[ECOS] AW: probably simple problem: ASSERT in Cyg_StdIostream

[ECOS] AW: probably simple problem: ASSERT in Cyg_StdIostream

[Manfred Wölfel]

In the case of buffered IO,  the constructor of Cyg_StdioStream 
constructs a Cyg_StdioStreamBuffer object io_buf. The constructor of 
Cyg_StdioStreamBuffer calls Cyg_StdioStreamBuffer::set_buffer that may 
fail for lack of memory. The returned error is silently ignored in the 
constructor of Cyg_StdioStreamBuffer. In this case the constructor of 
Cyg_StdioStream gives up, before the magic value is set.


-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

Re: [ECOS] AW: probably simple problem: ASSERT in Cyg_StdIostream

[Andrew Lunn]

On Tue, Apr 25, 2006 at 07:43:16AM +0200, Manfred W?lfel wrote:
> In the case of buffered IO,  the constructor of Cyg_StdioStream 
> constructs a Cyg_StdioStreamBuffer object io_buf. The constructor of 
> Cyg_StdioStreamBuffer calls Cyg_StdioStreamBuffer::set_buffer that may 
> fail for lack of memory. The returned error is silently ignored in the 
> constructor of Cyg_StdioStreamBuffer. In this case the constructor of 
> Cyg_StdioStream gives up, before the magic value is set.

So what are you suggesting? What is the fix? Could you provide a
patch?

   Thanks
        Andrew

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

Re: [ECOS] AW: probably simple problem: ASSERT in Cyg_StdIostream

[Andrew Lunn]

On Tue, Apr 25, 2006 at 10:43:36AM +0200, Manfred W?lfel wrote:
> No, I'm referring to 
> http://ecos.sourceware.org/ml/ecos-discuss/2005-04/msg00103.html and the 
> like. This is not the only topic that has been raised as a result of 
> that problem. I just wanted to make people that face this problem aware 
> of the need to break in  Cyg_StdioStreamBuffer::set_buffer. But if you 
> ask me for it, we have to give up completely, if memory allocation fails:
> 
> Index: packages/language/c/libc/stdio/current/include/streambuf.inl
> ===================================================================
> RCS file: 
> /cvs/ecos/ecos/packages/language/c/libc/stdio/current/include/streambuf.inl,v
> retrieving revision 1.6
> diff -u -r1.6 streambuf.inl
> --- packages/language/c/libc/stdio/current/include/streambuf.inl        
> 23 May 2002 23:07:16 -0000       1.6
> +++ packages/language/c/libc/stdio/current/include/streambuf.inl        
> 25 Apr 2006 08:41:32 -0000
> @@ -92,7 +92,8 @@
>     // unnecessary, but it is to ensure a defined state if e.g. the malloc
>     // in set_buffer() should fail
> 
> -    (void)set_buffer(size, new_buffer);
> +       Cyg_ErrNo err = set_buffer(size, new_buffer);
> +       CYG_ASSERT(err == ENOERR, "insufficient memory");
> } // Cyg_StdioStreamBuffer constructor
> 
> 
> Andrew Lunn wrote:
> >On Tue, Apr 25, 2006 at 07:43:16AM +0200, Manfred W?lfel wrote:
> >  
> >>In the case of buffered IO,  the constructor of Cyg_StdioStream 
> >>constructs a Cyg_StdioStreamBuffer object io_buf. The constructor of 
> >>Cyg_StdioStreamBuffer calls Cyg_StdioStreamBuffer::set_buffer that may 
> >>fail for lack of memory. The returned error is silently ignored in the 
> >>constructor of Cyg_StdioStreamBuffer. In this case the constructor of 
> >>Cyg_StdioStream gives up, before the magic value is set.

I don't think throwing an assert is the correct fix. It looks like
Cyg_StdioStream::initialize returns ENOMEM when the buffer cannot be
created. The constructor needs to return this error code so what ever
needs the stream knows there was no memory and propergate on the
error.

        Andrew

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss