On Tue, 18 Jan 2005, Thomas Klausner wrote:

> For the current state of affairs the policy is:
> . BUILDLINK_DEPENDS bump on major API changes and shlib major bumps
>   (i.e. when package stop building or when binary packages stop working
>   because the expected shlib major is not found)

I am not sure if that is how it currently or should be done.

The BUILDLINK_DEPENDS is bumped when packages depending on it stop
building (because of ABI change like you mentioned).

I may be reading your "BUILDLINK_DEPENDS bump on ... shlib major bumps"
and "when binary packages stop working because the expected shlib major
is not found" wrong.

But BUILDLINK_RECOMMENDED should be used for SONAME changes.

If we bump the BUILDLINK_DEPENDS for shlib major bumps, then the
IGNORE_RECOMMENDED=yes is useless.

Some examples I looked at:

x11/gtk2/buildlink3.mk
  BUILDLINK_DEPENDS.gtk2+=        gtk2+>=2.4.0
  BUILDLINK_RECOMMENDED.gtk2+=    gtk2+>=2.6.0nb1

graphics/png/buildlink3.mk
  BUILDLINK_DEPENDS.png+=         png>=1.2.4
  BUILDLINK_RECOMMENDED.png?=     png>=1.2.6nb1

graphics/tiff/buildlink3.mk
  BUILDLINK_DEPENDS.tiff+=        tiff>=3.6.1
  BUILDLINK_RECOMMENDED.tiff+=    tiff>=3.7.1nb1

>   This also incurs a recursive PKGREVISION and BUILDLINK_DEPENDS
>   bump for all dependencies because of how our tools work, currently.[*]

Or attempt to figure out what packages actually need the library and only
bump them.

> . BUILDLINK_RECOMMENDED bump for security fixes.

Okay.

 Jeremy C. Reed

 	  	 	 BSD News, BSD tutorials, BSD links
	  	 	 http://www.bsdnewsletter.com/