Re: Patch: New implementation of -Wstrict-aliasing

[Silvius Rus <rus at google dot com>]

Andrew Pinski wrote:
> [...]
> I also think the warning should not really be using %s at all, there are
> pretty-printers for a reason, they really should be quoted if %s is to
> be used.  
1. I switched to using %H do display location information, but got into 
problems.  It appears that only the first instance of %H is printed.

location_t location1, location2, location3;
...
warning (OPT_Wstrict_aliasing, "%H defined at %H is used at %H.",
        &location1, &location2, &location3);
Here is the last part of the output:
[...]  defined at  is used at .

2. Also, I am using "%s" to print a preceding "*" to make it clear when 
the warning involves pointers.  It is confusing to say that PTR is 
aliased to OBJECT, when in reality it is *PTR that is aliased to 
OBJECT.   This gets worse when OBJECT is a pointer itself.  Since the 
analysis is based on points-to information, I cannot always get accurate 
line numbers, so displaying as much information about the objects is 
important.  I do not understand how I could use pretty printers here.
I could write separate versions of the warning, with and without the 
"*", but with two variables that would mean 4 versions of the 
essentially same thing.

3. I could not find documentation for format extensions and it is not 
clear to me which of them I can use from the backend.  Could you please 
point me to it if there is any (or to locations in the source code)?

Thank you,
Silvius