Memphisd
commented
3 weeks ago For the first issue I responded in #152. Essentially lets restrict in both cases to a single execution of the estimators (removing the min) and increasing the tolerance to a higher value (I tested with 1.2, worked in 100 trials). We should include a comment that the higher tolerance compensates for the variance caused by the probabilistic nature of beullens runtime computation.
For the second issue:
For 1. maybe @FloydZ can give some feedback on this? I am not sure I yet understand the issue correctly.
For 2. I again suggest removing the correction factor and instead increasing the tolerance. I tested with tolerance equal to 1 and had no issues.
FloydZ
Dioprz
sonarcloud[bot]
Description
Help required
The migration of the LEEStimator has multiple issues for which I need help: @Javierverbel @Memphisd @FloydZ.
I will describe them in the same order I suggest you to look at them: by commit.
Note: Take in count that the LEEstimator have two testing files to migrate:
First commit: Migration of
test_le_beullens.sage. Here we have two issues:internal_estimator(calledt1there) isinfor not in the new framework. For this reason I only check the external estimator, as can be seen here: https://github.com/Crypto-TII/CryptographicEstimators/blob/611a0b9b7fbcedd898e48c4defbee30f659705f8/tests/external_estimators/ext_le.sage#L33-L41 While this doesn't broke anything, it is still a change so let me know if there is a better fix..yamlfiles:good_kat.yaml,failing_kat.yamlandfailing_kat_2.yaml. As expected, the prior have values matching the ones calculated by the internal estimator, but the second and third ones don't. (Possibly related: https://github.com/Crypto-TII/CryptographicEstimators/issues/152). If you want to test them, just change the name of any of them bykat.yamland runmake docker-runfollowed bypytest tests/test_kat.py.Second commit: Migration of
test_le_bbps.sage. Here we another two issues:1.
As discussed in some point, the "load" function used in Sage pollutes the namespace of the current execution. That means that if there are any function defined in the current execution namespace, and that same function is defined in a "loaded" file, the loaded function will override the first one.
This behavior is causing that when these two files are loaded in the same file
The
bbpskat generators doesn't work and complains about some "zero division". However, if you remove theattack_cost.sageload, then they works as expected (but beulleans doesn't). I have tested swapping its loading order and the result is the same.This means that
attack_cost.sagehas some function used by some kat generator of the bbps functions, that isn't defined incost.sage, and overwrites the default implementation thatbbpsfunctions expects.Probably the easier solution is to just find that function of
attack_cost.sage, and change the name by something else (ex.log->log_) and propagate the change where is appropriate.2. At https://github.com/Crypto-TII/CryptographicEstimators/blob/2c97a0ed6c926af8e3481872c46ceb034cad87d4/tests/external_estimators/LEEstimator/test_le_bbps.sage#L43-L63 we are correcting the value produced by the external estimator using the internal one. This is not supported at the moment, so we need to solve that too.
Testing tips
ext_sdfq -> _ext_sdfqandext_sd -> _ext_sd. This way the commandmake docker-generate katonly generate the kats ofext_le. Otherwise you will have to wait more time.BBPSand then runningmake docker-generate-kat, and making the same by commenting theBEULLENSsection too.Review process
Not ready yet.
Pre-approval checklist