Replace C NULL for iter stop with StopIteration

[plajjan]

No description provided.

[plajjan]

@nordlander I added a new test case which demonstrates some failure in the compiler to mark variables volatile. The new test case is in test/core_lang_auto/exceptions_volatile.act

First we can look at the working case, although note the two lines I commented out:

def fun2():
    a = "FOOBAR2" # easy to find in C source
    my_list = [1]
    complete = True
    #if 1 == 1:
    #    complete = False
    for r in my_list:
        complete = False
    return complete

kll@Kristians-MacBook-Air core_lang_auto % actonc --cgen exceptions_volatile.act | grep -A20 FOOBA
    B_str a = to$str("FOOBAR2");
    B_list my_list = B_mk_list(1, to$int(1));
    volatile B_bool complete = B_True;
    B_Iterator N_1iter = ((B_Iterator (*) (B_Iterable, B_list))W_27->$class->__iter__)(W_27, my_list);
    if ($PUSH()) {
        while ((B_True)->val) {
            B_int N_2val = ((B_int (*) (B_Iterator))N_1iter->$class->__next__)(N_1iter);
            B_int r = N_2val;
            complete = B_False;
        }
        $DROP();
    }
    else {
        B_BaseException N_3x = $POP();
        if ($ISINSTANCE0(N_3x, B_StopIteration)) {
        }
        else {
            $RAISE(N_3x);
        }
    }
    return complete;

most importantly we see volatile B_bool complete = B_True;. Now uncomment the lines, so we have:

def fun2():
    a = "FOOBAR2" # easy to find in C source
    my_list = [1]
    complete = True
    if 1 == 1:
        complete = False
    for r in my_list:
        complete = False
    return complete

and notice how volatile disappears. That doesn't look right...

kll@Kristians-MacBook-Air core_lang_auto % actonc --cgen exceptions_volatile.act | grep -A20 FOOBA
    B_str a = to$str("FOOBAR2");
    B_list my_list = B_mk_list(1, to$int(1));
    B_bool complete = B_True;
    if (ORD_B_int__eq__(to$int(1), to$int(1))) {
        complete = B_False;
    }
    B_Iterator N_1iter = ((B_Iterator (*) (B_Iterable, B_list))W_37->$class->__iter__)(W_37, my_list);
    if ($PUSH()) {
        while ((B_True)->val) {
            B_int N_2val = ((B_int (*) (B_Iterator))N_1iter->$class->__next__)(N_1iter);
            B_int r = N_2val;

[nordlander]

I've spotted the error. It's in Björn's recent additions to CodeGen, duplicating an error I also made in my recent mod to CPS (but was lucky enough to find and iron out!). Essentially, its root is the dubious nature of our utility function bound, which cannot tell a binding occurrence from a reassignment. There are alternatives in the compiler, but the honest truth is that the overloading of bound on Statements is a mess. :-(

I'll commit a fix to CodeGen tomorrow. But ideally we should also straighten up all code that deals with scopes in the compiler, for the benefit of our future selves! :-/

[plajjan]

@nordlander the telemetrify test is failing. Compilation and the (very small) test suite passes but then the final somewhat larger test of loading in a schema from a replay file fails:

/__w/acton/acton/telemetrify/telemetrify-core/out/bin/telemetrify.test maapi-schema
Current time: 2025-01-03T16:06:37.627548212Z
Starting up...
�S���MAAPI connected!!!! 
*d�l�m�adminh�a�aaa�d�systemajMAAPI user session started!!!! 
�������MAAPI loaded schema!!!! 
<$Actor -160 telemetrifyQ_nsoapiQ_schemaQ_SharedSchema at 0x7fef659a0f50>
MAAPI end user session failed: ValueError: Invalid `EObject` tag: 46
make: *** [Makefile:15: test] Error 1

I can only presume that control flow is messed up somehow. This is normally a very stable & deterministic test - I . Nonetheless, I have just cleared all CI caches and started a rerun just on the off chance there was an intermittent failure here.

[plajjan]

Same failure after clean rebuild... @nordlander

[plajjan]

🎆