memory leak

[DanGrayson]

This code

engineMemory = value Core#"private dictionary"#"engineMemory"
needsPackage "RandomCanonicalCurves"
ps = () -> run ( "ps u " | toString processID () )
leak = c -> (
    setRandomSeed "37862873";
    g := 9;
    p := 12347;
    numEx := 40;
    T := ZZ/p[t_0..t_(g-1)];
    for i from 1 to 100 do (
        IC := (random canonicalCurve)(g,T);
        minimalBetti IC;
    -- registerFinalizer (IC.cache.quotient,"--");
        if c then IC.cache.quotient = null;
        << "." << flush;);
    collectGarbage();
    print engineMemory();
    )
ps()
leak true
ps()
leak false
ps()

runnning under https://github.com/mikestillman/M2/tree/newres produces this totally perplexing result:

USER   PID  %CPU %MEM      VSZ    RSS   TT  STAT STARTED      TIME COMMAND
dan  75775   0.0  0.5  2917676  85244 s002  S+   10:39AM   9:50.42 /Users/dan/src/M2/M2-Macaula
....................................................................................................size of each slabs = 2040
total engine space allocated = 0k
           stash    k total   k in use       size     nalloc      inuse  highwater      freed
Finalizations of new resolutions:
# of res objects constructed/deconstructed=(100,100) #left = 0
# of res objects constructed/deconstructed=(100,100) #left = 0

USER   PID  %CPU %MEM      VSZ    RSS   TT  STAT STARTED      TIME COMMAND
dan  75775 102.8  0.6  2932208 101080 s002  S+   10:39AM  10:59.57 /Users/dan/src/M2/M2-Macaula
....................................................................................................size of each slabs = 2040
total engine space allocated = 0k
           stash    k total   k in use       size     nalloc      inuse  highwater      freed
Finalizations of new resolutions:
# of res objects constructed/deconstructed=(200,180) #left = 20
# of res objects constructed/deconstructed=(200,180) #left = 20

USER   PID  %CPU %MEM      VSZ    RSS   TT  STAT STARTED      TIME COMMAND
dan  75775 140.4  2.0  3177916 331080 s002  S+   10:39AM  12:14.92 /Users/dan/src/M2/M2-Macaula

Solving it would improve memory performance a lot.

[DanGrayson]

commit e7ee2bc4e5bd4978261c45987bb0e71da35cace9 will fix this

But we don't understand why.

[DanGrayson]

This file gives a simpler example of the memory leak:

https://github.com/Macaulay2/Workshop-2016-Warwick/blob/master/memoryLeaks/leak3.m2

Almost any change to this example diminishes the leak, but it's hard to believe that libgc is so easily thwarted by multiple cycles in the graph of pointers between memory blocks. It's a mystery.

We set up the computation for the nonminimal resolution, but don't start it. Observe that a preparatory gb is computed for it, nevertheless.