PPA for OpenSCAD 2021

Hi @tp3, we now know what happened in our unit tests so I thought I would report back.

We have our own OpenSCAD dictionary library inside the OpenFlexure Micrsocope project, and we test it really carefully because I am worried about edge cases causing things to fail. We also always have hardwarnings on, so it must run without warnings.

We use the C-style for loop list comprehensions quite a bit, iterating more than one value in the last section. An dummy example of some code that generates the warnings we were getting is:

 b= [1,2,3];
 a = [for (i=-1, j=0; i<len(b); i=i+1, j=j+b[i]) j];
 echo(a);

Here on the last iteration i becomes equal to b, so the loop will exit. But the next value of j is calculated before the loop exits. This means that OpenSCAD 2021 throws a the warning:

WARNING: undefined operation (number + undefined)

We have just adjusted our library so we no longer set off this warning. Our resulting code is actually nicer.

I am not sure if I would class this as unexpected behaviour from OpenSCAD. Warning about undef in arithmetic is so important, but then it does make it slightly harder to iterate over all values in an array. But perhaps there is a nicer way that I am missing.

---

Sent from the OpenSCAD mailing list archive at Nabble.com.
_______________________________________________
OpenSCAD mailing list
To unsubscribe send an email to [hidden email]

I did some timing tests with C-like for vs recursion and found that actually C-like for was slower.  I don't know how broadly applicable those test results were, but it seems like in many cases (most cases?) it's better to just use recursion.  

I do not understand how undef in arithmetic makes it "slightly harder to iterate over all values in an array".   I don't even see any connection between these things.  To iterate over all values in an array you do

for(entry=array) ....

Ronaldo wrote

That is a well-known design issue of the C-like for. The last iteration is
useless. See: https://github.com/openscad/openscad/issues/3227
The only way I know to handle the issue is to protect the expressions in
the loop avoiding illegal operations in the last iteration.

 b= [1,2,3];
 a = [for (i=-1, j=0; i<len(b); i=i+1, j= (i==len(b)) ? j : j+b[i]) j];
 echo(a);


Em dom., 25 de abr. de 2021 às 12:50, julianstirling <
[hidden email]> escreveu:

> Hi @tp3, we now know what happened in our unit tests so I thought I would
> report back.
>
> We have our own OpenSCAD dictionary library inside the OpenFlexure
> Micrsocope project, and we test it really carefully because I am worried
> about edge cases causing things to fail. We also always have hardwarnings
> on, so it must run without warnings.
>
> We use the C-style for loop list comprehensions quite a bit, iterating
> more than one value in the last section. An dummy example of some code that
> generates the warnings we were getting is:
>
>  b= [1,2,3];
>  a = [for (i=-1, j=0; i<len(b); i=i+1, j=j+b[i]) j];
>  echo(a);
>
>
> Here on the last iteration i becomes equal to b, so the loop will exit.
> But the next value of j is calculated before the loop exits. This means
> that OpenSCAD 2021 throws a the warning:
>
> WARNING: undefined operation (number + undefined)
>
>
> We have just adjusted our library so we no longer set off this warning.
> Our resulting code is actually nicer.
>
> I am not sure if I would class this as unexpected behaviour from OpenSCAD.
> Warning about undef in arithmetic is so important, but then it does make it
> slightly harder to iterate over all values in an array. But perhaps there
> is a nicer way that I am missing.
> ------------------------------
> Sent from the OpenSCAD mailing list archive <http://forum.openscad.org/>
> at Nabble.com.
> _______________________________________________
> OpenSCAD mailing list
> To unsubscribe send an email to [hidden email]
>

_______________________________________________
OpenSCAD mailing list
To unsubscribe send an email to [hidden email]

---

Sent from the OpenSCAD mailing list archive at Nabble.com.
_______________________________________________
OpenSCAD mailing list
To unsubscribe send an email to [hidden email]

Are you saying that has something to do with iterating over all the elements in an array?  

The problem with the orphan iteration in for loops is almost always a problem that requires some special handling, whether or not undef is silent in arithmetic.  The problem is the stupid behavior of c-style for, not undefs.   And in my experience the code is harder to read for c-style for than for recursion.  I think the quoted example, a vector sum code, is harder to read than a basic recursive version.  And also, vector sum can be done non-recursively by vector multiplication: b * [for(x=b) 1]

nophead wrote

You need to use the C style loop to be able to accumulate j in the loop. Or
use recursion.

On Sun, 25 Apr 2021 at 18:10, adrianv <[hidden email]> wrote:

> I did some timing tests with C-like for vs recursion and found that
> actually C-like for was slower.  I don't know how broadly applicable those
> test results were, but it seems like in many cases (most cases?) it's
> better to just use recursion.
>
> I do not understand how undef in arithmetic makes it "slightly harder to
> iterate over all values in an array".   I don't even see any connection
> between these things.  To iterate over all values in an array you do
>
> for(entry=array) ....
>
> Ronaldo wrote
> That is a well-known design issue of the C-like for. The last iteration is
> useless. See: https://github.com/openscad/openscad/issues/3227
> The only way I know to handle the issue is to protect the expressions in
> the loop avoiding illegal operations in the last iteration.
>
>  b= [1,2,3];
>  a = [for (i=-1, j=0; i<len(b); i=i+1, j= (i==len(b)) ? j : j+b[i]) j];
>  echo(a);
>
>
> Em dom., 25 de abr. de 2021 às 12:50, julianstirling <
> [hidden email] <http:///user/SendEmail.jtp?type=email&email=julian%40.co>>
> escreveu:
>
> > Hi @tp3, we now know what happened in our unit tests so I thought I
> would
> > report back.
> >
> > We have our own OpenSCAD dictionary library inside the OpenFlexure
> > Micrsocope project, and we test it really carefully because I am worried
> > about edge cases causing things to fail. We also always have
> hardwarnings
> > on, so it must run without warnings.
> >
> > We use the C-style for loop list comprehensions quite a bit, iterating
> > more than one value in the last section. An dummy example of some code
> that
> > generates the warnings we were getting is:
> >
> >  b= [1,2,3];
> >  a = [for (i=-1, j=0; i<len(b); i=i+1, j=j+b[i]) j];
> >  echo(a);
> >
> >
> > Here on the last iteration i becomes equal to b, so the loop will exit.
> > But the next value of j is calculated before the loop exits. This means
> > that OpenSCAD 2021 throws a the warning:
> >
> > WARNING: undefined operation (number + undefined)
> >
> >
> > We have just adjusted our library so we no longer set off this warning.
> > Our resulting code is actually nicer.
> >
> > I am not sure if I would class this as unexpected behaviour from
> OpenSCAD.
> > Warning about undef in arithmetic is so important, but then it does make
> it
> > slightly harder to iterate over all values in an array. But perhaps
> there
> > is a nicer way that I am missing.
> > ------------------------------
> > Sent from the OpenSCAD mailing list archive <http://forum.openscad.org/>
> > at Nabble.com.
> > _______________________________________________
> > OpenSCAD mailing list
> > To unsubscribe send an email to [hidden email]
> <http:///user/SendEmail.jtp?type=email&email=discuss-leave%40.openscad>
> >
>
> _______________________________________________
> OpenSCAD mailing list
> To unsubscribe send an email to [hidden email]
> <http:///user/SendEmail.jtp?type=email&email=discuss-leave%40.openscad>
>
>
> ------------------------------
> Sent from the OpenSCAD mailing list archive <http://forum.openscad.org/>
> at Nabble.com.
> _______________________________________________
> OpenSCAD mailing list
> To unsubscribe send an email to [hidden email]
>

_______________________________________________
OpenSCAD mailing list
To unsubscribe send an email to [hidden email]

---

Sent from the OpenSCAD mailing list archive at Nabble.com.
_______________________________________________
OpenSCAD mailing list
To unsubscribe send an email to [hidden email]