threads / bug / 48457

bug: SHA1 calculation on PPC machines when built with gcc older than 4.6

Subject: bug: SHA1 calculation on PPC machines when built with gcc older than 4.6

## tl;dr

2 messages between May 10, 2018 and May 10, 2018.

replies: 1people: 2as markdown or json

Ken Cunningham· May 10, 2018, 18:11 UTC · lore
Some vintage Apple PPC machines build a non-funtional version of git as of git 13.1 when using the stock gcc compilers that are installed with the OS; the SHA1 calculations are faulty. This can be repaired with a simple patch (attached).
Stock vintage Apple PPC machines come with gcc-4.0 or gcc-4.2. On MacOS 10.4 and earlier, or when not using Apple Common Crypto on 10.5, git uses the SHA1 calculation code from here <https://github.com/cr-marcstevens/sha1collisiondetection>. The code in <https://github.com/cr-marcstevens/sha1collisiondetection/blob/master/lib/sha1.c> tries to detect all systems that are BIG_ENDIAN, but the above noted systems fall through because they fail the tests.
It appears that the primary test:
#if defined(__BYTE_ORDER__) && defined(__ORDER_BIG_ENDIAN__)
only works as of gcc-4.6 and newer, so the code is built as LITTLE_ENDIAN on PPC with older gcc versions.
Issue report:
<https://github.com/cr-marcstevens/sha1collisiondetection/issues/40>
MacPorts bug report:
<https://trac.macports.org/ticket/54602>
The included patch to git fixes the issue on our testing.
Thanks for git!
Ken Cunningham
=====
diff --git a/sha1dc/sha1.c b/sha1dc/sha1.c
index 25eded1..5faf5a5 100644
--- a/sha1dc/sha1.c
+++ b/sha1dc/sha1.c
@@ -92,6 +92,10 @@
  */
 #define SHA1DC_BIGENDIAN
 
+#elif (defined(__APPLE__) && defined(__BIG_ENDIAN__) && !defined(SHA1DC_BIGENDIAN))
+/* older gcc compilers which are the default  on Apple PPC do not define __BYTE_ORDER__ */
+#define SHA1DC_BIGENDIAN
+
 /* Not under GCC-alike or glibc or *BSD or newlib or <processor whitelist> */
 #elif defined(SHA1DC_ON_INTEL_LIKE_PROCESSOR)
 /*

==========
Ævar Arnfjörð Bjarmason· May 10, 2018, 18:30 UTC · re: Ken Cunningham · lore

Re: bug: SHA1 calculation on PPC machines when built with gcc older than 4.6

On Thu, May 10, 2018 at 8:11 PM, Ken Cunningham <ken.cunningham.webuse@gmail.com> wrote:

Show 48 quoted lines
> Some vintage Apple PPC machines build a non-funtional version of git as of git 13.1 when using the stock gcc compilers that are installed with the OS; the SHA1 calculations are faulty. This can be repaired with a simple patch (attached).
>
>
> Stock vintage Apple PPC machines come with gcc-4.0 or gcc-4.2. On MacOS 10.4 and earlier, or when not using Apple Common Crypto on 10.5, git uses the SHA1 calculation code from here <https://github.com/cr-marcstevens/sha1collisiondetection>. The code in <https://github.com/cr-marcstevens/sha1collisiondetection/blob/master/lib/sha1.c> tries to detect all systems that are BIG_ENDIAN, but the above noted systems fall through because they fail the tests.
>
> It appears that the primary test:
>
> #if defined(__BYTE_ORDER__) && defined(__ORDER_BIG_ENDIAN__)
>
> only works as of gcc-4.6 and newer, so the code is built as LITTLE_ENDIAN on PPC with older gcc versions.
>
>
> Issue report:
>
> <https://github.com/cr-marcstevens/sha1collisiondetection/issues/40>
>
>
> MacPorts bug report:
>
> <https://trac.macports.org/ticket/54602>
>
>
> The included patch to git fixes the issue on our testing.
>
> Thanks for git!
>
> Ken Cunningham
>
>
>
>
> =====
>
> diff --git a/sha1dc/sha1.c b/sha1dc/sha1.c
> index 25eded1..5faf5a5 100644
> --- a/sha1dc/sha1.c
> +++ b/sha1dc/sha1.c
> @@ -92,6 +92,10 @@
>   */
>  #define SHA1DC_BIGENDIAN
>
> +#elif (defined(__APPLE__) && defined(__BIG_ENDIAN__) && !defined(SHA1DC_BIGENDIAN))
> +/* older gcc compilers which are the default  on Apple PPC do not define __BYTE_ORDER__ */
> +#define SHA1DC_BIGENDIAN
> +
>  /* Not under GCC-alike or glibc or *BSD or newlib or <processor whitelist> */
>  #elif defined(SHA1DC_ON_INTEL_LIKE_PROCESSOR)
>  /*

Thanks. As noted in https://public-inbox.org/git/87603xxc3k.fsf@evledraar.gmail.com/ patches like this should be sent to the upstream, it appears you just opened an issue there, but sent the patch here. Could you open a PR with upstream with this patch?

← back to recent threads