ports broken on amd64 [was: Re: Intel C2D COREs not used equally in FreeBSD 7.0-CURRENT i386]

[Mark Linimon]

[Subject changed and amd64@ added]

On Tue, Jun 12, 2007 at 03:23:11PM -0700, Darren Pilgrim wrote:
> Is there a list of known amd64-broken ports or have they all been 
> flagged in the tree with (NOT|ONLY)_FOR_ARCHS variables?

They are flagged BROKEN when kris notices a bad result from the build
cluster (pointyhat.freebsd.org).  For a long time, people have used those
*_FOR_ARCHS as a shorthand for conditionally marking them broken, in some
cases only because they didn't have an amd64 to test with.  These ports
should be changed over; my feeling is that the true use of *_FOR_ARCHS
should only be for "port cannot be made to work on this architecture".

In individual ports, BROKEN is used for compile failures, and IGNORE is
used for "does not actually work in this situation" (e.g. core dump), as
well as other cases such as CONFLICTS.

(This is the short summary; there is much more in the Porter's Handbook).

Part of my work for my BSDCan paper was to teach portsmon to create
dependency tree displays for ports that do not correctly package per-
buildenv.  I have it working statically but not on-demand; OTOH I am
still behind from my trip afterwards.  However, you can look at the
bar-chart comparison results as of a month ago at:

  http://people.freebsd.org/~linimon/portsmon2007/package_comparison.html 

which will lead you to the individual pie charts (e.g.):

  http://people.freebsd.org/~linimon/portsmon2007/package_status.amd64-6.html
  http://people.freebsd.org/~linimon/portsmon2007/package_status.i386-6.html

These _should_, but don't :-), lead you to the breakdown of failures by
buildenv:

  http://people.freebsd.org/~linimon/tmp/package_failures.amd64-6.html
  http://people.freebsd.org/~linimon/tmp/package_failures.i386-6.html

*note*: as a simplification, non-built packages that are leaves are not
yet shown in the latter two.  This is because if I put them all on there,
it gives graphviz indigestion.  This is a bug.

This is all Work-In-Progress.

Final note: both portsmon and FreshPorts currently only model the
"state" of a port on i386-6, so they give you a baseline, but not a way
to compare vs. e.g. amd64-6.  So you have to do a little investigative
work with either these charts (and portsmon pages for individual ports,
which will show you their build status across buildenvs), or the pointyhat
log pages.

mcl