Re: [v2] not responding to toolset

From: David Abrahams (dave_at_[hidden])
Date: 2005-03-14 13:23:43

Joao Abecasis <jpabecasis_at_[hidden]> writes:

> David Abrahams wrote:
>> Joao Abecasis <jpabecasis_at_[hidden]> writes:
>> 
>> 
>>>David Abrahams wrote:
>>>
>>>>in tools/quickbook/doc I did:
>>>>
>>>> bash -c "export ICU_PATH=/cygdrive/c/src/icu && bjam -a --v2 toolset=gcc pdf"
>>>>
>>>>But it continued to try to use vc7.1 until I moved
>>>>
>>>> using gcc ;
>>>>
>>>>to the top of my site-config.jam
>>>
>>>I had noticed this too... In order to have quickbook compiled only once 
>>>quickbook is always compiled with the "default" toolset. In practice, we 
>>>get an empty property-set:
>>>
>>> local quickbook-property-set = [ property-set.empty ] ;
>>>
>>>This is not the toolset specified on the command line but the first one 
>>>ever defined. (<variant>release is also explicitly added which may not 
>>>always be desirable...)
>
> I'm not sure how I can fix the above issue... The original motivation 
> for choosing a single toolset was to avoid having multiple quickbook 
> executables built for the sole purpose of generating docs.
>
> This design allows one to generate docs by specifying only the output 
> format on the command-line:
>
> bjam --v2 pdf
>
> Quickbook, boostbook et al. are implementation details.

Fine. But why not respond when the user requests a specific toolset
explicitly?

> There may be other ways to select the "default" toolset but, given my 
> limited knowledge of Boost.Build, I'm at loss here.
>
> Maybe the question could be which toolset an empty property-set selects. 
> Should a toolset explicitly stated on the command-line take precedence?

Yes.

>> Certainly *something* is making Cygwin GCC churn forever when it
>> compiles quickbook. Maybe it's the optimizer? Usually that sort of
>> thing is not the optimizer, but instead due to too much debug info
>> generated by templates...
>
> Under linux, the memory footprint (~100MB) and duration (2-4 mins) of 
> quickbook compilation with gcc doesn't change significantly going from 
> debug to release mode. With the intel compiler the memory footprint 
> increases a bit more, from 100 (debug) to 150MB (release).
>
> I suppose Cygwin GCC should parallel plain gcc behaviour.

Probably.

-- 
Dave Abrahams
Boost Consulting
www.boost-consulting.com