Re: [PATCH 0/5] core: update our SHA-1 docs, use sha1collisiondetection on OSX too

On Fri, Apr 22 2022, Junio C Hamano wrote:

Ævar Arnfjörð Bjarmason  [off-list ref] writes:

quoted

I wasn't able to find any on-list references to it being intentional,
but it appears that while we made the sha1collisiondetection variant
of SHA-1 the default in early 2017 we've never updated the OSX builds
to do likewise.

I don't know what various git packages for OSX to, but our vanilla OSX
distribution definitely uses Apple Common Crypto, and won't detect the
https://shattered.io attack.

This series changes that, and while doing so in 2/5 updates our
documentation and Makefile interface for the SHA-1 selection. Our
INSTALL file was still claiming we used OpenSSL's SHA-1 by default.

Then since we'd made sha1collisiondetection the default we hadn't
changed the code's default fallback to be that, it was still
block-sha1. Now our fallback behavior is "error" instead, which makes
it less likely that we'll get some foot-gun like the "OSX not using
sha1collisiondetection" again.

The 4/5 and 5/5 then remove the PPC_SHA1 implementation. I submitted
this before as [1], and the range-diff is to that submission (it
wasn't picked up). I think it makes sense as part of this general
SHA-1 cleanup.

Thanks for this effort.

I'd like to see somebody with "building Git for distributing to
macOS" background to comment (I am assuming that the mailing list
git-packagers@googlegroups.com is the way to reach them).

*Bump* in case anyone there would like to chime on this bit of OSX
portability.

I also sent an off-list E-Mail to Tim Harper today, and addresses which
I gather (from some old git-security@ traffic) are involved in packaging
the Apple Git shipped with OSX itself.

In the meantime do you mind if this topic were queued up? If there are
any lingering portability concerns getting it into CI and exposed to
anyone else building on OSX would be a good thing.

I don't see any reason for why we'd decide that OSX out of all our
supported platforms should be the only one where we're not mitigating
the SHAttered attack (and similar future attacks) by default.

So aside from any finer details of OSX portability the direction here of
building with sha1collisiondetection by default on OSX by default seems
like a safe bet, just as we do on the rest of our (checks out
config.mak.uname) 20+ supported platforms (per `uname -s`).