[M3commit] CVS Update: cm3 (Warning, type is NULL)

[Rodney M. Bates]

Tony Hosking wrote:
> On 3 Feb 2010, at 06:43, Jay K wrote:
> 
>> That is legal? What does it mean? What can you do it with t?
> 
> It's a variable of type NULL.  It can only hold the NIL reference. 
>  Agreed, it is not particularly useful.
> 
> The warning is there to distinguish:
> 
> VAR t: NULL
> 
> which a programmer presumably intended to write, from
> 
> VAR t:= complicated expression of type NULL (perhaps hidden through 
> various type renaming)
> 
> which the programmer possibly did not intend to write.

IMO, the appropriate thing for the front end to do is not to set the
type of r to NULL, which is a perfectly valid Modula-3 type, but
rather to an internal value named TypeUnknownBecauseOfAPriorError.
Then make all warning/error checks silently succeed when applied to
such unknown types, or anything else unknown in his way.

> 
>> I compiled all of cm3 (at least what isn't filtered out on NT386) and 
>> it didn't break anything.
>>  
>>  - Jay
>>
>>  
>> ------------------------------------------------------------------------
>> From: hosking at cs.purdue.edu <mailto:hosking at cs.purdue.edu>
>> Date: Wed, 3 Feb 2010 00:52:34 -0500
>> To: jay.krell at cornell.edu <mailto:jay.krell at cornell.edu>
>> CC: m3commit at elegosoft.com <mailto:m3commit at elegosoft.com>
>> Subject: Re: [M3commit] CVS Update: cm3
>>
>> Clearly the source code is not legal.  It's just that your change 
>> breaks legal code:
>>
>> VAR t: NULL;
>>
>> which should emit a warning but nothing else.
>>
>> The problem lies elsewhere...
>>
>>
>> On 3 Feb 2010, at 00:22, Jay K wrote:
>>
>>     The backend should be able to cope with almost anything, I guess.
>>     But is the source code really legal?
>>     Shouldn't the front end error?
>>     The front end should error on some things, obviously.
>>      
>>     Notice that all of the "internal errors" in this scenario are from
>>     the front end. Just the assertion failure is not.
>>      
>>      
>>     The front end doesn't handle its own stack correctly
>>     here, the errors come from its CG.m3 module that wraps
>>     the backend.
>>      
>>     If this code is truly illegal and has no valid meaning, I think
>>     this is an ok change.
>>     The NT386 backend I guess should be hardened against front end
>>     bugs though.
>>      
>>      
>>      - Jay
>>
>>      
>>     ------------------------------------------------------------------------
>>     From: hosking at cs.purdue.edu <mailto:hosking at cs.purdue.edu>
>>     Date: Tue, 2 Feb 2010 12:19:16 -0500
>>     To: jkrell at elego.de <mailto:jkrell at elego.de>
>>     CC: m3commit at elegosoft.com <mailto:m3commit at elegosoft.com>
>>     Subject: Re: [M3commit] CVS Update: cm3
>>
>>     That's the wrong approach.  A warning is not an error.  The
>>     backend should be able to cope with it...
>>
>>     Antony Hosking | Associate Professor | Computer Science | Purdue
>>     University
>>     305 N. University Street | West Lafayette | IN 47907 | USA
>>     Office +1 765 494 6001 | Mobile +1 765 427 5484
>>
>>
>>
>>
>>     On 1 Feb 2010, at 20:17, Jay Krell wrote:
>>
>>         CVSROOT: /usr/cvs
>>         Changes by: jkrell at birch. 10/02/01 20:17:28
>>
>>         Modified files:
>>         cm3/m3-sys/m3front/src/values/: Variable.m3 
>>
>>         Log message:
>>         test case e0\e029:
>>
>>         MODULE Main;
>>         TYPE Rec = RECORD a: INTEGER END;
>>         VAR  r := Rec;    (* warning: variable has type NULL *)
>>         BEGIN
>>         INC(r.a)
>>         END Main.
>>
>>         followed by internal errors and an assertion failure in m3back
>>
>>         Change warning to error, and then it never gets to the
>>         internal errors.
>>
>>
>>
>>
>>
>