Fuzzer not using seed corpus, likely needs to be named using the sha1 hash

[kevina]

I just downloaded a copy of corpus used by one of the fuzzes and no where in any of the inputs is there the string speeling and mistayks which leads me to believe that it is not using the seed corpus. I think this is because they need to be named using the sha1 hash.

@cmeister2 mostly FYI, I plan to fix this myself sometime Tuesday by using sha1sum on each file before creating the zip.

[cmeister2]

I'm reasonably certain that's not true. It's more likely that that input is not particularly interesting for the fuzzer and so it's used a different file which takes the same internal paths. There's a concept of "minimizing" corpus entries that the fuzzers do periodically; I could believe that these text strings have been minimized away.

What might be interesting instead is to supply a list of text strings as a .dict file, so the fuzzer is aware of those. Configuration option flags are probably good. I think all this needs is:

• a dictionary file named "<fuzz_target>.dict", each line containing a "string" entry (so, aspell_fuzzer.dict)
• possibly an "aspell_fuzzer.options" file, containing the following:

[libfuzzer]
dict = aspell_fuzzer.dict

[kevina]

I added some additional input which should of increased coverage, especially for the filters, but so far the coverage has not increased. It may be that I just need to give it another day. Any ideas?

[cmeister2]

Might help. Let's see where coverage is at tomorrow.

[kevina]

Upstream issue: google/oss-fuzz#2729