Summary: | REGRESSION(r19992-20004): WebKit ToT fails on G3 (dyld: Symbol not found: _WebCoreUnicodeDirectionFunction) | ||
---|---|---|---|
Product: | WebKit | Reporter: | Tony Mantler <nicoya> |
Component: | Tools / Tests | Assignee: | Nobody <webkit-unassigned> |
Status: | RESOLVED FIXED | ||
Severity: | Major | CC: | ap, ddkilzer, dev+webkit, mrowe |
Priority: | P1 | Keywords: | Regression |
Version: | 523.x (Safari 3) | ||
Hardware: | Mac (PowerPC) | ||
OS: | OS X 10.4 |
Description
Tony Mantler
2007-03-07 20:57:54 PST
Hi Tony! It's been a while since I've done any hacking on linux-mac68k. I believe this issue is a duplicate of Bug 10510, which means that you need to upgrade from Acrobat Reader 7 to Acrobat Reader 8 (or disable the Acrobat Reader plug-in) to use the nightly builds. Note that the first entry in Safari.crash.log is from 2007-02-28 and is probably a different issue. The rest of the crash log entries are related to this issue. Oops, didn't even notice I copied more than one crash log. Yes, disregard the first one in there, I think it's not even from a webkit nightly. I don't believe this issue is an exact duplicate of the bug you mentioned, as I do not have Acrobat or the Acrobat Reader plugin installed, since the built-in PDF functions in OS X have always been more than sufficient for my needs. Is there a specific directory I should double-check to make sure there isn't one accidentally installed? And yeah, it's been a while since I've been doing any linux-mac68k stuff myself. Small world we should run into each other again. :) (In reply to comment #2) > Is there a specific directory I should double-check to make sure there isn't one > accidentally installed? I would check these directories: /Library/Internet Plug-Ins/ ~/Library/Internet Plug-Ins/ You could also check which plug-ins are installed by going to "Help", then "Installed Plug-ins". It could be a plug-in other than Acrobat Reader doing this, although it would have to be something that loaded on your home page. If you set your default home page to be "Empty" via Safari's preferences, does the nightly still crash? The default homepage is already empty. The crash happens well before any browser window opens. The plugins are: QuickTime Plug-In 7.1.3 Java Plug-in (CFM) Shockwave Flash 9.0 r19 Quartz Composer Plug-In Verified Download Plugin VLC Multimedia Plugin Shockwave Flash 9.0 r20 Java Plug-in for Cocoa Shockwave for Director QuickTime Plug-in 7.1.5 Java Plug-in I'll try pulling some of those out a bit later to see if one of them is to blame. Well, I removed all the plugins and WebKit still crashes on launch. Any other ideas? (In reply to comment #5) > Well, I removed all the plugins and WebKit still crashes on launch. Any other > ideas? I'm out of ideas. Mark, do you have any more things to try? Tony, is this a relatively "clean" installation of Mac OS X 10.4.8 and/or a "clean" account? That depends what you mean by "clean". :) I just tried with a newly created user account and the crash was the same. The OS hasn't really been mucked around with much, but it's the same install since I got the machine way back in the 10.3 days. Just a shot in the dark, do you have a G3 machine to test on? Perhaps one of the build targets is broken. (In reply to comment #7) > The OS hasn't really been mucked around with much, but it's the same install > since I got the machine way back in the 10.3 days. I have a PB G4 that has been upgraded to Tiger (10.4) from Panther (10.3), although it works fine. > Just a shot in the dark, do you have a G3 machine to test on? Perhaps one of > the build targets is broken. Good thought! Unfortunately I do not have a G3 to test. Perhaps one of the other WebKit contributors (or Apple engineers) does. This does indeed crash with the same "Symbol not found" error on my PowerMac G3 with a fairly untouched, clean install of 10.4.8. Giving this bug the P1 treatment. I wonder if this regressed when Leopard support was added to WebKit ToT? Please try a WebKit nightly before r18849 to see if that "fixes" the issue. http://trac.webkit.org/projects/webkit/changeset/18849 Otherwise, a binary search of WebKit nightlies is probably the fastest way to find the regression point. http://nightly.webkit.org/ Per the original report, I've verified that the r19992 nightly works fine. A quick Trac check for r19992-r20004 doesn't show anything obvious. (In reply to comment #11) > Per the original report, I've verified that the r19992 nightly works fine. > A quick Trac check for r19992-r20004 doesn't show anything obvious. Could be an issue with the system that the nightlies are being built on, or perhaps a toolchain issue (ugh!). (Tony is "familiar" with fighting toolchain issues from the linux-mac68k project. :) Matt or Tony, do either of you have local development environments? It would be interesting (if not time-consuming) to create a local (debug) build of WebKit on a G3 to see if it has the same issue. Sadly my G3 happens to be a Rev-A B&W, which means I'm stuck with the original 6 gig root drive and no room to install xcode. And yes, I'm quite (painfully) familiar with toolchain issues. :) (In reply to comment #11) > Per the original report, I've verified that the r19992 nightly works fine. A > quick Trac check for r19992-r20004 doesn't show anything obvious. I'm leery of r20003 which was a change to libWebKitSystemInterface.a, and was later fixed by Mark in r20005. http://trac.webkit.org/projects/webkit/changeset/20003 http://trac.webkit.org/projects/webkit/changeset/20005 I'm building on my G3 now, I'll let you know how it goes when it's done in a week :p ToT built and runs fine on my G3 (ironically, I think it built faster than my G4...). I also built the nightly launcher in WebKitTools/ and that runs fine too, so it must be a change in the way the nightlies are built. I get the feeling this may be because I've switched to building the nightlies from Leopard. One thing that strikes me as suspicious is the reference to /System/Library/Frameworks/WebKit.framework. Symbol not found: _WebCoreUnicodeDirectionFunction Referenced from: /System/Library/Frameworks/WebKit.framework/Versions/A/WebKit Expected in: /Applications/WebKit.app/Contents/Resources/WebCore.framework/Versions/A/WebCore This suggests that the right WebKit.framework is not being loaded. Ok, I see why this is. Looking into a fix. This looks to be a toolchain issue. I've hacked a workaround into my nightly build process. Nightlies starting from r20103 should launch correctly on G3s again. r20103 works on my G3 now. Thanks! |