Re: namespace namespace

[Andrew Haley <aph at redhat dot com>]

Sam Ruby wrote:
> Andrew Haley wrote:
>> Sam Ruby wrote:
>>> Andrew Haley wrote:
>>>> Sam Ruby wrote:
>>
>>>>> Is there something more I can provide?  Will gcj/g++ also be fixed?
>>>> Well, I already suggested how you might fix gcjh.  It's a matter of
>>>> whether you want to be able to access namespace namespace from CNI or
>>>> merely not have header files that choke C++.  Why not try the fix?
>>> For the moment, I'm doing the equivalent of sed scripts to post process
>>> the header files (actually, I'm using Ruby -pi -e).
>>>
>>> I can certainly fix gcjh locally, but I want to share my results with
>>> others, and ultimately the fix should go into cvs.  I'm certainly
>>> willing to do my share of the work, but you already identified the
>>> actual fix.  What's the next step?
>>
>> Someone must apply the change, make a patch, and test the results.
> 
> Hopefully the attached test case will help?
> 
> I will confess that I don't feel confident enough to dive into the code
> and work out a suitable patch to the namespace mangling portion of the
> g++ compiler.  Proposing a patch to gcjh is something I could have
> handled, but you already have done the bulk of this work.

OK.

> I discovered this bug trying to use the classes provided with libgcj for
> the purposes that these classes were intended.  My code is public and
> (if successful) will be a part of a package which is made available
> under a MIT license.

OK, so it's worth my time fixing this.

>> Given that this looks real, and it is relevant to free software, I can
>> have a look.  However, I can't go back in time, so any change that I make
>> will not affect the gcj that's in distros for some time to come.
> 
> Understood.  But "download and build the latest" is still marked
> progress over "merge this patch, built it, and hope it works".  And
> within a year or so of being committed the fix should be in the
> distributions I care about.  Meanwhile, I have workarounds.

I've spent a little while this morning looking at this.  gcj's name
mangling logic is broken: it doesn't detect C++ keywords everywhere, just
in member names.  Even when it does find a C++ keyword and attempts
to append '$' it does so incorrectly, and generates an illegal mangled
name.  There's no way that any of this can ever have worked.

I'll fix it.

Andrew.