Re: "error: unable to construct..."

From: David Abrahams (dave_at_[hidden])
Date: 2004-12-14 11:05:31

Vladimir Prus wrote:
> On Tuesday 14 December 2004 17:23, David Abrahams wrote:
> 
>> >> P.S. It's disconcerting to see ../../../../ junk in the BBv2 output; I
>> >> had hoped we'd squash that by now. It's embarrassing enough in BBv1.
>> >
>> > It would be good to fix up, but is not possible in general. For
>> > example ../../../../libs/numeric/mtl/doc/index might not be the same as
>> > ".".
>>
>> Even when the current directory ends in libs/numeric/mtl/doc/index?
> 
> In that case, ".." is not necessary. But it *might* be possible that you're in 
> libs/python .......

But that's totally detectable, is it not?

>> I thought I understood that and knew that it had something to do with
>> symlinks, but now I don't see it.
> 
> Neither do I see how symlinks would affect the picture.
> 
>> But there's no need to add the ../../../ string to a local path in the
>> first place. In fact, if what you just said is true, it proves my 
>> point: adding that string may unintentionally cause us to point at a
>> different file.
> 
> We'd need to look where this string comes from and how to eliminate ".." when 
> possible. Some time in future...

Yes. Or we could automatically remove the string as appropriate based
on the identity of the current directory, something like:

[ make-relative [ normalize-path [ root-path $(PWD) : $(relative-path)
] ] $(PWD) ]

-- 
Dave Abrahams
Boost Consulting
http://www.boost-consulting.com