Potential bug in hash function

[georgeBacon23]

In the file TargetHashingClient.java, the function createHashForRule creates a hash for a rule by considering the rule's inputs and if necessary recursively hashing inputs that are rules. A hashmap ruleHashes is used to cache previously computed rule hashes. I think there is a bug here, as at least on my system, the MessageDigest object is reset when .digest() is called. So if the rule is in the cache, the first time it is found it will be reset when digest is called to update the parent's digest. Any other rules that use it will subsequently update with the wrong hash.

Am I missing something here?

[tinder-maxwellelliott]

@georgeBacon23 Thanks for reaching out. Do you have a demo project I can test this with? I have not seen this behavior myself.

Is this the line in question?

bazel-diff/src/main/java/com/bazel_diff/TargetHashingClient.java

Line 115 in 0f9f17d

digest.update(ruleInputDigest.digest());

? We are simply using ruleHashes to avoid recomputing rule hashes that have already been computed, as you can see on

bazel-diff/src/main/java/com/bazel_diff/TargetHashingClient.java

Line 99 in 0f9f17d

return existingMessage;

[georgeBacon23]

Hi, yes I believe when the method digest() is called on the MessageDigest object, the underlying bytes are reset. So next time it's retrieved you'll get a different result.

Sorry I don't have a project to hand that I can share. But I imagine any project with multiple targets which reference the same rule would show the issue. If you print the retrieved digest for the rule each time, it should have one value on the first call, then will be set to another for all subsequent calls.

The problem can be avoided if instead of caching the MessageDigest object you store a copy of the byte array, so something like:

ruleHashes.put(rule.getName(), digest.digest().clone())

Obviously you'd have to update method signatures.

[georgeBacon23]

I saw it by making a modification like below, where TEST_TARGET is one to look out for and bytesDigestToString is a helper function for printing the digest:

if (ruleInputDigest != null) {
    byte[] d = ruleInputDigest.digest();
    String name = inputRule.getName();
    if (name.equals("TEST_TARGET")){
        System.out.println("target, digest: " + name + " "+ bytesDigestToString(d));
    }
    digest.update(d);
}

Also you might see that a lot of targets have the same final hash.

[tinder-maxwellelliott]

I have reproduced this issue locally, getting a fix up now