From: Reece Dunn (msclrhd_at_[hidden])
Date: 2005-10-24 04:29:16
Zbynek Winkler <zw-bjam <at> robotika.cz> writes:
> Reece Dunn wrote:
> >If you, Volodya or anyone else has a better solution feel free to add it
> >to the discussion.
> The response files are "just a workaround" for a limited command line
> length, right?
Yes. They allow you to pass defines and include directories that would go beyond
the command line limit.
> If this is true, would it be worth to consider just creating the
> response file in the action? I mean that the response file would not be
> a target at all, the build engine would never see it... The information
> to create the rsp is already there and it would have the right semantic
> (recreate the rsp each time the target is rebuild) and we could easily
> keep the rsps around...
The problem is that the response file generation logic is currently tied into the
rule+action architecture, so removing them as a target would (currently) break
this. I am unsure how tightly they are bound to this arcitecture, so don't know
the impact of removing them as a target.
Another thing is that the response file is placed in the debug/gcc, etc. build
directory, but this can be worked around by using $(>).rsp.
Your idea seems sound in theory, I would need some time to work on the
implementation as I only have a limited knowledge of how response files work.
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