"my(...)" scoping question

[hermann@stamm-wilbrandt.de]

Running this GP script
https://github.com/Hermann-SW/RSA_numbers_factored/blob/main/pari/all_rsa.foursquare.gp

all is fine:

$ gp -q < all_rsa.foursquare.gp
...
2048 [28, 
158197984288196117699097355897947500264458455190848222964138664214873348733353898043756194523483689509093385743402168573089458743256253901017830264069606012704497883085623209455523704906878923700142603965513457096959088734082854043425944788477935555382589380809038501009740812793032665236884977009980514922487, 
13011773238477059231321994244187743729468594581843753487082884034603212622185946597112257777907861613001484624197063577339843512098442794834693496703933881475402708427339094714589624392731377779402228770755581888942318831433232790010161732825667949844267841944332119616646371761146340753071052451512068626398, 
0]~
  ***   last result computed in 3,373 ms.
all asserts OK
hermann@7600x:~/RSA_numbers_factored/pari$



I added "if(n%8!=7,assert(F[4]==0));" as last line into main "{ ... }"
block (to prove that really sum of 3 squares got computed):

{
  foreach(rsa,r,[l,n]=r;
    my(F=foursquarep(n));
    print(l," ",F);
    assert(vecsum([x^2|x<-F])==n));
    if(n%8!=7,assert(F[4]==0));
}

Although output for RSA-2048 as well as vecsum assert are good, the
following assert "thinks" F is a t_POL:

...
2048 [28, 
158197984288196117699097355897947500264458455190848222964138664214873348733353898043756194523483689509093385743402168573089458743256253901017830264069606012704497883085623209455523704906878923700142603965513457096959088734082854043425944788477935555382589380809038501009740812793032665236884977009980514922487, 
13011773238477059231321994244187743729468594581843753487082884034603212622185946597112257777907861613001484624197063577339843512098442794834693496703933881475402708427339094714589624392731377779402228770755581888942318831433232790010161732825667949844267841944332119616646371761146340753071052451512068626398, 
0]~
  ***   at top-level: ...)==n));if(n%8!=7,assert(F[4]==0))
  ***                                             ^--------
  ***   incorrect type in _[_] OCcompo1 [not a vector] (t_POL).
  ***   last result computed in 0 ms.
all asserts OK
hermann@7600x:~/RSA_numbers_factored/pari$


Moving "my(F)" before foreach loop makes it work again like initially 
shown:

{
  my(F);
  foreach(rsa,r,[l,n]=r;
    F=foursquarep(n);
    print(l," ",F);
    assert(vecsum([x^2|x<-F])==n));
    if(n%8!=7,assert(F[4]==0));
}


Is the t_POL assertion a bug, or do I understand "my(...)" scoping 
wrong?


Regards,

Hermann.