On Wed, Feb 27, 2008 at 11:17:34AM -0800, Mike.Sullivan at sun.com wrote:
> >From Nicolas.Williams at sun.com Wed Feb 27 10:57:41 2008
> 
> >> Then again, if you're putting it into the proto area anyway then moving &
> >> deleting I don't know why you just do the install into the right spots in
> >> the first place, then delete the bits you don't want? Though perhaps
> >> there's a lot more to delete individually then.
> >
> >Because that means knowing what those bits are.  What if the next
> >version of SQLite3 installs lemon (a parser generator)?
> 
> If you're already cherry-picking what you want don't you already
> have that problem? Or are you saying you _don't_ want it to install

No, because the things I want are stable -- they won't go away.

> more things like that, which you then forget to delete? Well if

That, yes.

> you don't want to ship lemon, but it does, I'd actually side with
> it as I tend to think if we claim we're shipping package 'foo' then
> we really did ship everything you'd get if you built it yourself.

But I think that's something the next integrator needs to be on the hook
for.

> Yeah I agree with not shipping things like static libraries in
> most cases, but at least that's usually controlled by a configure option.

Right.  In this case there's an option for it, but it doesn't work
(i.e., configure takes a --disable-static option, but the makefile
doesn't honor it).

> >I'd much rather have full control over what gets installed.
> 
> You do either way :) And I'd think that this part of the work is
> the easy part of doing upgrades. After all, if the Makefiles can
> be told to install into $ROOT properly then pkgmk and protocmp will
> notice when things are added/removed. When you have a hardcoded list you
> won't know unless something is deleted, and it will still be easy
> for you to miss, say, a new header file that needs to be shipped
> because an existing one now requires it.

I still have the problem that I have to run *two* configure && gmake
installs, one for 32-bit and one for 64-bit and I don't want to have to
make sure that they don't step on each other (there's lots of paths to
check).

Folks, I'm trying to make things simple.  And all I see is complications
arising here.  I find it a lot simpler to install into a tmp DESTDIR and
cherry pick.  I'll stick with it unless someone pulls rank (Mike, I
guess that'd be you :)

> The hard part of upgrades for me is determining if there are any
> incompatible changes, the paperwork, the testing...more so if anything
> is using it. The packaging is pretty simple in comparison.
> 
> Of course I won't force you to do it either way, this is all just
> my opinion from my minimal work in sfw.

OK, so I'll keep it using tmp DESTDIR.

BTW, I've put many hours into trying this various ways, and over the
weekend too.  I really, *realy* don't want to experiment anymore before
integrating.  Think of future volunteers.

> btw your install script shouldn't really be creating the usr/lib/*
> symlinks if they aren't there. You're supposed to have run
> 'make setup' at the top level if you're building by hand, and if
> you haven't done that you should have worse problems than the
> links :)

OK.

Reply via email to