Subject: Re: efficiency
To: David Maxwell <>
From: David Laight <>
List: current-users
Date: 11/05/2003 22:24:39
> I tried to write some changes to (can't recall the details now)
> and it took a while to realize why the information I needed to make the
> change wouldn't have been available until a few steps later in

I'm not even sure Luke knows everything that goes one, some is magic,
some just mystical.

> Likely, the trouble with de-sequencing delete and compile is that you
> would need a (maintained) in-order list of dependencies, because you
> need to delete/build, following the tree from the bottom up. I suspect
> that list is difficult to create/maintain.

For 'normal' use an update build is fine, these run relatively quickly.

> Deleting and mkobjdirs could probably be merged. Simply make the targets
> set some variables, and create a new 'updatetree' target that actually
> does the walk of the source tree, and does each function in order
> (delete, then mkobjdir ;-) in each location.

I'd go for removing the mkobjdir pass, and hacking make to create the
objdir when it tries to 'cd' into it.

> Depending is probably tough to split from its own pass in make - or, put
> another way, building dependencies may hit enough other dirs that I'm
> not sure the cache would help much.

dependall already does the depend and all (aka build) bits in one pass.
(except in the bits of teh tree where it decides not to bother with the
depend part!).


David Laight: