Bug: setting elements of nested arrays sometimes goes wrong

[dc42]

When the 'set' command is used to change the value of an element of an array within an array, the assignment can go wrong. See https://forum.duet3d.com/topic/35725/question-array-assignment.

[dc42]

I identified two issues. The minor one is incorrect calls to lock the heap. This is fixed in commit 702c47b. The major one is that at ArrayHandle.cpp(91) the recursive call to InternalAssignIndexed may trigger compacting garbage collection, which may cause the elements of the array (one of which is being assigned) to be moved.

[dc42]

Code from user Nine Mile which is claimed to reproduce this:

Setup:

global mosET = { 0.0, null }
global mosTT = { vector(limits.tools, global.mosET) }

Some time later:

if { exists(param.X) }
    if { global.mosTT[param.P][1] == null }
        set global.mosTT[param.P][1] = { param.X, 0.0 }
    else
        set global.mosTT[param.P][1][0] = { param.X }

if { exists(param.Y) }
    if { global.mosTT[param.P][1] == null }
        set global.mosTT[param.P][1] = { 0.0, param.Y }
    else
        set global.mosTT[param.P][1][1] = { param.Y }

[dc42]

Here's a macro based on the code from Nine Mile that reproduces it. When invoked with a small Q parameter, it appears to work without error. When invoked with Q values greater than about 20 it either crashes or reports array index out of bounds.

var dummy = vector(50,0.0)
var mosET = { 0.0, null }
var mosTT = { vector(40, var.mosET) }
set var.dummy = 0.0

var X = 42
var Y = 24

while iterations < param.Q
    if { var.mosTT[iterations][1] == null }
        set var.mosTT[iterations][1] = { var.X, 0.0 }
    else
        set var.mosTT[iterations][1][0] = { var.X }

    if { var.mosTT[iterations][1] == null }
        set var.mosTT[iterations][1] = { 0.0, var.Y }
    else
        set var.mosTT[iterations][1][1] = { var.Y }

[dc42]

Believed fixed now in 3.5-dev. Andy S has confirmed that the fix works in his test setup. Now waiting for feedback from the user who posted the original thread.

[dc42]

User reported that it looks good, see https://forum.duet3d.com/post/339680.