Test kem interface against KATs #38
Conversation
src/lib.rs
Outdated
|
|
||
| std::thread::Builder::new() | ||
| .stack_size(10 * 1024 * 1024) | ||
| .spawn(|| { |
There was a problem hiding this comment.
concerning the thread stack size increase, I attempted to write a small wrapper function using generics in order to just wrap the three KEM function calls and not the other code, however could not get it to work out, maybe you have some suggestions to fix that:
fn increase_stack_size_wrapper<F: Fn() -> R + Send>(function: F) {
std::thread::Builder::new()
.stack_size(10 * 1024 * 1024)
.spawn(|| {
function
})
.unwrap()
.join()
.unwrap();
}To provide a bit more context: the problem seems to be with setting the Result type of the generic function, since it needs to implement Box<dyn error::Error>> aswell as Send + Sync 🤔
Colfenor
changed the title
…nor/classic-mceliece into feature/test-kem-interface-vs-KATs
• dependencies are imported often w.r.t. some feature guards • the use of dependency functionality is also feature guarded • these should be the same set of feature guards
The updated version exposes the same API and runs the same set of tests,
but the tests have been moved to separate modules.
According to semver-checks, no API breakage was found after running:
cargo semver-checks --only-explicit-features --baseline-rev main
--baseline-features mceliece348864
--current-features mceliece348864
Sadly, we cannot move these tests into the tests/ directory,
because the public API does not expose functionality like "fn verify".
As such, we cannot use verify from files in tests/.
Even if we migrate "fn verify" to tests/, there is still AesState.
Since AesState is used in so many place, we can only migrate
the tests to tests/ by code duplication (which I do not want to do).
Furthermore:
• Address TestData with `TestData` after an `use` statement
instead of `crate::TestData` to make it easier to move TestData around
• a lot of feature guard fixes
• KATKEM tests pass for me with "ulimit -s 12048",
but if you additionally pass feature="alloc" now,
the public key will end up on the heap instead of the stack now.
• I ran all tests in all configurations and made sure no performance
regressions occur.
|
All my performance and correctness tests passed now. My final latest commit tries to move out test functionality from Recognize that
|
|
I found one more cleanup: if there is a module-level feature guard, I remove the corresponding local feature guards. This makes the source code less redundant. No semantic difference are given. |
Colfenor
force-pushed
the
feature/test-kem-interface-vs-KATs
branch
from
b293955 to
81cb9fc
Compare
Colfenor
force-pushed
the
feature/test-kem-interface-vs-KATs
branch
from
81cb9fc to
4045c08
Compare
No worries, I'm glad I could find it :) |
…nor/classic-mceliece into feature/test-kem-interface-vs-KATs
this PR aims to close #37
kemfeature inside the already existing functions of creating & verifying request/response files, augmenting theTestcasestruct, without duplicating much code.The consequences are that the test function needs to run with increased stack size & that I had to refactor out the
katkem.rsfunctionality intolib.rsplus change the bash script which calls the test functionsI'm not quite sure if this is a great choice, however I couldn't achieve the desired
kemtesting otherwise since it's functions are private and feature restricted inlib.rsTODO: update readme md5 hashes, and section about katkem