Re: [boost] [1.34.0] serialization, spirit, borland and msvc-6.5

From: Roland Schwarz (roland.schwarz_at_[hidden])
Date: 2007-01-28 11:22:49

Vladimir Prus schrieb:

> 2. The 'out-of-tree' spirit 1.6 does not seem to work, for several
> reasons:
> 
> 	- There's no V2 mechanism to plug out-of-tree spirit 1.6. Both
> 	Roland and Nicola have patches to fix this.
> 
> 	- For borland, it requires patching Spirit 1.6 to change
> 	include statements from ""-includes to <>-includes.

My latest findings let me believe that there are even more issues with 
the out of tree usage. Having the entire serialization depend on the 
out-of tree 1.6 does give yet more errors.
The good news is that having only the parts of serialization depending 
on spirit including the out of tree spirit seems to work. At least 
msvc-6.5 is compilable, and testing results in a much better figure 
(still not perfect though).

I managed this by only having xml_[w]grammer.cpp and xml_[w]iarchive.cpp 
including the out of tree spirit.

I am still struggling with the <>/"" issue for borland. I hope Nicola 
will be able to provide further input.

> 	1. Nicola sends Roland and me his patched spirit.

I obtained the above results with SPIRIT_RC_1_6_2 branch. Perhaps it is 
a good idea to apply the patch to this branch? (If boost policy allows 
this.)

> 	2. Roland sets up regression tests that use the
> 	patched spirit 1.6. He changes the name of toolsets
> 	to indicate that patched spriit is used, say:
> 
> 		msvc-6.5_spirit1.6
> 
> 	and does a test run.

Ok.

> 
> 	3. If results are good enough, or can be
> 	made good enough in reasonable time, we make this
> 	patched spirit available from SF -- it can be either new
> 	release of the spirit project, or new download from the
> 	boost project with prominent warning that it's just
> 	for msvc-6.5 and borland users.

See my above comment about patching SPIRIT_RC_1_6_2.

> 
> 	4. We also mark serialization library as unusable for
> 	msvc-6.5 and borland, and include a link for
> 	the procedure to download spirit 1.6 and set it up.

Hmm, not sure. What about making the tests expecting to fail. A 
regression run that unexpectedly passes will show up with a description 
why this happened. (Out of tree spirit.) So the user will see more green 
lights.

> 2. If yes, Nicola and Roland, can you work on this plan?

Yes for me.

Roland