From: Vladimir Prus (ghost_at_[hidden])
Date: 2003-06-09 07:05:30
David Abrahams wrote:
> >> What about replacing the complicated implementation in path.jam with
> >> something based on normalize-raw-paths?
> > The problem is that 'normalize-raw-paths' does something less --- it
> > cannot join two paths together.
> Most of what join is doing happens after the split.
Actually, that happens only when join is called from 'make'. In other
situations, like "join a/b/c ../d", nothing is split.
> You realize that
> normalize-raw-paths is really based on simplify-path-tokens from
No. I forgotten about that rule.
> > I'll probably see how path.join can be improved, by anyway: unless
> > path.jam is moved to "kernel", "modules.jam" cannot benefit from it.
> >> > Besides, the email you've responded to questioned the need to call
> >> > 'normalize-raw-paths' at all. Do you have any comments on this?
> >> I put it in there because I wanted to be able to move modules from
> >> "tools/new" to tools/xxx as follows:
> >> 1. Copy the file to tools/xxx
> >> 2. Modify the version in tools/new to generate an error when
> >> executed, to be sure it isn't getting used.
> > Why not "cvs rm -f tools/new/something.jam"? Or just "rm
> > tools/new/something.jam"?
> I'm not sure, honestly.
> >> 3. Run all the tests.
> >> The problem in that case is that other modules still in tools/new
> >> would always pick up the wrong version without even trying.
> >> It could be seen, I suppose, as a transitional measure.
> > I'm probably missing something, but you can move files without this
> > measure.
> It's true.
Does that mean we can just add directory of importing module to the *end* of
search path, and avoid the call to 'normalize-raw-paths' altogether?
Boost-Build list run by bdawes at acm.org, david.abrahams at rcn.com, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk