orbit-python-list

[Orbit-python-list] Problem with parsing of recursive structs in IDL

[Brad C. <cha...@ar...>]

Hello all;
I ran into a problem with CORBA._load_idl() getting a Bus error and
core dump while parsing an IDL that I work on. After a bit of
narrowing down, it appears as if this is a minimal IDL that will
cause the problem:

struct RecursiveStruct {
  sequence <RecursiveStruct> inner_struct;
};

interface ProblemInterface {
  attribute RecursiveStruct problem_item;
};

The idea is that you have a struct that has, as one of its member, a
list of itself (in the IDL that I'm working on this is used to descend
a sort of "tree" of items). When a struct like this is present inside
an interface, it will cause the problem. gdb gives a trace that looks
like:

#0  0x283c6dba in _thread_leave_cancellation_point () from /usr/lib/libc_r.so.4
#1  0x283c705d in _thread_leave_cancellation_point () from /usr/lib/libc_r.so.4
#2  0x283c7781 in malloc () from /usr/lib/libc_r.so.4
#3  0x283bf179 in calloc () from /usr/lib/libc_r.so.4
#4  0x28461dc7 in g_malloc0 () from /usr/local/lib/libglib12.so.3
#5  0x283ff8be in alloc_typecode () at idl.c:917
#6  0x28400538 in get_struct_typecode (tree=0x81af5c0) at idl.c:1172
#7  0x28401525 in get_typecode (tree=0x81af5c0) at idl.c:1431
#8  0x283ffe59 in get_ident_typecode (tree=0x81af240) at idl.c:1043
#9  0x28401565 in get_typecode (tree=0x81af240) at idl.c:1439
#10 0x2840082a in get_sequence_typecode (tree=0x81af600) at idl.c:1219
#11 0x2840158d in get_typecode (tree=0x81af600) at idl.c:1445
[...]

#s 5-11 will than repeat over and over again in the backtrace,
seeming to indicate that orbit-python gets itself into a loop while
parsing this.

This error happens for me with orbit-python 0.3.0, ORBit 0.5.8, on both
FreeBSD and NetBSD machines.

I hope this report gives enough info for someone with a knowledge of
idl.c; please let me know if you need more information (or would
rather I submit this to the bug-tracker instead of sending it here).

Thanks to all for orbit-python -- it has been really great to see all
the new features/improvements happening on it!

Brad

[Orbit-python-list] Re: Problem with parsing of recursive structs in IDL

[Jason T. <ta...@li...>]

* Brad Chapman (cha...@ar...) [010828 21:57]:
> I ran into a problem with CORBA._load_idl() getting a Bus error and
> core dump while parsing an IDL that I work on. After a bit of
> narrowing down, it appears as if this is a minimal IDL that will
> cause the problem:
> 
> struct RecursiveStruct {
>   sequence <RecursiveStruct> inner_struct;
> };
> 
> interface ProblemInterface {
>   attribute RecursiveStruct problem_item;
> };

Yep, this certainly causes nasty things.  In fact, orbit-idl also
segfaults on this IDL.  Maybe ORBit2's IDL compiler also has this
problem?

I'll have a look at the nature of this problem over the weekend.  If
it's nothing significant then it will be fixed for 0.3.1, which should
be released hopefully early next week.

BTW, you shouldn't really need to use _load_idl().  The only time you
need to use _load_idl() is when the preprocessor is broken, in which
case that should get fixed too.  (Well, another valid case to use
_load_idl() is when you want to load an IDL that is new since the CORBA
module was imported.  Mind you, I could probably work up a hack to get
around this problem too.)

Cheers,
Jason.

Re: [Orbit-python-list] Re: Problem with parsing of recursive structs in IDL

[Christian R. R. <ki...@as...>]

On Tue, 28 Aug 2001, Jason Tackaberry wrote:

> BTW, you shouldn't really need to use _load_idl().  The only time you
> need to use _load_idl() is when the preprocessor is broken, in which
> case that should get fixed too.  (Well, another valid case to use

Or when the IDL isn't in the IDLPATH, and you specify the full path.

Take care,
--
Christian Reis, Senior Engineer, Async Open Source, Brazil.
http://async.com.br/~kiko/ | [+55 16] 272 3330 | NMFL

[Orbit-python-list] Re: Problem with parsing of recursive structs in IDL

[Mark M. <ma...@sk...>]

Hi Jason,

On Tue, 28 Aug 2001, Jason Tackaberry wrote:

> > struct RecursiveStruct {
> >   sequence <RecursiveStruct> inner_struct;
> > };
> >
> > interface ProblemInterface {
> >   attribute RecursiveStruct problem_item;
> > };
>
> Yep, this certainly causes nasty things.  In fact, orbit-idl also
> segfaults on this IDL.  Maybe ORBit2's IDL compiler also has this
> problem?

	No, ORBit2's IDL compiler has fixed this problem. If my memory
serves me correctly, there's a fix for this in ORBit HEAD as well.

	Oh, and the Perl bindings have this problem too ;)

Good Luck,
Mark.

Re: [Orbit-python-list] Re: Problem with parsing of recursive structs in IDL

[Brad C. <cha...@ar...>]

Hey all;
Thanks for the responses on this.

[I talk about a seg-fault problem caused by self-referencing
structs]

Jason:
> Yep, this certainly causes nasty things.  In fact, orbit-idl also
> segfaults on this IDL.  Maybe ORBit2's IDL compiler also has this
> problem?

Doh, thanks -- I didn't even think of looking at orbit-idl. Well now
it makes sense why I'm seeing the problem :-).

Mark McLoughlin:
>  No, ORBit2's IDL compiler has fixed this problem. If my memory
> serves me correctly, there's a fix for this in ORBit HEAD as well.

Thanks for the info. I grabbed ORBit out of CVS (the stable
revision), and it still seems to have the problem for me. Don't know
if this'll help save any time looking at it with ORBit-python.

Jason:
> BTW, you shouldn't really need to use _load_idl(). 

Well, my use case is kind of what Christian was suggesting. I'm
actually emulating stubs and skeletons (by hacking up __init__.py in
one of my modules), so I just use the _load_idl() to grab the idl
from a predefined location. It doesn't really affect my portability
in this case, because I'm already going for portability between
multiple ORBs. Thanks for the advice, though!

[My minimal test case]
Christian:
> Minimal, but 3vil. Do you have a simple test script I can integrate
> to the test-suite so we guarantee it's fixed and stays fixed 
> over versions?

Hmmm, I'm not exactly sure if this is what you want, but the
attached script will cause the seg-fault with the IDL I posted. I'm
not sure if you want seg-faulting scripts in the Test Suite though
:-)

Brad

Re: [Orbit-python-list] Re: Problem with parsing of recursive structs in IDL

[Christian R. R. <ki...@as...>]

On Tue, 4 Sep 2001, Brad Chapman wrote:

> Hmmm, I'm not exactly sure if this is what you want, but the
> attached script will cause the seg-fault with the IDL I posted. I'm
> not sure if you want seg-faulting scripts in the Test Suite though

It's exactly what I want. Test scripts that SEGV are the latest fashion in
ORBit-Python; I've just discovered another three last week with using
__import__() after an import CORBA. That's fun.

Take care,
--
Christian Reis, Senior Engineer, Async Open Source, Brazil.
http://async.com.br/~kiko/ | [+55 16] 272 3330 | NMFL

Re: [Orbit-python-list] Re: Problem with parsing of recursive structs in IDL

[Christian R. R. <ki...@as...>]

On Tue, 4 Sep 2001, Christian Robottom Reis wrote:

> It's exactly what I want. Test scripts that SEGV are the latest fashion in
> ORBit-Python; I've just discovered another three last week with using
> __import__() after an import CORBA. That's fun.

Summary: I want more SEGVs!

I received your Any() patch for ORBit-Python, but since my refcounting
skills are a drag, I would like to ask you for a favor. If you could send
me a minimal testcase or a suggestion of one that triggers the problem, I
can gdb through and understand what is happening? Deal?

Take care,
--
Christian Reis, Senior Engineer, Async Open Source, Brazil.
http://async.com.br/~kiko/ | [+55 16] 272 3330 | NMFL

Re: [Orbit-python-list] Re: Problem with parsing of recursive structs in IDL

[Brad C. <cha...@ar...>]

Christian:
> Summary: I want more SEGVs!

All right, if you insist. :-)

> I received your Any() patch for ORBit-Python, but since my refcounting
> skills are a drag, I would like to ask you for a favor. If you could send
> me a minimal testcase or a suggestion of one that triggers the problem, I
> can gdb through and understand what is happening? Deal?

Sure, attached is a test that should cause either segfaults (does
this on linux mandrake) or lots of memory errors about free() 
(on NetBSD). The patch I sent makes this simple test run smoothly.
Let me know if you need more.

Brad
-- 
PGP public key available from http://pgp.mit.edu/

Re: [Orbit-python-list] Problem with parsing of recursive structs in IDL

[Christian R. R. <ki...@as...>]

On Tue, 28 Aug 2001, Brad Chapman wrote:

> narrowing down, it appears as if this is a minimal IDL that will
> cause the problem:
>
> struct RecursiveStruct {
>   sequence <RecursiveStruct> inner_struct;
> };
>
> interface ProblemInterface {
>   attribute RecursiveStruct problem_item;
> };

Minimal, but 3vil. Do you have a simple test script I can integrate to the
test-suite so we guarantee it's fixed and stays fixed over versions?

> The idea is that you have a struct that has, as one of its member, a
> list of itself (in the IDL that I'm working on this is used to descend
> a sort of "tree" of items). When a struct like this is present inside
> an interface, it will cause the problem. gdb gives a trace that looks
> like:

By list of itself you mean "list of things of the same type as it", right?

Take care,
--
Christian Reis, Senior Engineer, Async Open Source, Brazil.
http://async.com.br/~kiko/ | [+55 16] 272 3330 | NMFL