Re: Patch for vacpp-tools.jam

From: Markus Schöpflin (markus.schoepflin_at_[hidden])
Date: 2003-03-05 07:59:31

David Abrahams wrote:

> Sorry it took so long to reply to this...
> 
> Markus Schöpflin <markus.schoepflin_at_[hidden]> writes:
> 
>>David Abrahams wrote:
>>
>>>Extending a toolset to override installation directory requires
>>>that the directory be set as a variable "on" the target, or you
>>>will just end up with one global setting for installation directory
>>>and you'll just use *one* of the actual toosets (whichever one gets
>>>used to generate the last target). Try it with -n -a and inspect
>>>the command-lines to see what I mean.
>>
>>I don't understand what you mean. I have been using this with two
>>toolsets (va5 and va6) which extend vacpp and just override VA_BIN to
>>genrate the regression tests and I can't find anything wrong with
>>it. 
> 
> Did you try running regression tests with both toolsets at once,
> using more than one target? Do the command-lines actually invoke the
> two different compilers you want in each case?

Nope, I just used the current shell script to run the regression 
tests. The script invokes bjam seperately for each toolset. I didn't 
try to bjam -sTOOLS="va5 va6" and I think I see how this could cause 
problems.

>>How should it look like to be correct?
>>
>>>I also worry a bit about the use of very generic global variable
>>>names which appear to be picked up from the environment (e.g. CXX).
>>
>>They are set by bjam build flags and it has been like that for quite a
>>while now. IIRC, I asked on the list if this is ok.
>>
>>
>>>The philosophy for most toolsets has been to try to avoid relying
>>>on environment variables whenever possible. I think that this
>>>criticism applies to the current state of the toolset also.
>>>
>>Hmm, should I just rename the variables?
> 
> I'm not sure. It probably doesn't matter all that much; I hope we'll
> be on to BBv2 by the time the next release branch is made...

Agreed. If you don't touch the current vacpp toolset it works as it 
worked before. And it gives me the possibility to run the regression 
tests for VA5 and VA6 by just extending the current toolset as each 
toolset gets it's own regression run by the current shell script.

Markus