ssh_decode_sequence_multi & compiler error

[karel-m]

I have got this report:

cc ... -c ltc/misc/ssh/ssh_decode_sequence_multi.c -o ltc/misc/ssh/ssh_decode_sequence_multi.o
{standard input}: Assembler messages:
{standard input}:100: Error: Incorrect register `%rbx' used with `l' suffix
{standard input}:101: Error: Incorrect register `%rbx' used with `l' suffix
*** [ltc/misc/ssh/ssh_decode_sequence_multi.o] Error code 1

It is probably: gcc 4.2.1 20070831 patched [FreeBSD]

Full report http://www.cpantesters.org/cpan/report/341fe940-d829-11e8-9067-b381f67cb40a

And ... I have absolutely no idea what's going on.

Cc: @rmw42

[rmw42]

Well, since we don't have any assembly language in there, that's gotta be a compiler bug...

[sjaeckel]

Could it be related to the invocation of LOAD {32,64}H() while dereferencing a pointer?

That's the only part of the code which pulls in assembly

[rmw42]

Ah, hadn't seen that... it's nothing like the error, though:

#define LOAD64H(x, y)          \
asm __volatile__ (             \
   "movq (%1),%0\n\t"          \
   "bswapq %0\n\t"             \
   :"=r"(x): "r"(y): "memory");

[rmw42]

It also successfully compiled ciphers/camellia.c, encauth/gcm/gcm_gf_mult.c and hashes/sha2/sha512.c, which all use LOAD64H(), so I can't see it being the inline asm for that macro...

[rmw42]

I guess the only way we'll get to the bottom of this is if someone can find a FreeBSD box with that compiler on it. Replacing -o ltc/misc/ssh/ssh_decode_sequence_multi.o with -S -o ssh_decode_sequence_multi.s in the CC command will get the asm it was generating, and the issue is around line 100...

[sjaeckel]

I know that'd be only a temporary solution, but do you even provide an API in Perl for those ssh functions?
If not, couldn't you simply disable their compilation?

[karel-m]

Cc @eserte

Hi Slaven, we are discussing here a compiler error from your cpantester report related to my perl module CryptX. The error is basically a failure to compile one file from libtomcrypt library - see http://www.cpantesters.org/cpan/report/341fe940-d829-11e8-9067-b381f67cb40a

Is the compiler and/or the platform of this cpantester machine somehow unusual/special? Does the same cpantester machine complain/fail in a similar way with other libraries/modules?

[karel-m]

couldn't you simply disable their compilation?

The irony is that we hit here the reverse of #449 which can be called: "ecc_sign_hash_ex + ecc_verify_hash_ex forces linking SSH decode/encode code" 😄

So maybe after #450 or something like that I can simply exclude ssh_(en|de)code_sequence_multi.c

[eserte]

I checked on another (real, non-VM) freebsd9 machine, and the same compiler error happens here, too.
What works on both machines is to use the installed gcc 4.8 compiler:

perl Makefile.PL CC=gcc48

And I never saw this kind of error while compiling other stuff.

[karel-m]

Further investigation revealed that the trouble making line is:

LOAD32H(size, in);

plus the fact that size is unsigned long but should be ulong32.

So it is fixable, not a compiler bug.

[karel-m]

@eserte may I ask you for a quick check whether CryptX-0.061_003 doesn't fail anymore?

http://cpan.metacpan.org/authors/id/M/MI/MIK/CryptX-0.061_003.tar.gz

[karel-m]

All green - http://fast-matrix.cpantesters.org/?dist=CryptX%200.061_003

Closing.