Re: [Valgrind-developers] Bug 73655 fix

[Robert W. <rj...@du...>]

> Sounds plausible.  Is it documented?

Phngh.

> I'm not sure if the comments in
> vg_intercept.vgi now match with the reality.  A precis of your description
> could be what's needed...

Okey dokey.  I'll put a better description extracted from my message
into vg_intercept.c.base.

> Also, I see in the generated vg_intercept.c lines like this:
> 
>   int VG_INTERCEPT(soname$3Alibc$2Eso$2E6$3A__raise)(int)
> 
> How does that work?  What's with the dollar signs?

VG_INTERCEPT is a C macro that just sticks a magic preamble onto the
symbol so that we can pick it up automatically when we're scanning the
symbol table.  The dollar symbols are legal in symbols in gcc, believe
it or not.  We use them to delimit bits of the symbol.  The two numbers
following it are hex digits of a character that isn't legal in a symbol,
so $3A == ':' and $2E == '.'.  So this function is:

  soname:libc.so.6:__raise

This is plucked apart by the scanning code to set up the intercept
table.  This basically says "the function called
_vgi__soname$3Alibc$2Eso$2E6$3A__raise is an intercept for __raise in
the shared object called libc.so.6."

Ugly, but it works. :-)  If you can think of a cleaner mechanism, I'd
like to hear about it.

Regards,
 Robert.