In v0.8 and prior, the expose-internals feature, when enabled, would export several low-level functions that generally shouldn't be used directly. Keeping those functions behind an "I know what I'm doing" feature flag was IMO a good compromise, since it gave the crate secure defaults but also let users who really need those functions access them without patching the crate.

Unfortunately, with v0.9 and #304, those functions were moved to various other places and hidden behind multiple layers of pub(crate), meaning downstream users of rsa have no way at all to use them anymore (even though this commit message acknowledges that "it is expected that the functions inside algorithms crates might be useful (and used) by other parties"). The expose-internals feature flag was removed too, but none of this was documented in CHANGELOG.md.

I now find myself unable to upgrade one of my projects from v0.8 to v0.9 because it relies on the raw encrypt() and decrypt() functions. The project (which unfortunately isn't open source, or I'd link to it) decrypts firmware images for a commercial piece of hardware, the format of which I reverse engineered. As such, I have no control over what crypto operations it uses: the encryption scheme was defined by the vendor, and my code needs to implement it as is—bad choices and all. One of the bad choices the vendor made was to encrypt blocks of data with no padding whatsoever, which is why I need encrypt().

Would you accept a PR to bring expose-internals back? The functions themselves don't have to move back—internals could just re-export them from their new locations, and we could remove pub(crate) selectively on only the ones that make up the internal API. I personally only need encrypt() and decrypt(), so I'd be fine leaving the rest hidden unless/until someone demonstrates a compelling need for them.