Closed RandallSteck closed 4 months ago
Thanks! I will take a look at this later today and report back.
This all looks great! Could you re-open it against the libacvp_2_1_0-throttle branch? Will approve it there.
Thank you for the contributions! 😄 Its also very nice to know libacvp is being tested on such a wide variety of platforms elsewhere too.
Andrew
Closing this pull request to reissue against 2_1_0-throttle.
Many fixes/cleanup to comments, extra CRLFs, and some to data output
I have addressed several data validation issues where libacvp differed from ACVP protocol min/max values. I've altered max values for some buffers/arrays (e.g. filename max length to better accommodate windows long filenames) and increased some fields marked "arbitrary" where they length proved insufficient in some cases.
I have been compiling libacvp on a wide variety of systems, include small/embedded/IOT, and along the way I've made minor fixes to address compiler warnings and hardware/compiler "quirks". One operational environment failed Hash/Shake MCT testing, and it is due to a compiler's misinterpretation of the Rightmost Outer Bits (ROB) calculation. I've change it to be explicit and avoid "pointer math".
I'm happy to change/amend as needed. I have more "feature" oriented changes in future pull requests.