Logger.moon caused RNG seed fixed to 0

[Lambholl]

When I was generating subtitles from script which contains math.random, I found that each time when I started up Aegisub, the random function always gives same results.
Then I found when I added math.randomseed(0) to script, the results are always the same as the first-time-run after startup, so it could be sure that RNG seed was set to 0 each time Aegisub starts up.
So the only one code that affects random seed could be found is in DependencyControl/Logger.moon:

unless seeded
    timer.sleep 10 for i=1,50
    math.randomseed(timer\timeElapsed!*1000000)
    math.random, math.random, math.random
    seeded = true
    -- timer gets freed on garbage collection
    timer = nil

So it could be preferred that timeElapsed always returns a 0 which caused the issue.

[FichteFoll]

Afaik, the random seed is reset by Aegisub on each karaoke templater invocation (potentially by executing it in an isolated environment?) and the fact that it always results in the same "randomness" is actually a feature and not a bug. If you want different results in your script, you need to change the seed yourself inside the script.

I theorize this because I have observed this functionality many years before Dependency Control was even created, however I do not have a definitive source (right now) as for whether this functionality is intended.

[Lambholl]

Afaik, the random seed is reset by Aegisub on each karaoke templater invocation (potentially by executing it in an isolated environment?) and the fact that it always results in the same "randomness" is actually a feature and not a bug. If you want different results in your script, you need to change the seed yourself inside the script.

I theorize this because I have observed this functionality many years before Dependency Control was even created, however I do not have a definitive source (right now) as for whether this functionality is intended.

I would like to consider this as a bug because using the current timestamp is a normal and intended way to generate a seed. The goal of the code in Logger.moon is explicitly to provide a non-deterministic, unique seed upon each startup to ensure different random sequences.

Which could prove this is the intended behavior and not just a workaround, the script executed math.random() three times immediately after generating the seed. This is a well-known practice called "warming up" the PRNG. Its sole purpose is to quickly improve the statistical quality of the subsequent random numbers after a fresh seed has been set. This practice only makes sense if the seed itself is expected to be different each time. If the environment were designed to always provide a fixed seed (like 0), then "warming up" the RNG would be entirely pointless, as the entire sequence, including the first few numbers, would be identical every time. The presence of this code is a strong indicator that the developer's intention was for the seeding process to be non-deterministic.

The theory that timer\timeElapsed() might be returning 0 is a very plausible root cause for the bug I was experiencing. If the high-resolution timer hasn't been initialized properly or starts counting from the exact point the Lua state is initialized, it could indeed return 0 or anything that could be transfered to 0 on its first call, making math.randomseed(0) a de facto outcome.

In conclusion, the explicit intent of the Dependency Control code is to provide a fresh seed. The fact that it currently fails to do so due to a faulty time measurement should be classified as a bug in the implementation, not an affirmation of an intended feature. The ideal fix would be to find a more reliable method of obtaining a unique seed value within the Aegisub environment.

I found changing this file could not affect the randomseed, so the real factor needs more researching.