internal breaking change 2021.01 vs 2019.05

[mjk-at-sag]

module print (emboss=$preview)
    if (emboss) { children(); }
    else { difference() { children(0); children([1:$children-1]); } }

module background ()
    if (false) cube([75, 20, 10], true);

print($preview) {
    background();
    color("black")
        translate([0, 0, 4.5]) linear_extrude(1)
        text(str(version()), valign="center", halign="center");
  }

1. open with 2021.05

2. F6 - render

3. read "WARNING: No top level geometry to render"

4. F7 - export

5. read "ERROR: Nothing to export! ..."

6. open with 2019.05

7. F6

8. F7

9. success

--- Want to back this issue? **[Post a bounty on it!](https://app.bountysource.com/issues/97022585-internal-breaking-change-2021-01-vs-2019-05?utm_campaign=plugin&utm_content=tracker%2F52063&utm_medium=issues&utm_source=github)** We accept bounties via [Bountysource](https://app.bountysource.com/?utm_campaign=plugin&utm_content=tracker%2F52063&utm_medium=issues&utm_source=github).

[thehans]

@mjk-at-sag This is a consequence of changes from #3342 and how empty geometries are now handled. The bug was in 2019 version, and is now fixed. You are subtracting your text from an empty child background(); so the result is empty geometry.

[mjk-at-sag]

@thehans Very well, I accept that. Now I don't understand why the following two designs produce the same results when previewed (F5) and different results when rendered (F6).

This produces the now correct result:

module print (emboss=$preview)
    if (emboss) { children(); }
    else { difference() { children(0); children([1:$children-1]); } }

module background ()
    if (false) cube([75, 20, 10], true);

print($preview) {
    //////// background();       //////////////// <========
    color("black")
        translate([0, 0, 4.5]) linear_extrude(1)
        text(str(version()), valign="center", halign="center");
    color("red")
        translate([0, 0, -0.5]) linear_extrude(1)
        text(str(version()), valign="center", halign="center");
}

This produces the old incorrect result:

module print (emboss=$preview)
    if (emboss) { children(); }
    else { difference() { children(0); children([1:$children-1]); } }

module background ()
    if (false) cube([75, 20, 10], true);

print($preview) {
    if (false) background();     ////////////////// <=======
    color("black")
        translate([0, 0, 4.5]) linear_extrude(1)
        text(str(version()), valign="center", halign="center");
    color("red")
        translate([0, 0, -0.5]) linear_extrude(1)
        text(str(version()), valign="center", halign="center");
}

[thehans]

OK, I see. The second example does appear to be a bug when rendered with F6.

[kintel]

Looking at this, it looks correct: The designs happen to produce the same results when previewed (F5) and different results when rendered (F6) because the design is written to handle previews differently from render.

[jordanbrown0]

The preview/render difference is presumably related to the $preview.

But the 2019-vs-later difference is, as @thehans described above, based on a change in the behavior of empty geometries.

Here's a test case:

module dummy() {
}

difference() {
    dummy();
    cube(10);
    cube(10, center=true);
}

2019.05 ignores the dummy child, for at least some purposes, and yields a cube with the corner cut out. 2021.01 and later subtract two cubes from nothing, and yield nothing.

Note: in all three, an "if" is counted as a child, but if it fails it yields no geometry. Or something like that. If you change the dummy(); to if (false) dummy();, all three yield the cube with corner cut off; the failing "if" does not count as the first child of the difference.

module dummy() {
}

difference() {
    if (false) dummy();
    cube(10);
    cube(10, center=true);
}

Similarly, if the first child of the difference() is an echo(), it is ignored for geometry purposes and yields the cube-with-cut-corner. However, if the first child of the difference is a module that calls echo() and does nothing else, in 2019.05 it's ignored and in 2021.01 and later that does count for geometry purposes, and you end up with nothing.

---

My humble opinion is that they are all wrong, though 2021.01 and later are better. Every module invocation, including echo, failing if, and for-that-executes-zero-times, should yield geometry, perhaps empty. Nothing should peek at the resulting geometry, in any way, to handle it specially if it is empty.

[jordanbrown0]

Further clarification: they should all yield exactly one child, which might be empty geometry.