasis on correctness and performance.
Installation
In nodejs:
npm install adler-32
In the browser:
<script lang="javascript" src="adler32.js"></script>
The browser exposes a variable ADLER32
Usage
-
ADLER32.buf(byte array or buffer)
assumes the argument is a set of 8 bit unsigned integers (e.g. nodejsBuffer
or simple array of ints) -
ADLER32.bstr(binary string)
interprets the argument as a binary string where thei
-th byte isstr.charCodeAt(i)
-
ADLER32.str(string)
interprets the argument as a standard JS string
Testing
make test
will run the nodejs-based test. To run the in-browser tests, run a
local server and go to the ctest
directory. To update the browser artifacts,
run make ctest
.
To generate the bits file, use the adler32
function from python zlib:
>>> from zlib import adler32
>>> x="foo bar baz٪☃🍣"
>>> adler32(x)
1543572022
>>> adler32(x+x)
-2076896149
>>> adler32(x+x+x)
2023497376
Performance
make perf
will run algorithmic performance tests (which should justify certain
decisions in the code).
js-crc has more performance notes
Bit twiddling is much faster than taking the mod on Safari and older Firefoxes.
Instead of taking the literal mod 65521, it is faster to keep it in the integers
by bit-shifting: 65536 ~ 15 mod 65521
so for nonnegative integer a
:
a = (a >>> 16) * 65536 + (a & 65535) [equality]
a ~ (a >>> 16) * 15 + (a & 65535) mod 65521
The mod is taken at the very end, since the intermediate result may exceed 65521
Magic Number
The magic numbers were chosen so as to not overflow a 31-bit integer:
F[n_] := Reduce[x*(x + 1)*n/2 + (x + 1)*(65521) < (2^31 - 1) && x > 0, x, Integers]
F[255] (* bstr: x \[Element] Integers && 1 <= x <= 3854 *)
F[127] (* ascii: x \[Element] Integers && 1 <= x <= 5321 *)
Subtract up to 4 elements for the unicode case.
License
Please consult the attached LICENSE file for details. All rights not explicitly granted by the Apache 2.0 license are reserved by the Original Author.