Minutes from 16 Aug 2004 release wranglers call

[Notes taken by Leon Shiman]

wrangler's mtg: monday, 16 august 2004

attendees:

jim gettys
kevin martin
stuart kreitman
leon shiman
andy richter
alan coopersmith
dan mcnichol
kristian hogsberg

sk: made dummy file for xevie, because imake macros require one. 

sk: talk about putting legitimate header file in that . don't have wire.h 
proto.h . will try to fix that today.

jg: is nice convention.

km:status

improvements in composite over weekend. spent weekend with xtestsuite.
xtest fails with composite enabled. many failures to be addressed. is
off by default. failed almost every test. there are bugus issues having
to do with timing acc to keith. there seem to be many things. jg asks if
the test suite is at fault, bec it has no strong synchronization. does a
get image. can't you run the suite more slowly?

are the vsw4 in same place as config files for vsw5?

need to look into testfailures for composite.

.pc.in also caused trouble with xevie, sk wants guidance. kp won't have
net access. alan c. might be able to help. kristian will pos info to
wrangler's list.

sk can make hdr of hdr files to point; cojld also look at proto and wire
files. is it necesssary? sk will look into; not top priority. mainly get
into separate library. current users of xevie don't need 1.1. there is
1.1 xevie, there is some working code there,but needs to be hashed
around re multiple clients using it, with lock for unique user. this is
now xevie 1.0, and disabled by default bec of those issues.

since complete, will now move over to documentation bug. would like any
results re hdr files. pick up on wednesday. look for insert into release
notes. putting bugs into 999 when they are ... 999 bug is the release
notes bug, to assure that all issues are covered.

re 1060- info: mike raised issues on wrangler's list. xprint went in and
changed interface to xaw and added code to clients without condition.
interface change wasn't noticed, and so now builds break without that
info. so will remove code from xaw and let roland make separate lib for
after code freeze. want kristian to summarize patch changes: first,
xevie changes were small. a few clients changed. two approaches
reverting changes to older versions; otherwise compile without xprint.
changes to lib imake files. build xaw without. bunch of xpxxx clients.
so build now only if xprint there. all details in the change log of
1060.

we all are to blame for process, not noticing proposals.  consequences
of what to do about this. problem of hanlding shared libraries. roland
can have a libxawprint libraries. plan is to revisit this once roland is
avaiable and there is time to explore thoroughly. there is other issues:
use of separate libraries. we may not want more monolithically.  roland
has been trying to pull more into the tree. we need more discussion.
we'll leave it off until roland is able to get back to it. R200 bug is
also serious.

LS: This is not a long term policy decision, which needs time and
thought to resolve.

1029 - hard failure when socket dirs not chown'd to root. changes cause
hard failure. RH can work around it. can creat dir in start up scripts.
if there can be removed and rc init scripts can create it.

a true make install can do that too.  be sure that make install creates
them, doc in release notes, get vendors to send us pam modules, etc.
want to ensure that if accidents happens... want vendors to send in
solutions.

Question: do we make a helper? see alan cox. Second: do we need to put
through cert? re helper: in or out? should we talk to cert, and ask them
what to do. this determines whether we want to document in release
notes. we do need make install change. in full install there's no
problem in deleting and re-creasting. if fullinstall, hopefully not
running X, bec may cause trouble. Is it wise to remove them?  j: trying
to start a new app by UDS's will cause trouble. Can do a check, looking
to see if correct. unconditional removable is bad. prob with xfs. not
everyone runs font servers. xfs would be among the first to go.
propose: if correct, leave; if not correct, repalce. make install runs
as root. may need to create. if they exist, then chg permissions and
ownership. what should the correct permissions be? jg: xserver creates
them the right way, i think. three dirs:.x11-unix, .font-unix,
.ice-unix. owner might be whatever UID the xserver is running as.
claiming root is false. same ownership as xserver, and .font owned by
font-server. currently fails if not root. now get failure bec ownership
is wrong when xserver starts. can solve at redhat. worried about other
platforms.

jg and km will talk to cert. how document and disseminate needs to be
cleared up. no strong feeling to revert behaviour. hard-fail is
conservative thing to do.

594- large pixmap from ghostview. kp suggested this could be out of
memory problem by kernel. on systems with small amts of memory, fails
immediately. very large swap is ok. if you run mozilla, you'll find
150mb easily. problem is old code and requires huge pixmaps. protocol
says 32kx32k. should we lose sooner? not a soluble problem. won't happen
with old apps. gnome system monitor reports pixmap usage. xserver should
not crash, even when demands are too large. bad alloc failure is ok. is
it a platform dependent option to be warned. different unixes have
different policies re allocation issues.

"optimistic allocation" policy needs to be thought through. there may be
a real bug here now. should we remove it as a blocker. we can put a size
limit, and then return a bad alloc error. should it be in create pixmap,
window, etc. there may be more of these. audit create pixmap in server??
talked wit KP about putting in dispatch code. jg in favor of giving bad
alloc when >32768. some cases may not be able to properly handled. this
needs to be handled carefully. don't want to create partial and possibly
incorrect solution. will leave bug open, but will be off blocker list if
there is not another real bug. put on doc bug 999 optimistic allocators
reminder. defer larger issue re regions vs pixmaps.

770- main R200 on 3d. eric noted working on soultion. leave for now.

964- chkd in chg but leaving bug open for better changes. MH notified.

those open issues will now go to a hold-open bugs. find a convenient
number.

993 - xevie lib mkoving to its own lib. sk checked in main chgs. still
accepting chgs for hdr issue. desire to work out hdr issues.  no longer
a blocker.

999 still a blocker. soren and diego said they might help. 

1032- need soln to darwin crashes. main blocker bug has been fixed. but
move to held open.

1057- simple chg being tested on 64 bit systems w/ dri. also checking on
others. once in suggest that move over.

1065- same as R200, so think R100 can be similarly fixed. can't turn
render on without fix. consdier R100/R200 a blocker bug. want both of
these open. Andy Richter will check on nvidia drivers.

1072- kbd driver fallback. alaising kbd to old keyboard driver. kristian
has patch. working out details. once finished, looks good.

prob is that testing is incomplete. ok with code freeze. still not
total. check-ins only with permission. i've spoken with blocker issues.
if there are others, i'll manage it. so permission process is get bug
and patch into bugzilla, then mail wrangler's to add. kevin in chg. need
comments about level of need.

deron johnson is unclear about brancdhes and labels.

km: i'll continue to work on trunk until release. for this release,
since time is short, tempted to continue on trunk until release, then
branch -- but please send out message that no check-in on head without
approval. we're in code freeze. note will be sent. there are branch and
regualr tags. snapshot tags are not for branch points. just noting ckout
points. ckout on head is ckout without any tags.

will continue on trunk for now. will then tag tree, and send notice.

TESTIING:

trying to get vsw4 working from lsb. lots of problems. so since i have
another soln: a test suite in dri tree, and xreg script in top dri
level. strongly considering moving from lsb-vsw4 to dri test. used with
xfree86 and dri for a while. can document differences between tests.
looking for simple soln. jg got it to run. prob is doing one test, and
running with dummy driver and anything other than xvfb. xreg is script
allowing various tests: xtest, xperf, etc. is just a bourne shell
script. updated to run basic xtest, stores results and summary. runs
multiple depths. makes tests easy. prob with vsw4 is ease of operations.
if no objections, I'll update wiki status page with details for use. not
aware of any license issues. has been in xf86 for years. been using it
since 95. only used on linux so far.

several test failures so far. 8bit test fails on xvfb. A greast number
of tests fail for XDrawText. standard tests fail for XDrawArcs.

tracking down test failures, and radeon driver, problemss with grab
keyboard. no successful test run so far. build and install run fine. but
conformance and xtest on real drivers don't.

TINDERBOX:

no updates for now. leave it for now, but needs to be pursued.

DOCUMENTATION:

tools for building are antique. need modern tools. need someone in
charge of this. need to move from troff and old sgml. eric raymond's
doclifter. need team to help. need modernized style sheets, etc.


SUMMARY

tree tagged this pm. will point to new xtest site. builds: dan reported
fixes. dan mcnichol will add to bugzilla. ibm-config files can go in
late. alan is building but no yet testing on solaris.