test: fix session-ticket, non-blocking-io tests on 32 bit

[jmayclin]

This test fixes two session tests that were failing on 32 bit platforms. Session_ticket was failing because a time_t overflowed when converting from seconds to nanoseconds.

Non-blocking was failing because of different integer comparison logic on 32 bit vs 64 bit time_t.

Resolved issues:

Makes progress on #3948

Description of changes:

s2n_self_talk_nonblocking_test

This test was failing because of a broken macro. The macro had always been broken, but became differently broken on a 32 bit platform.

#define LESS_THAN_ELAPSED_SECONDS(start_time, max_time) ((start_time - time(0)) < max_time)

This macro is a little bit backwards, because it is small_number - big_number, which results in the left side of the comparison always being negative.

The left side of the comparison is time_t types and the right side (max_time) is uint32_t.

On a 64 bit platform, time_t is an int64_t which can represent the entire range of uint32_t. During comparison uint32_t is safely upsized so that time_t:-1 < uint32_t:300 evaluates to true.

On a 32 bit platform, time_t is normally an int32_t. Comparison between int32_t and uint_32t results in the int32_t getting "upgraded" to an unsigned integer type, so the negative number becomes a very big number, resulting in the condition evaluating to false. This leads the client to an early exit, which results in the test failing.

s2n_session_ticket_test

There was a current time unix_seconds time_t that we tried to convert to nanoseconds before upsizing to 64 bits. The fix was to cast to an int64_t before converting to nanoseconds.

Testing:

Tested locally with the following

cmake . \
	-Bbuild \
	-DCMAKE_BUILD_TYPE=DEBUG \
	-DCMAKE_TOOLCHAIN_FILE=cmake/toolchains/32-bit.toolchain

cmake --build ./build -j $(nproc)

CTEST_PARALLEL_LEVEL=$(nproc) make -C build test

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

[jmayclin]

This PR has gotten a bit large to review as a single unit, so I'll be closing it and splitting up the changes into smaller PRs. Edit: wrong PR, this is already the refactored PR. 😄

[jmayclin]

Waiting for #3977 to rebase and merge this, since it will validate the 32 bit fixes in CI.

[jmayclin]

Hmm, looks like the Nix Integration tests failed despite all of the normal integration tests passing? Generally I've only seen the nix integration tests gives false positives (the tests reported a failure but nothing was wrong), so I don't feel any immediate need to revert this. But we should look into this next week and confirm.

[jmayclin]

Even in the case that the nix tests are actually failing, I don't think we'd revert this commit. This commit's impact is very strictly limited to unit tests files, and I have high certainty that it does not touch any of the code that gets exercised in the Integration Tests.