| Message ID | patch-1.1-1f4e39be97b-20221020T225305Z-avarab@gmail.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | Makefile: use sha1collisiondetection by default on OSX and Darwin | expand |
Ævar Arnfjörð Bjarmason <avarab@gmail.com> writes: > Junio: I see in the meantime you've queued your own > https://lore.kernel.org/git/cover-v3-0.9-00000000000-20221020T223946Z-avarab@gmail.com/; > which is currently in "seen". Yes, as I said, I intend to merge it to 'next' in tomorrow's pushout, and then fast track all three topics (Peff's "-O0", Eric/Ævar's "use git_SHA_CTX abstraction", and "osx-clan uses sha1dc") down to 'master'. As you chose to make this topic hostage to the other multi-part topic, which is likely to be slowed down and require rerolls for typofixes and possible bikeshedding, by the time this topic becomes ready, it is likely that it would already be in 'master' and you'd have to rebase on that. Isn't this step of much higher importance than the other multi-part topic? I do not see why you chose to take it a hostage to the other one. Let's all learn to give priorities to produce sufficiently focused fixes that also sufficiently cover important issues. Frills and niceties can come on top later.
On Thu, Oct 20 2022, Junio C Hamano wrote: > Ævar Arnfjörð Bjarmason <avarab@gmail.com> writes: > >> Junio: I see in the meantime you've queued your own >> https://lore.kernel.org/git/cover-v3-0.9-00000000000-20221020T223946Z-avarab@gmail.com/; >> which is currently in "seen". [B.t.w. that link is wrong, I meant https://lore.kernel.org/git/xmqq35bitooc.fsf@gitster.g/] > Yes, as I said, I intend to merge it to 'next' in tomorrow's > pushout, and then fast track all three topics (Peff's "-O0", > Eric/Ævar's "use git_SHA_CTX abstraction", and "osx-clan uses > sha1dc") down to 'master'. As you chose to make this topic hostage > to the other multi-part topic, which is likely to be slowed down and > require rerolls for typofixes and possible bikeshedding, by the time > this topic becomes ready, it is likely that it would already be in > 'master' and you'd have to rebase on that. > > Isn't this step of much higher importance than the other multi-part > topic? I do not see why you chose to take it a hostage to the other > one. Let's all learn to give priorities to produce sufficiently > focused fixes that also sufficiently cover important issues. Frills > and niceties can come on top later. I don't see why we're in any rush to get this change down to "master" (either this [1], or the base [2]), nor your jc/ci-osx-with-sha1dc[3]. Your list in [4] has two fixes for issues on "master" which we should get there sooner than later. But the 3rd is just addressing a CI blind spot that we've had for at least 2 years, and since 2017 if we're talking about the default SHA-1 backend on OSX. Yes, we had some unportable code sneak in recently, but there's no reason I can see for why we'd expect that to happen tomorrow. It's been one such issue from 2017 until now, so at this rate we should expect the next one in 2027, not next week :) In any case, I figured you might want to fast-track it anyway, or whatever, which is why I crafted this series to give you easy options given your [4]. Namely (and in no particular order): A. You can queue the base topic[2] and this [1] on top of "seen" and your jc/ci-osx-with-sha1dc and they won't conflict. B. This applies directly on "master" if you peel off the first hunk. If you wanted this faster than the "doc" change it could be queued like that, and we could fix the docs later. C. You could eject your [3] and we could let this simmer at the normal rate, but that's assuming the "no rush" outlined above. D. You could proceed with your [3], and I can eventually rebase on top of it (we'll probably want to undo the nothing-to-do-with-SHA-1 change, presumably?) E. You don't need to pick this up at all at this time. Nothing's broken now that hasn't been broken for the years by us not using DC_SHA1 on OSX by default. 1. https://lore.kernel.org/git/patch-1.1-1f4e39be97b-20221020T225305Z-avarab@gmail.com/ 2. https://lore.kernel.org/git/cover-v3-0.9-00000000000-20221020T223946Z-avarab@gmail.com/ 3. https://lore.kernel.org/git/xmqq35bitooc.fsf@gitster.g/ 4. https://lore.kernel.org/git/xmqqtu3yqhm2.fsf@gitster.g/
diff --git a/Makefile b/Makefile index 36d6bffd1f1..fb4f240e28b 100644 --- a/Makefile +++ b/Makefile @@ -529,10 +529,8 @@ include shared.mak # Define BLK_SHA1 to make use of optimized C SHA-1 routines bundled # with git (in the block-sha1/ directory). # -# Define NO_APPLE_COMMON_CRYPTO on OSX to opt-out of using the -# "APPLE_COMMON_CRYPTO" backend for SHA-1, which is currently the -# default on that OS. We'll define NO_APPLE_COMMON_CRYPTO on Mac OS -# 10.4 or older ("Tiger", released in early 2005). +# Define APPLE_COMMON_CRYPTO_SHA1 to use Apple's CommonCrypto for +# SHA-1. # # === SHA-256 backend === # @@ -1873,7 +1871,7 @@ ifdef NO_POSIX_GOODIES BASIC_CFLAGS += -DNO_POSIX_GOODIES endif -ifdef APPLE_COMMON_CRYPTO +ifdef APPLE_COMMON_CRYPTO_SHA1 # Apple CommonCrypto requires chunking SHA1_MAX_BLOCK_SIZE = 1024L*1024L*1024L endif @@ -1890,7 +1888,7 @@ ifdef BLK_SHA1 LIB_OBJS += block-sha1/sha1.o BASIC_CFLAGS += -DSHA1_BLK else -ifdef APPLE_COMMON_CRYPTO +ifdef APPLE_COMMON_CRYPTO_SHA1 COMPAT_CFLAGS += -DCOMMON_DIGEST_FOR_OPENSSL BASIC_CFLAGS += -DSHA1_APPLE else diff --git a/ci/lib.sh b/ci/lib.sh index 1b0cc2b57db..fda7e008f26 100755 --- a/ci/lib.sh +++ b/ci/lib.sh @@ -264,6 +264,9 @@ macos-latest) esac case "$jobname" in +osx-clang) + MAKEFLAGS="$MAKEFLAGS APPLE_COMMON_CRYPTO_SHA1=Yes" + ;; linux32) CC=gcc ;;
When the sha1collisiondetection library was added and made the default in [1] the interaction with APPLE_COMMON_CRYPTO added in [2] and [3] seems to have been missed. On modern OSX and Darwin we are able to use Apple's CommonCrypto both for SHA-1, and as a generic (but partial) OpenSSL replacement. This left OSX and Darwin without protection against the SHAttered attack when building Git in its default configuration. Let's also use sha1collisiondetection on OSX, to do so we'll need to split up the "APPLE_COMMON_CRYPTO" flag into that flag and a new "APPLE_COMMON_CRYPTO_SHA1". Let's also change the CI for "osx-clang" to test with the new APPLE_COMMON_CRYPTO_SHA1 knob ("osx-gcc" uses the new sha1collisiondetection default). In practice this will spot issues like the one noted in [4], as testing with just two backends should be enough to spot unportable code. Ideally we'd have other CI jobs to test the various SHA-1 combinations, but for now we have better CI coverage than before. 1. e6b07da2780 (Makefile: make DC_SHA1 the default, 2017-03-17) 2. 4dcd7732db0 (Makefile: add support for Apple CommonCrypto facility, 2013-05-19) 3. 61067954ce1 (cache.h: eliminate SHA-1 deprecation warnings on Mac OS X, 2013-05-19) 4. 32205655dc7 (fsmonitor OSX: compile with DC_SHA1=YesPlease, 2022-10-19) Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> --- The 1st hunk here depends on the "base" topic which prepares the docs for this small change: https://lore.kernel.org/git/cover-v3-0.9-00000000000-20221020T223946Z-avarab@gmail.com/ But otherwise this applies on "master". Junio: I see in the meantime you've queued your own https://lore.kernel.org/git/cover-v3-0.9-00000000000-20221020T223946Z-avarab@gmail.com/; which is currently in "seen". That patch will merge smoothly with this one, both textually and semantically, but if we have this it's all that's needed to flip the default and give us pretty much the same OSX CI coverage. "Pretty much" because an unstated effect of your patch is to disable all use of the "Apple Common Crypto" library on "osx-clang", which includes (but is not limited to) giving us another SHA-1 backend. Range-diff: 1: 392fabdb456 < -: ----------- fsmonitor OSX: compile with DC_SHA1=YesPlease 2: 7ae22276aa7 < -: ----------- Makefile: create and use sections for "define" flag listing 3: 78ef8636c57 < -: ----------- Makefile: really use and document sha1collisiondetection by default 4: f1fb9775b33 < -: ----------- Makefile: rephrase the discussion of *_SHA1 knobs -: ----------- > 1: 1f4e39be97b Makefile: use sha1collisiondetection by default on OSX and Darwin Makefile | 10 ++++------ ci/lib.sh | 3 +++ 2 files changed, 7 insertions(+), 6 deletions(-)