Re: Shortcomings of HPD '#' identifier qualification syntax

[Andrew Cagney <cagney at redhat dot com>]

Sami and I were brainstorming some #-parsing vs lexing problems friday

We came up with a rough rule of thumb that given a #, everything until 
until a space (and a finite small set of characters) would be included 
in the token.  Without this something like:
 (fhpd) print #libc#environ  + #libc.co#errno
would not end how you'd expect.  Similarly, this gives us a way to 
handle the "." in
 (fhpd) print #solib#astruct .member
 (fhpd) print #solib#afile.c#10
The consequence is that extra tokens such as "@" and ":" can be made to 
work.

Andrew

Petr Machata wrote:
> HPD syntax misses certain features that I'd like to use in ftrace (and
> who knows, maybe fhpd users would appreciate them, too).  So here goes
> my bikeshed problem:
>
> 1) You can't specify ELF symbol version,
> 2) You can't specify PLT entry
>
> Number 1 can be solved by simply allowing the symbol name to have a form
> of "symbolname@version".  "@" is good for that purpose, because it's
> non-ambiguous, and already used by readelf to refer to symbol versions.
>
>   
i noticed, will help with cut/paste.
> Number 2 can be resolved in any number of ways.  In ftrace, I currently
> output traced PLT entries like "#soname#plt:symbolname".  This looks
> nice, seems to be easy to remember, but would be hard to parse in the
> context of ?: ternary expression.  objdump uses notation of the form of
> "symbol@PLT", which would collide with symbol version syntax.  Maybe
> something along the lines of ".plt:symbol"?
>   

> Opinions?  Ideas?
>
> PM
>