Test failures on big endian architecture.

[xnox]

s390x is 64bit big endian.
test.txt

[pmienk]

Assume fixed by #445, thanks!

[xnox]

Nope, it is not =) #445 is compile failure. This is the test-suite failure that one observes after applying #445 fix.

[narodnik]

Is it possible to get safe access to a VPS with this arch to fix this issue? This looks like an endianness issue. See the errors like [7f8000 != ff8000] and [817f != 017f] Thanks.

[pmienk]

I assume it's in large part due to a lack of conditionality in the endian
functions. I didn't see any simple way to get travis-ci to cover big
endian systems and don't have one myself. Was hoping to make a checkin in
the next day or two and see if it addresses the issue.

On Sun, Aug 21, 2016, 1:19 AM RojavaCrypto notifications@github.com wrote:

Is it possible to get safe access to a VPS with this arch to fix this
issue? This looks like an endianness issue. See the errors like [7f8000
!= ff8000] and [817f != 017f] Thanks.

—
You are receiving this because you modified the open/close state.
Reply to this email directly, view it on GitHub
#446 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AGL5OG1Ca46M74ox5N9QXMBDrpbmeJYwks5qiAn9gaJpZM4Jof3Y
.

[xnox]

@RojavaCrypto i am not aware of any public access to any big endian systems. E.g. powerpc (32bit, big endian) or the s390x (64bit, big endian). Whilst arm64 is wide spread, and hardware has support for arm64be - nobody has a linux port for arm64be, everyone does little endian only.

[pmienk]

It appears I was wrong as to the involvement of the to_ and from_ endian functions as for the most part they rely on bit shift operators which should provide platform independence. I've corrected what looks to be an oversight I made there w.r.t. endianness, but doubt it has much to do with the failures seen.

[evoskuil]

Looking at the test failures, these are the most interesting:

test/utility/endian.cpp(98): fatal error: in "endian_tests/from_little_endian_stream_unsafe_eof_stream_partial_read": critical check expected == value has failed
test/utility/endian.cpp(128): fatal error: in "endian_tests/from_little_endian_stream_unsafe_valid": critical check expected == value has failed

Notice that all of the BE tests succeed, as do some of the LE tests. Unfortunately Boolean boost macros are used in these two tests, instead of integer value macros, so we can't see detail of the failures. I've created a PR to address this issue and better factor and standardize the endianness tests.

One of the three LE stream tests succeeds but it is a negative test with no value comparison - the output of its from_little_endian_stream_unsafe call is unused. Therefore the common denominator of the endian test failures is from_little_endian_stream_unsafe on BE architecture. Given that LE streaming is used extensively it could be the cause of some (or all) of the other failures.

[evoskuil]

Looking at recent history, it looks like the immediately-above issue has been addressed by 8c89522. This should resolve the two endian test failures and certainly a number of others. @xnox could you please re-execute the test pass with latest code?

[evoskuil]

Based on the objectives of simplicity and full test coverage, the SHA1 and RIPEMD160 implementations have been replaced with architecture-independent implementations. Given that tests in these implementations were failing in this platform, this should also provide resolution for that part of the issue.

#462

[evoskuil]

A problem of invalid (overflowing) test data in the script_number implementation has been resolved by #459. The implementation has been modified to handle the full int64_t domain and to throw when that domain is over/under-flowed. The previous behavior was assertion-based and is documented to vary by architecture. It is likely that a large number of test failures resulted from this issue on this platform.

[evoskuil]

There are no known outstanding test failures or identified issues pertaining to endianness. Given that we do not have a BE test platform we await feedback.