Re: [boost] [build] bootstrap.sh is still broken

Subject: Re: [boost] [build] bootstrap.sh is still broken
From: Eric Niebler (eniebler_at_[hidden])
Date: 2013-10-30 16:16:05

On 10/29/2013 10:36 AM, Steven Watanabe wrote:
> On 10/29/2013 10:08 AM, Eric Niebler wrote:
>> On 10/29/2013 9:55 AM, Steven Watanabe wrote:
>>>> As a curiosity, I'll
>>>> note that the project-root.jam created by "bootstrap.bat gcc" contains a
>>>> "using msvc ;" and not "using gcc ;", which seems wrong to me.
>>>
>>> bootstrap.bat sets msvc unconditionally.
>>> The language for batch scripts is really
>>> horrible, so working this out correctly
>>> is likely to be very painful.
>>
>> It is a horrible language that I know quite well. If you write the
>> pseudo-code, I can translate it into command script for you.
> 
> In tools/v2/build/engine/build.bat:
> 
> if --guess-toolset in $argv
>   call :Guess_Toolset
>   return $BOOST_JAM_TOOLSET
> 
> Also, make sure that local variables don't escape.
> I don't know how to pass the return value through setlocal.

The hackish way to do this is to write the "return value" into a file,
and then read it back in.

> (I just noticed this in bootstrap.bat:
> REM Ideally, we should obtain the toolset that build.bat has
> REM guessed. However, it uses setlocal at the start and does not
> REM export BOOST_JAM_TOOLSET, and I don't know how to do that
> REM properly. Default to msvc for now.)
> 
> in bootstrap.bat:
> 
> TOOLSET = find argument --with-toolset=xxx

Doable.

> if not $TOOLSET
>   set TOOLSET=`tools/build/v2/engine/build.bat --guess-toolset`

This doesn't seem right. build.bat takes a toolset as an optional
argument. If the user specified --with-toolset=xxx to bootstrap.bat,
then shouldn't xxx get passed as the toolset argument to build.bat?

-- 
Eric Niebler
Boost.org
http://www.boost.org