[M3devel] narrow still failing..

[Rodney M. Bates]

On 01/21/2014 11:11 AM, mika at async.caltech.edu wrote:
> Did you try using my bdd.tgz ?
>
> I am getting this error on:
>
> ancient PM3 on FreeBSD4
>
> head CM3 on AMD64_LINUX

Hmm.  I can't get any errors compiling bdd and bddtest on my AMD64_LINUX system with
a freshly updated head.  Hacking around, I tried to use the binaries in bdd.tgz,
but my .M3SHIP wants a libbdd.so.5, not there.  Some of the *.[im]o files are
slightly different in size when I recompiled, but I think there is an explanation
for that.

I can't build bddtest with the release compiler, because it needs caltech_parser,
and that won't build in the release.  This combination of [non]failures seems
pretty strange.  I am also wondering about a problem with type checks in the
runtime.


>
> few months old CM3 on ARM_LINUX (Raspberry Pi)
>
> I doubt you will get it on LINUXLIBC6 if you don't get it on the others.
>
> I already tried rearranging the files (the implementation and interfaces
> are split in a funny way across BDD.i3, BDDImpl.i3, BDD.m3, BDDImpl.m3
> right now).  With no effect.  I don't know what it is about my code that's
> tripping up the compiler.
>
> It's also a "decent" BDD package so someone (me I guess, but anyone
> else who's interested is welcome to) might want to add it to CM3... :-)
> Kind of a pain if you can't print out the debugging info though!
> (Actually the workarounds are OK.)
>
>      Mika
>
> "Rodney M. Bates" writes:
>>
>>
>> On 01/20/2014 10:18 PM, mika at async.caltech.edu wrote:
>>> Were you able to get it to fail on any of your computers?
>>
>> No, the only times I got narrow failures were due to errors
>> in my trial cases.
>>
>> I have only tried AMD64_LINUX, but should be able to get this onto a
>> LINUXLIBC6 system.
>>
>>>
>>> Your Z fails for me---CM3 and PM3 both.
>>>
>>> (579)truffles:~/bsd/refany/src>../AMD64_LINUX/refany
>>>
>>>
>>> ***
>>> *** runtime error:
>>> ***    An explicit or implicit NARROW() operation failed.
>>> ***    file "../src/Main.m3", line 30
>>> ***
>>
>> Well that torpedos both my theories.
>>
>> But it is hard to imagine that bad code generated for an implicit narrow
>> on an assignment could ever have gone unnoticed so long.  I thought it
>> might be more believable if it only happened when narrowing an actual
>> parameter before passing it.
>>
>>>
>>> Abort
>>> (580)truffles:~/bsd/refany/src>cat Mai
>>> Main.m3~  Main.m3
>>> (580)truffles:~/bsd/refany/src>cat -n Main.m3
>>>        1  MODULE Main;
>>>        2  IMPORT BDD, IO;
>>>        3
>>>        4  PROCEDURE P() : REFANY =
>>>        5    BEGIN
>>>        6      VAR x : REFANY;
>>>        7      BEGIN
>>>        8        x := BDD.New("a");
>>>        9        Z(x);
>>>       10        RETURN x
>>>       11      END
>>>       12    END P;
>>>       13
>>>       14  PROCEDURE Q() =
>>>       15    BEGIN
>>>       16      IO.Put(P() & "\n")
>>>       17    END Q;
>>>       18
>>>       19  (*
>>>       20  PROCEDURE Z(VAR x : REFANY) =
>>>       21    BEGIN
>>>       22      TYPECASE x OF
>>>       23        BDD.T(b) => x := BDD.Format(b)
>>>       24      END;
>>>       25    END Z;
>>>       26  *)
>>>       27  PROCEDURE Z(VAR x : REFANY) =
>>>       28     VAR b: BDD.T;
>>>       29     BEGIN
>>>       30       b := x;
>>>       31       x := BDD.Format(b)
>>>       32     END Z;
>>>       33
>>>       34
>>>       35
>>>       36  BEGIN
>>>       37    Q()
>>>       38  END Main.
>>> (581)truffles:~/bsd/refany/src>
>>> "Rodney M. Bates" writes:
>>>> What does this do?
>>>>
>>>> PROCEDURE Z(VAR x : REFANY) =
>>>>     VAR b: BDD.T;
>>>>     BEGIN
>>>>       b := x;
>>>>       x := BDD.Format(b)
>>>>     END Z;
>>>>
>>>> The common property of the failing cases seems to be the necessity
>>>> of a runtime narrow check when passing x to BDD.Format.
>>>>
>>>> So far, I haven't been able to reproduce the failure on AMD64_LINUX,
>>>> with even more irrelevant (one would think) stuff pared out.  I moved
>>>> T, Root, and Format into main, removed all but one fields of T and Root,
>>>> replaced BDD.New by NEW(Root) and removed Symtab and most of the body
>>> >from Format.
>>>>
>>>> Looking at your code, I agree it should work.
>>>>
>>>>
>>>> On 01/19/2014 04:01 PM, mika at async.caltech.edu wrote:
>>>>> As in the ISTYPE/TYPECASE example, the following works:
>>>>>
>>>>>
>>>>> PROCEDURE Z(VAR x : REFANY) =
>>>>>      BEGIN
>>>>>        TYPECASE x OF
>>>>>          BDD.T(b) => x := BDD.Format(b)
>>>>>        END;
>>>>>      END Z;
>>>>>
>>>>> (244)rover:~/refany/src>../FreeBSD4/refany
>>>>> a
>>>>>
>>>>>
>>>>> This does NOT work:
>>>>>
>>>>> PROCEDURE Z(VAR x : REFANY) =
>>>>>      BEGIN
>>>>>        TYPECASE x OF
>>>>>          BDD.T(b) => x := BDD.Format(x)
>>>>>        END;
>>>>>      END Z;
>>>>>
>>>>>
>>>>> (242)rover:~/refany/src>../FreeBSD4/refany
>>>>>
>>>>>
>>>>> ***
>>>>> *** runtime error:
>>>>> ***    NARROW failed
>>>>> ***    file "/big/home/mika/refany/src/Main.m3", line 22
>>>>> ***
>>>>>
>>>>>      use option @M3stackdump to get a stack trace
>>>>> Abort
>>>>>
>>>>>
>>>>> Same result PM3 / CM3.
>>>>>
>>>>>         Mika
>>>>>
>>>
>