Potential heap memory regression between 0.3 and 0.6

[kvark]

I'm looking at the heap memory regression in Gecko that happened upon updating parking_lot and rand crates. Parking lot doesn't appear to have any related changes, judging by Amanieu/parking_lot#143. This try CL appears to fix the regression in Gecko, which leaves me with Rand to look at. Any ideas about what could have caused the memory usage increase?

[dhardy]

It might be the RNG used by thread_rng, which in the current release is Hc128Rng, however IIRC the old Rand 0.3 used Isaac64Rng on 64-bit platforms, thus not much different unless you are talking about 32-bit here (see table).

It happens that we've already switched this algorithm out for ChaCha (a new implementation not quite the same as in that table, but still much smaller memory usage), thus Rand 0.7 may fix this (release is imminent, on the scale of a couple of weeks).

I suggest you wait until 0.7 is available and test again. If it is still a problem, then you'll have to be much more specific about where exactly the memory usage is coming from. The only code doing explicit allocations is (1) rand_core::Error, (2) the seq module and (3) a few less common distributions.

[dhardy]

The 0.7.0-pre.1 release should let you test this.

[dhardy]

Is this still an issue? Since you haven't replied in a while I'll close, but if it is I can re-open it.