Re: c++/9942: Function Overload Resolution calls wrong function onvariable length parameter lists

[Wolfgang Bangerth <bangerth at ticam dot utexas dot edu>]

> How often are you finding that having both preprocessed and raw c++ files
> would be useful?

Not very often. At least not often enough that this...

> Perhaps we should amend thebug reporting instructions along
> the lines of
> please provide
> * preprocessed .ii file
> * a *single* .cc file that #includes stl headers (but not any of your own)

...would be worthwhile. I think the second point is just too impractical, 
since one would have to hand-include all non-standard headers, and you 
know how entangled these can be in larger projects. I'd expect we get more 
unusable attempts at that than helpful stuff.

Besides, when you have a little experience, most of the time reducing test 
cases is spent in reducing user code -- which I don't know and have to 
find my way around in. It's not so difficult to spot the boundaries 
between user code and system/stl headers, so once you've reduced user code 
to 20 or so lines, you'd just block delete 1000s of lines, which is about 
as complicated as removing #include statements.

There is only one thing that I'm worried about with .ii files: if someone
created her .ii files with something pre-3.4, then we can't compile it
with the new parser any more (due to all the missing this-> stuff in
pre-3.4 STL headers). So if we get a report against 3.2, and I find that
it is fixed in 3.2.2 or 3.3 by running these compilers against the whole
.ii file, I can only check that it is also fixed in mainline by reducing
the testcase to something that doesn't have any STL stuff in it anymore.  
That's tedious, and usually only yields the expected result that mainline
is also fixed...

W.

-------------------------------------------------------------------------
Wolfgang Bangerth             email:            bangerth at ticam dot utexas dot edu
                              www: http://www.ticam.utexas.edu/~bangerth/