Just a thought, but shouldn't FuzzyReader rather be a composite of io.Reader and io.ByteReader?

See here: https://github.com/glaslos/tlsh/wiki/API
There should be also a small snippet in the README demonstrating usage of the API.

Once the API is stable, create a PR for: https://github.com/trendmicro/tlsh#ports

TLSH_BUCKETS: determines using 128 or 256 buckets, more is better
TLSH_CHECKSUM_1B: determines checksum length, longer means less collision

Sample app should be able to calculate the distance between to hashes.

Not sure where the extra allocation in BenchmarkHash-4 is coming from.
benchcmp bench_1.8.3.test bench_1.9.test

benchmark                     old ns/op     new ns/op     delta
BenchmarkPearson-4            0.32          0.29          -9.38%
BenchmarkFillBuckets-4        7545          7416          -1.71%
BenchmarkQuartilePoints-4     2128          1792          -15.79%
BenchmarkHash-4               12653         12651         -0.02%
BenchmarkModDiff-4            0.29          0.29          +0.00%
BenchmarkDigestDistance-4     36.4          36.1          -0.82%
BenchmarkDiffTotal-4          44.7          44.3          -0.89%

benchmark                     old allocs     new allocs     delta
BenchmarkPearson-4            0              0              +0.00%
BenchmarkFillBuckets-4        3              3              +0.00%
BenchmarkQuartilePoints-4     0              0              +0.00%
BenchmarkHash-4               7              8              +14.29%
BenchmarkModDiff-4            0              0              +0.00%
BenchmarkDigestDistance-4     0              0              +0.00%
BenchmarkDiffTotal-4          0              0              +0.00%

benchmark                     old bytes     new bytes     delta
BenchmarkPearson-4            0             0             +0.00%
BenchmarkFillBuckets-4        4197          4197          +0.00%
BenchmarkQuartilePoints-4     0             0             +0.00%
BenchmarkHash-4               4317          4376          +1.37%
BenchmarkModDiff-4            0             0             +0.00%
BenchmarkDigestDistance-4     0             0             +0.00%
BenchmarkDiffTotal-4          0             0             +0.00%

Hi

this is great - I added a link to it.

But you have made the license MIT - which is not actually allowed under the license of TLSH.
Could you make the LICENSE.txt file the same as
https://github.com/trendmicro/tlsh/blob/master/LICENSE

Cheers
jono

• Comparison with C implementation
• How to use the package
• Some Performance numbers
• Comparison with SSDEEP

The current API looks like it takes a reader and reads the entire file by itself. This is not efficient when compared to other hashes in Go which offer a Writer interface. The advantage with a Writer interface is that the caller can supply repeated buffers to multiple hashes at the same time thereby allowing for a single read of the file, and fanning out to multiple hashes. This is most efficient for slow readers (e.g. zip file members etc) which repeated reading of the same file is too expensive.

Please consider to refactor the API to make it more similar to existing hashes to comply with the standard Hash interface:

https://pkg.go.dev/hash#Hash

I am not that familiar with this hash or if this is even possible for this specific algorithm so please feel free to dismiss this FR :-)

Increase test coverage >95%

https://www.cs.toronto.edu/~norouzi/research/papers/multi_index_hashing.pdf

Minimal data length for TLSH calculation is 256bytes.

Hi

I noted that this repo accidently lost the NOTICE.txt file
Keeping this is a requirement of the Apache license

https://github.com/trendmicro/tlsh/blob/master/NOTICE.txt

Cheers
jono