Support text-based encoding (e.g. Base64) for inputs files

[Marcono1234]

Currently the inputs files generated by Jazzer are binary. This can lead to issues when adding them to a Git repository (e.g. as regression test resources) because Git might erroneously consider them to be text and performs line ending conversion, see also #958 (comment).

Would it be possible to support text-based encoding such as Base64? Possibly as opt-in via a command line flag or a junit-platform.properties entry.
Maybe you could also use a file extension suffix such as .base64 (e.g. crash-abcdef.base64) so that Jazzer recognizes and decodes it automatically when reading the inputs (even if writing of Base64 encoded files is not opted-in).

The disadvantage would be that these files would be larger compared to the binary ones. And it might need custom post- and pre-processing when interacting with libFuzzer?

[fmeum]

If this requires explicit opt-in via a config file anyway, could a copy & pastable .gitattributes snippet work just as well?

[Marcono1234]

Fair point; if it was opt-in the users might notice it too late as well (once they already encountered issues with Git modifying the files), so it might not be much better than the .gitattributes approach.

#965 has added a note for .gitattributes to the README.
Maybe the pattern there (and in docs/dev-junit-implementation-details.md) could be more specific to not apply to all test resources¹, probably something like:

/src/test/resources/**/*Inputs/** binary

But I am not complete sure.

Footnotes

1. Was my fault though for not suggesting this more specific pattern from the beginning already. ↩