[Caml-list] Compiler killer code?

[Caml-list] Compiler killer code?

[Alessandro Baretta]

Is the type inference algorithm guaranteed to termintate? I 
ask because I have accidentally attempted to evaluate a 
(meaningless) function which almost hanged my Athlon.

The killer code is the following:

type t = t -> t
let f (x:t) :t = f f


If I type this into the toplevel, it starts to allocate 
memory by the tens of megabytes, until I have to kill it to 
prevent a system crash. I let it reach approximately 128MB 
before killing it. (BTW, that was with ocaml 3.04+13 
2002-06-05.)

Is the language really supposed to accept such garbage as 
that which I wrote?

Alex

-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners

Re: [Caml-list] Compiler killer code?

[William Lee Irwin III]

On Sat, Jun 08, 2002 at 04:32:13PM +0200, Alessandro Baretta wrote:
> Is the type inference algorithm guaranteed to termintate? I 
> ask because I have accidentally attempted to evaluate a 
> (meaningless) function which almost hanged my Athlon.
> The killer code is the following:
> type t = t -> t
> let f (x:t) :t = f f
> If I type this into the toplevel, it starts to allocate 
> memory by the tens of megabytes, until I have to kill it to 
> prevent a system crash. I let it reach approximately 128MB 
> before killing it. (BTW, that was with ocaml 3.04+13 
> 2002-06-05.)
> Is the language really supposed to accept such garbage as 
> that which I wrote?

I get this instead:

# type t = t -> t;;
The type abbreviation t is cyclic

Cheers,
Bill
-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners

Re: [Caml-list] Compiler killer code?

[William Lovas]

On Sat, Jun 08, 2002 at 07:34:02AM -0700, William Lee Irwin III wrote:
> > Is the type inference algorithm guaranteed to termintate? I 
> > ask because I have accidentally attempted to evaluate a 
> > (meaningless) function which almost hanged my Athlon.
> > The killer code is the following:
> > type t = t -> t
> > let f (x:t) :t = f f
> > If I type this into the toplevel, it starts to allocate 
> > memory by the tens of megabytes, until I have to kill it to 
> > prevent a system crash. I let it reach approximately 128MB 
> > before killing it. (BTW, that was with ocaml 3.04+13 
> > 2002-06-05.)
> > Is the language really supposed to accept such garbage as 
> > that which I wrote?

I wonder if it really would have crashed your system, or if O'Caml would
have simply died with an `out of memory' error.

> On Sat, Jun 08, 2002 at 04:32:13PM +0200, Alessandro Baretta wrote:
> 
> I get this instead:
> 
> # type t = t -> t;;
> The type abbreviation t is cyclic

If you use recursive types (`ocaml -rectypes'), it typechecks fine:

    # type t = t -> t;;
    type t = t -> t
    # let rec f (x : t) : t = f f;;
    val f : t = <fun>

In fact, you don't even need to define the type abbreviation:

    # let rec f x = f f;;
    val f : 'a -> 'b as 'a = <fun>

This is O'Caml 3.04 -- if it fails to typecheck in the CVS version, then
i'd say either the semantics of recursive types have changed or a bug has 
been introduced.

cheers,
William
-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners

Re: [Caml-list] Compiler killer code?

[Alessandro Baretta]

William Lee Irwin III wrote:
> I get this instead:
> 
> # type t = t -> t;;
> The type abbreviation t is cyclic
> 
> Cheers,
> Bill

Here's what I do.

         Objective Caml version 3.04+13 (2002-06-05)

# type t = t -> t ;;
type t = t -> t
# let rec f (x:t) = f f ;;
Interrupted.

Question for the developers: are "-rectypes" automatically 
enabled in the CVS version I am using?

Alex

-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners

Re: [Caml-list] Compiler killer code?

[Xavier Leroy]

>          Objective Caml version 3.04+13 (2002-06-05)
> 
> # type t = t -> t ;;
> type t = t -> t
> # let rec f (x:t) = f f ;;
> Interrupted.

This looks like a bug in the development version.  Without the
-rectypes flag, the declaration of t should be rejected, like it is in
OCaml 3.04.  Interestingly, the development version handles your
example just fine when -rectypes is set.

A note to users of the development version: we appreciate your
testing, but please keep the latest official release around to check
weird behaviors, and when you observe a discrepancy like the above,
report it to caml-bugs@inria.fr.

- Xavier Leroy
-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners

Re: [Caml-list] Compiler killer code?

[Blair Zajac]

Xavier Leroy wrote:
> 
> >          Objective Caml version 3.04+13 (2002-06-05)
> >
> > # type t = t -> t ;;
> > type t = t -> t
> > # let rec f (x:t) = f f ;;
> > Interrupted.
> 
> This looks like a bug in the development version.  Without the
> -rectypes flag, the declaration of t should be rejected, like it is in
> OCaml 3.04.  Interestingly, the development version handles your
> example just fine when -rectypes is set.
> 
> A note to users of the development version: we appreciate your
> testing, but please keep the latest official release around to check
> weird behaviors, and when you observe a discrepancy like the above,
> report it to caml-bugs@inria.fr.

I think this points to something that should be considered for
future Ocaml development, that being a comprehensive test suite
to check the language.  This would also help test porting of the
language to new OSes.

All of the major compilers I can think of, Perl, Python, GCC, etc,
have a test suite and I think from a company's point of view looking
to use the language, a test suite written in Caml would give additional
credibility and a sense of professionalism that would only help
adoption rates.

A good model for the test suite would be Perl's test suite which checks
every aspect of the language and its modules and is written in Perl
itself.

Is there any interest in this?

Best,
Blair

-- 
Blair Zajac <blair@orcaware.com>
Web and OS performance plots - http://www.orcaware.com/orca/
-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners

Re: [Caml-list] Compiler killer code?

[Jacques Garrigue]

From: Alessandro Baretta <alex@baretta.com>
>          Objective Caml version 3.04+13 (2002-06-05)
> 
> # type t = t -> t ;;
> type t = t -> t
> # let rec f (x:t) = f f ;;
> Interrupted.

That's a bug: the compiler is never supposed to loop. At least in the
core language.
As already pointed out, the bug is on the first line: you shouldn't be
able to define such a type without -rectypes.

> Question for the developers: are "-rectypes" automatically 
> enabled in the CVS version I am using?

No. This is just a bug. Now fixed.
As Xavier remarked, questions regarding the CVS version should go to
caml-bugs@pauillac.inria.fr as most people on this list do not use it.
An extra advantage is that then we can do bug tracking, and your bug
id would appear in the CVS log.

Thank you for this bug report. This helped correcting on overlooked
bug.

Jacques Garrigue
-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners

Re: [Caml-list] Compiler killer code?

[Alessandro Baretta]

Jacques Garrigue wrote:
>>Question for the developers: are "-rectypes" automatically 
>>enabled in the CVS version I am using?
> 
> 
> No. This is just a bug. Now fixed.
> As Xavier remarked, questions regarding the CVS version should go to
> caml-bugs@pauillac.inria.fr as most people on this list do not use it.
> An extra advantage is that then we can do bug tracking, and your bug
> id would appear in the CVS log.
> 
> Thank you for this bug report. This helped correcting on overlooked
> bug.
> 
> Jacques Garrigue
> 

You are more than welcome. Thank you both, instead, for the 
continual improvements to O'Caml. I enjoy programming in 
O'Caml more than in any other language.

I'll keep in mind your comment about the caml-bugs reporting 
system. If I notice any strange behavior in the CVS version 
I'll ask you guys through that address.

Alex

-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners

[Caml-list] Re: Compiler killer code?

[Michal Moskal]

On Sun, Jun 09, 2002 at 11:08:19AM -0700, Blair Zajac wrote:
> A good model for the test suite would be Perl's test suite which checks
> every aspect of the language and its modules and is written in Perl
> itself.
> 
> Is there any interest in this?

And what about 2 stage bootstrap, with comprasion of generated object
files? Or maybe it's not so needed as it is in gcc? Anyway it would be
easier to make Makefiles comparse object files then write the testsuite,
but I don't know how usefull might it be.

-- 
: Michal Moskal ::::: malekith/at/pld-linux.org :  GCS {C,UL}++++$ a? !tv
: PLD Linux ::::::: Wroclaw University, CS Dept :  {E-,w}-- {b++,e}>+++ h
-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners

Re: [Caml-list] Compiler killer code?

[Xavier Leroy]

> I think this points to something that should be considered for
> future Ocaml development, that being a comprehensive test suite
> to check the language.

No one can object to this -- just like no one can object to "regular
exercice at the gym is good for you"; it's just that few have the
fortitute to actually do it.  Developing a good test suite takes about
as much effort as writing the software being tested.

For OCaml, we have some test suites for a few specific libraries and
run-time facilities, but that's about it.  Compilers and type-checkers
are even harder to test than run-time libraries.

So, if anyone wants to tackle this challenge, that would be very
useful indeed, but be aware that it's a huge task.

- Xavier Leroy
-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners

Re: [Caml-list] Re: Compiler killer code?

[Xavier Leroy]

> [Discussion of test suite]
> And what about 2 stage bootstrap

It's the "bootstrap" entry of the OCaml Makefile, and we've been doing
this for years.  (We compare executables and gcc compares object
files, but that doesn't make any significant differences.)

It is a good test, but by no means a complete test
suite.  In particular, it tests correct source code, while the hardest
thing to test in a type-checker and compiler are the error cases
(like the initial bug report that triggered this discussion, where
source code that should have been rejected was not).

- Xavier Leroy
-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners