-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

MDJ 2003.05.11 (May 11, 2003)
=============================

   Copyright 2003, GCSF Incorporated.  All rights reserved.


Top of the Day
--------------

**More Mac OS X news**

  The new Mac OS X 10.2.6 update does what a bug fix update
  should: it fixes critical customer bugs quickly in a small
  update that's understandable and almost completely documented,
  we think.  It solves problems with USB hubs that showed up with
  10.2.5, even though they weren't Apple's fault. It fixes
  AppleShare file corruption, though the user-level release notes
  don't admit it. There are OpenGL improvements, support for a new
  phone in Address Book, some printing improvements, and that's
  about it. In today's double-length feature, we examine the
  updates, clarify the confusing Mac OS X Server 10.2.6 update,
  and debunk the myths already circulating about exactly what the
  update and its updaters do. Renew your acquaintance with the
  state of Mac OS X art today.

**Up next**

  Our first look at the best new disk utility of the year:
  Alsoft's DiskWarrior 3. Stay tuned!


Mac OS X 10.2.6: A small, efficient update
------------------------------------------

**6MB revision fixes hot-button 10.2.5 issues**

  Imagine for a moment that you write USB device drivers for Apple
  Computer.

  In Mac OS X, device drivers are kernel extensions - code that
  runs inside the Mach Kernel's address space, just like Photoshop
  plug-ins run in Photoshop's address space. If a Photoshop
  plug-in crashes, it can bring down the entire application, but
  if the kernel crashes, the OS halts. The kernel can't
  "unexpectedly quit." Those crashes are kernel panics, and they
  bring down the entire system.

  In short, kernel panics are bad. Kernel-resident code therefore
  must be exceptionally bulletproof.

  So you're working on improving the USB support in the operating
  system, and you wind up in the code for when the Universal
  Serial Bus is reset. It happens more often than you might think,
  including when devices are attached or removed, or perhaps even
  when drivers load or unload. You're looking for some problems in
  the response to a bus reset. The USB specification requires that
  all devices respond to an enumeration request after the bus
  resets, even before they're assigned device IDs and the bus is
  ready for normal traffic.

  Maybe the enumeration code somewhere else is not working
  properly and that's interfering with recognizing some devices
  when they're plugged in. Or maybe some devices only respond
  properly if enumerated after bus reset. You can't enumerate the
  bus twice to solve the problem later, but as you debug it, you
  realize you don't have to. The specification says enumeration
  _must_ work here, so you change the code to rely on that. The
  old code allowed for devices that didn't respond, but the
  specification says they _have_ to respond, so you cut out the
  old code that caused trouble on some devices.

  You could just leave it that way, but since you're a trained
  programmer, you probably put in a line that raises an exception -
  an error condition - if a device doesn't respond properly. You
  raise the error because if your code gets to that point, the USB
  bus would be in an unsupportable state and your code wouldn't
  know how to continue. When you're about to go wandering in the
  weeds, it's better to raise a flag than cause more trouble. Any
  error would panic the kernel, but the line of code that raises
  the exception should never get called - the USB specification
  _requires__devices to respond. Specifications exist to mandate
  such rules.

  So you build your change. It seems to work, and the original
  problem devices now behave. You send it off to Apple's test
  engineers, who try it with a few hundred USB devices, and none
  of the engineers find any problems. It gets seeded to external
  developers, and they don't report any specific problems. You
  therefore consider it another problem solved and go about more
  USB business. You look forward to the next release of Mac OS X,
  so people who want this fix can have it.

  One day, your fix ships as part of Mac OS X 10.2.5. Of course,
  there are always reports of individual problems with any new OS
  release, but most of them are hard to verify, or are from people
  blaming the OS for bad programs or bad middleware. It's only the
  patterns that are scary.

  Understandably, you're not happy when a pattern emerges.

  The customer support line and multiple Web sites are reporting a
  kernel panic several minutes after booting, but only if some USB
  devices are connected. You get this sinking feeling in your
  stomach, because that's _your_ area, and you changed the code
  for this release. So you go back and look, and you see no
  problems: your code is up to spec and well-tested.

  The customer service folks find a few common threads in the
  complaints, so they send along the names of a few USB hubs that
  several people mentioned as causing kernel panics when
  connected. You run out to Fry's and buy one, you plug it in, and
  sure enough, the kernel panics.

  Your heart _sinks_, because people using your code are seeing
  their systems crash when they didn't before. You have to find
  out what's causing this and fix it in the very next update.

  After some debugging, it turns out that it's not your fault
  after all. A fair number of USB hubs use some chips or software
  that disobeys the specification by not responding to enumeration
  requests after bus reset. The specification clearly says they
  _must_ do that, but they don't. The code path you removed to fix
  other devices no longer handles these errant ones because, well,
  you removed it. These peripherals do exactly what they're not
  supposed to do, so they fall through to the case that should
  never happen. Your code raises an exception and panics the
  kernel.

  Of course, you spend some time being mad at stupid USB hub
  makers, since the entire reason to buy their product is that it
  allegedly meets a standard you've shown it to violate. What's
  the point in buying a "USB" device if it doesn't work with USB?
  But customers have these defective hunks of plastic, and telling
  them "Go buy better USB devices" isn't really an option.
  Instead, you buy some of these substandard hubs for the testing
  lab, hoping the extra testing doesn't require hiring more
  engineers. You go back and put in code to accommodate the
  defective devices, and you wonder in idle moments if you should
  have found the problem before it shipped even though it wasn't
  your fault.

  Keep that up for five years and you get a nice gold Apple-shaped
  pin with five small diamonds. You can sell it for about 0.05% of
  the cost of a house in Cupertino.

**The tiny update**

  This description is entirely imaginary but based on a true
  element: USB hubs that fail to enumerate after reset as
  _required_ by the USB specification panic the kernel [1] in Mac
  OS X 10.2.5. It's not the operating system's fault, but people
  with the hubs don't want to hear that, so Apple fixed it. That
  and another nasty bug are the primary reasons Apple released Mac
  OS X 10.2.6 on 2003.05.06, less than one month after the
  2003.04.10 release of Mac OS X 10.2.5 (MDJ_ 2003.04.14). The
  company followed that two days later with Mac OS X Server
  10.2.6, an update so similar it almost makes your head spin. Mac
  OS X 10.2.6 is available as a 6MB updater [2] from Mac OS X
  10.2.5 or an 86MB "Combo" updater [3] for any system since Mac
  OS X 10.2. Mac OS X Server 10.2.6 is also available as a 4MB
  updater [4] from Mac OS X Server 10.2.5 or as a 103MB "Combo"
  updater [5] for any version since Mac OS X Server 10.2.

  [1] <http://docs.info.apple.com/article.html?artnum=25440>
  [2] <http://www.info.apple.com/kbnum/n70173>
  [3] <http://www.info.apple.com/kbnum/n70174>
  [4] <http://www.info.apple.com/kbnum/n120216>
  [5] <http://www.info.apple.com/kbnum/n120217>

  These updates are small, compared to 23.3MB for November's Mac
  OS X 10.2.2, 50.9MB for December's Mac OS X 10.2.3, 40.6MB for
  February's Mac OS X 10.2.4, and 38.2MB for the recent Mac OS X
  10.2.5. The Mac OS X 10.2.6 user-level release notes [6]
  describe only six changes, almost but not quite everything in
  the update. The Mac OS X Server 10.2.6 release notes [7] offer a
  few more changes that also apply to the non-server version, plus
  some mysteries.

  [6] <http://docs.info.apple.com/article.html?artnum=107531>
  [7] <http://docs.info.apple.com/article.html?artnum=107531>

  Here's what we found, and as is now usual, you can grab our
  hierarchical OPML outline of all the files in either the Mac OS
  X 10.2.6 [8] or Mac OS X Server 10.2.6 [9] updates if you wish.

  [8] <http://www.macjournals.com/special/1026.opml>
  [9] <http://www.macjournals.com/special/1026server.opml>

**Server shenanigans** -- Comparing those two OPML files shows
  something remarkably annoying: the Mac OS X 10.2.6 and Mac OS X
  Server 10.2.6 updates are _identical_ save for one file. The
  server installer updates
  "/System/Library/CoreServices/ServerVersion.plist", the file
  that identifies the installed operating system as Mac OS X
  Server instead of plain Mac OS X Server. Every other file that
  Mac OS X Server 10.2.6 installs is also installed by the Mac OS
  X 10.2.6 updater. The archive sizes are different due to
  compression differences, different installer scripts (the server
  updater makes sure you're running it on Mac OS X Server),
  different "Read Me" files, and so on. The actual operating
  system files installed by the two updaters are the same.

  This is annoying largely because Apple seems unable to make up
  its corporate mind on what a "Mac OS X Server" update really is.
  The Mac OS X Server 10.2.5 update, for example, contained only
  files specific to Mac OS X Server, and could only be installed
  on a machine that already had the Mac OS X 10.2.5 Update applied
  to it, even though many server-related changes were implemented
  in the core operating system (MDJ_ 2003.04.14). For 10.2.6,
  however, server administrators are wasting time applying the
  non-server update first, as the only different file they'll get
  is the ServerVersion.plist file. Updates are already mysterious
  enough: Apple needs to decide if server updates require
  similarly numbered non-server updates or not, and stick with
  that decision.

  It's also bad for regular users because, as usual, Apple
  documents some changes that apply to every system only in the
  server version's release notes, usually a few days after the
  fact. We've integrated those changes into this description, but
  honestly, average users shouldn't have to read server release
  notes to figure out what changed on their iMacs or iBooks.

**Applications** -- The only changed application of any kind in
  Mac OS X 10.2.6 is, once again, Address Book, but the program
  itself is not changed: only the ABDeviceCommandSets.plist file
  inside Address Book's telephony bundle, found within the
  application's package. That XML file contains modem strings to
  control functions on several Bluetooth phones, the mechanism
  behind Address Book's ability to dial your Bluetooth phone on
  request. The revised file supports the Sony Ericsson T610 phone.
  The only other revised files inside Address Book are XML files
  to bump the version of both the Telephony bundle and the Address
  Book application itself, but again, no code changed. As part of
  this, the Address Book public framework (in
  "/System/Library/Frameworks") also has revised version numbers,
  but again, no code-level changes. None.

**System** -- All of Mac OS X 10.2.6's system-level changes are
  implemented either in a revised Mach kernel file
  ("/mach_kernel") or in six subdirectories of the
  "/System/Library" hierarchy: there are no changes to Unix
  components like CUPS or Samba, no revised utilities, no updated
  preference panes, or anything like that. In fact, one of those
  subfolders has only a trivial change: the only revised file in
  "/System/Library/CoreServices" is SystemVersion.plist. It
  defines this release as Mac OS X 10.2.6, build 6L60. (The same
  folder has the revised ServerVersion.plist file in the Mac OS X
  Server 10.2.6 update.)

  Other folders also have only lightweight changes. The only
  changed file in "/System/Library/Printers" is
  libLW8Converter.dylib, one we believe the system uses to convert
  LaserWriter 8-specific printing routines used by some Carbon
  applications (mostly from Adobe) to something that Mac OS X's
  CUPS-based architecture can use. Mac OS X 10.2.6 changes that
  conversion so that such programs can now use PostScript-based
  OpenType fonts through these LaserWriter 8-specific routines.
  The System Configuration framework that Mac OS X Server uses
  gets a revised bundle for controlling USB-based uninterruptible
  power supplies, and that change (both code and version data) is
  the only revision in the "/System/Library/SystemConfiguration"
  folder.

  Three more public frameworks in "/System/Library/Frameworks"
  change in addition to the previously-mentioned Address Book
  version changes. The massive umbrella Application Services
  framework is revised in Mac OS X 10.2.6, but only to change the
  QuickDraw framework that implements the Mac OS X-native version
  of the Mac's classic graphics package. The documented change to
  improve Maya Auto Paint and Paint Effects is either in QuickDraw
  or OpenGL, but we know of no other change that can be tied to
  QuickDraw.

  The equally-massive Core Services public framework also sees
  updates in Mac OS X 10.2.6, but only to two of its component
  frameworks: the Carbon Core and OS Services frameworks. Apple
  packages many of its most common frameworks in larger umbrella
  frameworks for the convenience of programmers: a program that
  binds to Core Services automatically gets the Carbon Core and OS
  Services, along with Core Foundation Networking and Web
  Services.

  We think these frameworks may fix a Mac OS X 10.2.5 problem that
  disabled non-Roman scripts (writing systems and text input) if
  English was your default language. As it turned out, if you
  wanted to use a script like Traditional Chinese or Korean, you
  had to have that script's default font installed. If you didn't,
  the script wouldn't even show up in the International preference
  pane or the Input menu (the one with a flag). On the other hand,
  Apple now documents [10] those problems as caused by bad
  international-related cache files that weren't properly cleared
  during the Mac OS X 10.2.5 software update. The Mac OS X 10.2.6
  installer takes care of them, or you can do it yourself by
  following Apple's instructions.

  [10] <http://docs.info.apple.com/article.html?artnum=107534>

  That's all the user-level release notes say might be related,
  but the server notes go a bit farther, saying the "Core OS" was
  updated to "address a rare condition in which users cannot
  complete the login process." No further details are available.

  Mac OS X 10.2.6's last updated framework is for OpenGL. That's
  almost certainly behind the documented improvement for Unreal
  Tournament 2003 on GeForce2 MX and GeForce4 MX graphics chips,
  and possibly behind Maya improvements as well. Apple only
  recently documented [11] OpenGL changes in Mac OS X 10.2.5,
  including support for the ATI Radeon 9700 Pro graphics
  processor, some new OpenGL extensions, and lots of bug fixes.
  The framework has revised OpenGL dynamic libraries, and revised
  plug-in bundles for 16-bit and 24-bit color (technically, five
  bits per channel and eight bits per channel) for 16-bit and
  32-bit data. The "/System/Library/Extensions" folder, home of
  kernel extensions, has revised GeForce2 MX and GeForce3 OpenGL
  drivers as part of these fixes.

  [11] <http://developer.apple.com/qa/qa2001/qa1258.html>

**The server changes** -- That folder also contains a revised
  USB driver family kernel extension as part of the misbehaving
  USB hub fix, documented also in the server release notes as
  "improve[d] compatibility with some USB devices, providing
  greater system reliability." Although that's the last documented
  change, there's one more revised kernel extension: the driver
  for PC Cards. Removing and reinserting the Novatel Wireless CDMA
  1.9 GHz PC Card under Mac OS X 10.2.5 didn't work properly; Mac
  OS X 10.2.6 fixes that, and MacFixIt readers report other PC
  Card improvements as well.

  On the other hand, there's a change in the update that's not
  documented at all in the user-level release notes: a revised
  AppleShare file system. More specifically, Mac OS X 10.2.6
  contains a new version of the AppleShare file system kernel
  extension, found in the "/System/Library/Filesystems/AppleShare"
  folder. There have been third-party reports of corruption in
  copying files to AppleShare volumes under Mac OS X 10.2.5, as
  well as continued reports that some versions of AppleShare don't
  get along with Microsoft's UAM (user authentication method) for
  its Windows servers. Only in the Mac OS X Server 10.2.6 release
  notes does Apple own up to both bugs and admit that the update
  fixes them.

  The server notes also say that the update provides "more
  reliable software installation for a client computer that is
  installing from a file located on a remote AFP server volume,"
  but that probably applies to the AppleShare file system (the
  client software that mounts servers), not the AppleShare file
  server (the program that shares your disks either as File
  Sharing or as the AFP File Service on Mac OS X Server). It's
  just that not many people install software over file sharing
  compared to installing from dedicated file servers, so it's not
  much of an issue for most users.

  The server release notes also say the update provides an
  "improved CUPS printing experience for Hewlett Packard LaserJet
  8150 CN laser printer[s], allowing full text and image print by
  default." It's hard to see how that can be true: no part of CUPS
  is revised in 10.2.6, and the only printing change is to the
  LaserWriter 8 compatibility library, as mentioned. Perhaps it's
  a change that only affects applications using the older
  LaserWriter 8 routines, but elsewhere Apple strongly implies the
  LaserWriter 8 change is all about OpenType fonts for such
  programs. CUPS was revised in 10.2.5, though, and perhaps this
  change is only now getting documented, though it would be pretty
  underhanded of Apple to imply an older change is actually a
  newer one.

**That's all**

  That's the sum total of Mac OS X 10.2.6, and honestly, that's
  enough for a bug fix update. Monthly updates of more than 30MB
  are not only amazingly difficult on the 70% of US Internet users
  that connect via dial-up, they change so much of the system with
  so little documentation that careful professionals can hardly
  know what they're getting into with each revision.

  Alas, this hasn't stopped a lot of fascinating but ultimately
  silly speculation. For example, MacFixIt readers have said [12]
  that the update makes Adobe FrameMaker display white space
  instead of gray space at the beginning and end of documents.
  Although it's possible this has something to do with the changes
  in Mac OS X's QuickDraw framework, FrameMaker is a Classic
  application, and Classic did not change.

  [12] <http://www.macfixit.com/article.php?story=20030509064645537&pid=0&type=article>

  The site also had extensive descriptions [13] of "changes"
  between Address Book 3.0.3 (v224) in Mac OS X 10.2.5 and Address
  Book 3.0.3 (v225) in Mac OS X 10.2.6, including new editing
  features. But remember: there are no code-level changes in
  Address Book or its framework. None. The only changes are to
  modem strings and files that determine the version numbers. It's
  quite impossible for the update to provide the kinds of changes
  MacFixIt describes.

  [13] <http://www.macfixit.com/article.php?story=20030508084651448>

  There's also some confusion circulating about whether there's a
  difference between applying the smaller updater or the
  mega-combo updaters that work on any Jaguar system. MWJ_ and
  MacFixIt reader Philip Tejada applied each kind of updater to an
  "identical copy of [Mac OS X] 10.2.4" on two partitions, and
  when done, found "over 2800 files different on the two systems
  after the updates." MacFixIt used this as support for its recent
  and oft-repeated contention that you're better off using the
  Combo updaters than the smaller updaters designed for the
  just-prior version of the system software.

  However, Tejada copied MDJ_ on his findings, and MacFixIt
  extensively misquoted [14] him. First, Tejada was trying to
  track down the crashing bug in Microsoft's UAM that Apple's
  server notes now admit was introduced in Mac OS X 10.2.5. For
  Tejada, the problem only appeared on systems that used the
  MacFixIt-recommended "Combo" updater, _not_ the individual
  10.2.4 to 10.2.5 updater. Secondly, Tejada's experiment was in
  updating 10.2.4 to 10.2.5, not 10.2.5 to 10.2.6.

  [14] <http://www.macfixit.com/article.php?story=20030507074551748>

  MacFixIt actually changed his text to say he worked on "two
  identical partitions with an identical copy of 10.2.5 on each,"
  when Tejada's message says he worked on "two identical
  partitions with an identical copy of 10.2.4 on each," applying
  the _10.2.5_ updaters. Not only that, MacFixIt changed Tejada's
  words to support the site's "Combo" update recommendations when
  Tejada's experiment clearly showed the "Combo" updater caused
  his problem. Given this pretty much inexcusable misquoting of
  Tejada to support its preconceptions, it's now difficult to take
  anything MacFixIt says about "Combo" updates seriously.

  That doesn't answer Tejada's question, and to be honest, we're
  not sure of the total answer. Particularly, we're following up
  on how his original 10.2.4 system was created: the differences
  file shows that it did not have some standard parts of Mac OS X
  10.2 installed on it before the updaters were applied, such as
  iChat. The Combo updater contains an updated version of iChat
  from Mac OS X 10.2.2 (MDJ_ 2002.11.26), but the Mac OS X 10.2.5
  updater does not. That explains why the Combo updater writes a
  new copy of iChat but the single-version updater does not. We do
  not know if iChat is the only standard part of 10.2 that was not
  installed on Tejada's systems, nor how that would affect the
  installer.

  We're not completely surprised by different system binaries
  after two different updates because one of Mac OS X's primary
  performance mechanisms, prebinding [15], changes binary code
  files. Many of the files marked different in Tejada's experiment
  are prebound binaries, so the Installer's "Optimizing System
  Performance" pass could easily have changed those binaries. In
  Mac OS X 10.2, the system may also update a file's prebinding on
  its own for performance reasons. Since the results of prebinding
  depend on the libraries any binary code needs, which may depend
  on other binaries or applications, and so on, the results are
  somewhat unpredictable on an individual basis. If Tejada's
  system was missing some libraries from 10.2 that the Combo
  installer added but the individual 10.2.5 update did not, it
  could have affected many prebound code files.

  [15] <http://developer.apple.com/tools/projectbuilder/Prebinding.html>

  Tejada's systems had many non-binary differences as well. Some
  well-meaning MacFixIt readers said these were probably
  permission problems, but Tejada used the diff command, and diff
  only compares file contents. It does not flag two files as
  different if their contents are the same but permissions differ.
  The differences Tejada found could depend on the composition of
  the reference 10.2.4 systems he updated; we're attempting to
  follow up on that.

  At the moment, we've seen nothing that makes us reluctant to
  recommend either updater, with the choice depending on your
  needs. If you're running Mac OS X 10.2.5, use the 6MB updater.
  If you're running an older version, use the 103MB "Combo"
  updater. Mac OS X 10.2.6 fixes a small set of annoying bugs in
  Mac OS X 10.2.5, and does so quickly and without confusing us
  too much. We approve.


-----------------------------------------------------------------

  MDJ_, The Daily Journal for Serious Macintosh[tm] Users, is
  published by GCSF, Incorporated.

  Publisher:           Matt Deatherage    <mattd@macjournals.com>
  Staff:               Justin Seal       <justin@macjournals.com>
                       Nathaniel Irons    <irons@macjournals.com>
                       John C. Welch     <jwelch@macjournals.com>
                       Jerry Kindall    <kindall@macjournals.com>
                       John Gruber       <gruber@macjournals.com>

  Copyright (c) 2003 GCSF, Incorporated.  All rights reserved.
  All trademarks are the property of their respective holders
  and owners.

  The symbol **[D]** indicates potential conflicts of interest,
  but for readability, the necessary disclaimer has been omitted
  from the text. Such disclaimers may be found online at
  <http://www.macjournals.com/disclaimers.html>.

  MDJ_ contains news, information, strong opinion, parody, biting
  sarcasm, and things you need to know.  Those easily offended
  should seek information elsewhere.

  Humans often answer the telephone between 10 AM and 6 PM Central
  (US) Time, Monday through Friday. Voicemail is available at any
  hour.

  This file is formatted as setext.  For more information, send
  email to <setext@tidbits.com>.  A file will be returned shortly.
  It is also digitally signed using PGP technology to verify the
  integrity of the transmission.
  
  Our DH/DSS corporate PGP key maybe obtained at
  <http://www.macjournals.com/pages/gcsf/gcsf_keys.html#Anchor-GCSF_DSSKey>.
  
  GCSF, Incorporated.
  P.O. Box 1021
  El Reno, OK  73036-1021
  (405) 262-1399
  <mdj@macjournals.com>


-----BEGIN PGP SIGNATURE-----
Version: PGP 8.0.2

iQA/AwUBPsBgprH4QMSEyVHCEQKpGwCgpauRvUGQZe/H8Z/EKmBCJ+3fQnoAoPsb
LkF1eMKA6uILF4Xm1OYf4QGe
=xeyr
-----END PGP SIGNATURE-----