security/testssl.sh: Fix build on !amd64

[tobik]

Summary

Only use enable-ec_nistp_64_gcc_128 on amd64 since it requires a
64-bit little-endian architecture and fix the build on !amd64

ecp_nistp224.c:43:9: error: unknown type name '__uint128_t'
typedef __uint128_t uint128_t;	/* nonstandard; implemented by gcc on 64-bit
	^

PR:		220403
Reported by:	[e-mail]
Approved by:	??? (mentor)
Differential Revision:	https://reviews.freebsd.org/D11436
MFH:		2017Q3

Test Plan

poudriere testport 10.3/i386, 11.0/i386, 11.0/amd64 ok

[mat]

Maybe it could be conditional on it being i386 ?

[tobik]

Maybe it could be conditional on it being i386 ?

Sure, I have copied security/openssl's approach now.

[mat]

security/testssl.sh/Makefile

53–59 ↗

(On Diff #30352)

Mmmm, this is absolutely not the same thing as "does not work on i386". Also, it makes it possible to disable it on amd64.

I was more thinking in the way of

.if ${ARCH} == "i386"
CONFIGURE_ARGS+= blah
.else
CONFIGURE_ARGS+= blurb
.endif

[tobik]

security/testssl.sh/Makefile

53–59 ↗

(On Diff #30352)

The real problem is that enable-ec_nistp_64_gcc_128 requires a 64-bit
little-endian architecture, so it'll likely break on powerpc, armv6,
... too.

it makes it possible to disable it on amd64.

Sure, but this isn't a problem on amd64 and it'll still build fine
without it since it just disables an optimization.

But I can use something like this you think it's better than using
options:

.if ${ARCH} == "amd64"
CONFIGURE_ARGS+=	enable-ec_nistp_64_gcc_128
.else
CONFIGURE_ARGS+=	no-ec_nistp_64_gcc_128
.endif

[mat]

security/testssl.sh/Makefile

53–59 ↗

(On Diff #30352)

That looks better, yes. (But then again, the review was "fix on i386", it should be "fix on !amd64")

[tobik]

• Use .if ${ARCH} instead of options to only enable-ec_nistp_64_gcc_128 on amd64