Minutes from 13 Aug 2004 Release Wranglers call

[Kevin E Martin]

[Notes taken by Leon]

on call:

deron johnson
stuart kreitman  
kevin martin
jim gettys
keith packard
alan coopersmith
dan mcnichol
leon shiman

STATUS REVIEW
-------------
composite is working, damage still has small bugs

in composite, problem was that render was buggy and hadn't been
integrated.

rendertest not run yet, should be done, need some work moved from
modular tree. ones found were the render clipping was broken. that is
checked in.

problem with getting damage events.

propose extending code freeze 'til monday morning. for code freeze,
make it monday morning, after freeze would like to take official lock on
tree (kevin). all changes after that must be in bugzilla, and kevin has
to approve. basic idea: cut RC1 on monday morning. code in tree, all
have to do is add snapshot numers change to 901, after that i'll cut
RC's intermittently, closing bugs as they are known to be fixed.

kevin would like to update test matrix with RC agaisnt which they have
been tested, changing color accordingly.

monday morning at call time: 11am edt. then tree will be tagged, and no
more changes after that except ones approved from bugzilla.

jim asked to put RC in the wiki page, will be a top of page; will also
put in tag.

tinderbox can make this easier for next release.

km: took about 8 hours of wall time, 30 minutes of km's time to test
builds. will be time consuming. be prepared for people who signed up
and can't do it. may have to back some out of test marix.

RC's expected about every 3 days.

mike harris will get into rawhide, and kristian hook to continue to
build his; see his FDO page.

km: i'll contact the rest.

BLOCKER LIST:
-------------
594: bounds wrong on ps file. kp: problem with 32-64k hard to use
bounds check bec of ints and shorts. prob is rep of coordinate space of
drawables, and for the effective coord space, which is 0-32k. jg: paper
published about deficiencies.

permits geometries basd alloc for >32k pixmaps. several unpublished
fixes. km will deal with this.

687: matrox: there is more history than report gives. patch is ok, alan
cox ref in bug is to xfree86. traced back to alan.

770: km: one of issues with dri. never have perfect code. if eric can
fix, will allow fix, if he can't, won't worry about it. if have fix
before shippping, will do. marking: keep in blocker list

964: km: have reproduced. looking at matthieu's code to merge mult error
msgs into 1, think code is ok, but worry that xprt is tickling it, and
we're just fixing symptom. leave on blocker list for now. orig was to
limit # of error msgs to var/log. prob of brute force magic cookie from
outside -- chances near zero. can check patch in and deal later.  think
this is prob may be with xprt and not audit code. so will back out
matthieu's change until we have a better solution. now: patch worse than
bug. will add note that if there are other fixes found later will look
at them.

991: visuals for composite. km: fbconfig. fine to leave it as is. since
ext is disabled by default. may change in future. useful in glx, normal
visuals for non-users, others will have metadata structure popinting
back to visuals. is a design change. builtin compositing mgr works with
this visual. can have apps use state in compoisiting mgr check stastge
and olny compositie if ARGB. compositing extension is not broken. only
when you have external CM can it go awry. kp: in dix set all alpha bits
to 1. expose visual, bec is a new visual new depth. require new depth.
in xserver no semantics, so need convention. but doesn't

flash player mistakenly uses this visual. flash has a bug. doesn't know
what visual mozilla is using. kluge around reqd kp: i'll add an xlib
patch, whhich will permit Mozila use. in redhat can modify mozilla env
variable.

kp will write this up for bugzilla. incl doc on xlib change. once closed
will transfer to release notes bug. don't close 'til chagne from 351 to
999.

993: sk move xevie to own lib. will be put in this morning, so will
close today. xevie1.0 .so1.0 . xevie may change in next few months. is
an example for users. got spec on wiki. is 1.1 spec, so will edit saying
that 6.8 doesn't have block feature

999: need to get going

1024: paul didn't get to it before leaving. suggest that alexander
should commit, hoping that it can be tested. best til alex returns.
committed for now.

1029: same security bug. who can use xtrans to create dirs: any x app,
or Ice. just servers. only dirs created are X11 and ICE. if just these
two dirs asre created, then might fix in startup scripts.  kp: what do
we do: cox suggestion? can create a smart app, dealing with security
there, or can try to be smarter in our solution, open to sugg's. either
leave as is, documenting it, so that users will update script, or back
out and allow insecure behaviour. no time to explore secure secondary
app. alan cox revoked claim of easy solution. need to document, but who
do we send doc to? a cert advisory? is easy for us to change in redhat.
for others may not be so easy.

sysems wihtout setuid will have to back out change. for others, don't
know how will affect. that's my worry. i know that i'll get bug reports
at rh, from people who build and then have failures. if those dirs get
the wrong permissions, we're in trouble. do systems fail hard.

leave as open questin 'til monday. must resolve on monday.

1032: damage. doesn't crash it, but still broken. is this major blocker
still? fix would be good, leave it. can move off blocker list later.

1041 - has sinmple fix. suggest applying, think was oversight. think
won't fail, so will apply patch.

1050 - related patch for randr. patch looks innocuous. stores monitor
sizes for ease of update. will apply both.

1042 - tell Kristian to apply patch today

1055 - mike's issue. better be able to build without xprint. includes
1060. 1055 is easier. logic is consistent. alan coopersmith will test
fix today.

1060 - is harder of 2. need to back out or make conditional.  Xaw.
selfcontained. one file and header change. is symptom of bigger problem.
need to monitor. shouldn't extend api without architecture discussion.
think can be external library. can propose external library. propose
pullling code out. if it can be done as an external lib ok. can't be
stubbed out. xaw is sometimes used. allow roland to add this after code
freeze. even if hooks in xaw are required, after review. no other known
depdencies on XP.  will pull code at this point and encourage roland to
build separate library, and add it.

1062 - will add fix.

add VT switching to run-test list.

1065 - leave as is for now.

1068 - new kbd driver: new driver only works on linux and bsd. problem
of breaking use deprecated unless linux or bsd. alan will file a bug and
put it in

1068 - new kbd driver. don't compile kbd driver in. can print warning.

SUMMARY:
--------
km: majority done by monday.

Deron's issue: adding code to server

code for project LG, enables my display server, allows me to test
composite in my env. changes mostly in events.c new lib libXLGT. changes
mostly in dix, some in mi validate tgree, some in cw. have it in ifdefs
mostly, build lgswitch.

kp: should be on branch.  SK: deron wants to put his hooks i now. can't
we put hooks in later. DRI development

deron will create branch and check in, and will then close bug. will
then merge with head later.

dan mcn: hard getting stuff to build on aix. seem to be many small
problems.

tinderbox is outbound only. firewall issues with tinderbox

NEXT CALL MONDAY