Experimental Physics and
Industrial Control System

Benjamin Franksen <[email protected]> · Thu, 20 Sep 2007 14:50:40 +0200

On Wednesday 19 September 2007 15:36, Benjamin Franksen wrote:
> On Tuesday 18 September 2007 17:03, Eric Norum wrote:
> > On Sep 18, 2007, at 9:59 AM, Jeff Hill wrote:
> > > Hi Ben,
> > >
> > > You might try commenting out iocInit in the startup script, and
> > > then after
> > > the IOC boots up type "sp iocInit". When it fails type "tt <task
> > > id>" to get
> > > the stack trace. If that doesn't help you might also build for
> > > debugging and
> > > attach the Tornado debugger to the IOC.
> >
> > This is running on linux so Benjamin needs only to run the IOC under
> > gdb.
>
> I have run it under ddd and valgrind and attached the outputs. I used
> asyn4-8 and stream2-2. The problem /appears/ to be in stream support
> (contrary to what I thought at first). OTOH, the many many timeouts are
> completely inexplicable to me and suggest that there is something wrong
> with asyn/vxi-11, too. Anyway, I am completely lost; I have great
> difficulty understabnding the relevant stream (C++) code.

Ok, I've found one bug in stream version 2-2 that leads to a crash. A format 
of "%*[whatever>]"
gets converted internally to "%0[whatever>]%n" and then only /one/ pointer 
arg is provided to sscanf, instead of two. The method where sscanf crashes 
is in StreamFormatConverter.cc:

int StdCharsetConverter::
scanString(const StreamFormat& fmt, const char* input,
    char* value, size_t maxlen)
{
    int length = -1;
    if (fmt.flags & skip_flag)
    {
        if (sscanf (input, fmt.info, &length) < 0) return -1;
    }
    ....

Since skip_flag is set only for %* formats, it is wrong to replace it with 
%0. Thus the parse method must be changed, like this (see comments):

int StdCharsetConverter::
parse(const StreamFormat& fmt, StreamBuffer& info,
    const char*& source, bool scanFormat)
{
    if (!scanFormat)
    {
        error("Format conversion %%[ is only allowed in input formats\n");
        return false;
    }
    if (fmt.flags & (left_flag|sign_flag|space_flag|zero_flag|alt_flag))
    {
        error("Use of modifiers '-', '+', ' ', '0', '#'"
            "not allowed with %%%c conversion\n",
            fmt.conv);
        return false;
    }
    if (scanFormat && fmt.prec >= 0)
    {
        error("Use of precision field '.%d' not allowed with %%%c input 
conversion\n",
            fmt.prec, fmt.conv);
        return false;
    }
    // fix: bfr 20/09/07
    if (fmt.flags & skip_flag ) {
        info.printf("%%*[");
    }
    else {
        info.printf("%%%d[", fmt.width);
    }
    // end fix
    while (*source && *source != ']')
    {
        if (*source == esc) source++;
        info.append(*source++);
    }
    if (!*source) {
        error("Missing ']' after %%[ format conversion\n");
        return false;
    }
    source++; // consume ']'
    info.append("]%n");
    return string_format;
}

However, the IOC crashes anyway, because this is just one of many bugs. I'll 
send more debugger output soon.

Cheers
Ben
-- 
"Programming = Mathematics + Murphy's Law" (E.W.Dijkstra)

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System