70635 - Crash triggering host-only install URL

Status: RESOLVED WONTFIX

(Core Graveyard :: Installer: XPInstall Engine, defect)

Description

[Jimmy Lee]

Build: 2001-03-01-08-Mtrunk(LINUX)

1. From browser location field, enter:
javascript:InstallTrigger.install({"test":"ftp://ftp.netscape.com/"},function(ur
l,status){alert(url+' '+status);})
Related to Bug 51098

2. From confirm dialog, click OK button

RESULT:
Crash after displaing "true" from browser window.  Windows and Mac write "true" 
to browser window with no crash.

EXPECTED RESULT:
No crash.

TalkBack Incident ID 27137667

Stack Trace:
libxpinstall.so + 0x1fe5b (0x40339e5b) 
libnecko2.so + 0x51ea (0x412311ea) 
libnecko.so + 0x2fc54 (0x40852c54) 
libnecko2.so + 0x57cf (0x412317cf) 
libnecko.so + 0x1b302 (0x4083e302) 
libnecko.so + 0x1a6cc (0x4083d6cc) 
libxpcom.so + 0x75633 (0x400bd633) 
libxpcom.so + 0x75556 (0x400bd556) 
libxpcom.so + 0x76319 (0x400be319) 
libwidget_gtk.so + 0xcdcf (0x404fbdcf) 
libwidget_gtk.so + 0xcb8d (0x404fbb8d) 
libglib-1.2.so.0 + 0xe52a (0x4069152a) 
libglib-1.2.so.0 + 0xfbe6 (0x40692be6) 
libglib-1.2.so.0 + 0x101a1 (0x406931a1) 
libglib-1.2.so.0 + 0x10341 (0x40693341) 
libgtk-1.2.so.0 + 0x8c209 (0x405bd209) 
libwidget_gtk.so + 0xd2bc (0x404fc2bc) 
libnsappshell.so + 0xd57a (0x4036d57a) 
main1() 
main() 
libc.so.6 + 0x181eb (0x4023e1eb)

Comment 1

[Jimmy Lee]

Nominating for beta.  I need somebody to look at this more closely to determine 
what the real problem might be.

Comment 2

[dbragg]

Interesting note on this bug.  if you take off the trailing slash on Dan's test
(ftp://ftp.netscape.com instead of ftp://ftp.netscape.com/) Mozilla doesn't
crash. Instead you get an error -207 (Can't read archive) and the program
continues to run just fine.  You also get just ftp:// in the Confirm dialog
instead of ftp://ftp.netscape.com.

Comment 3

[dbragg]

After some preliminary investigation into this bug it's clear that this will not
be a quick fix.  It needs to be triaged to nsbeta1+ for me to continue working
on it.

Comment 4

[Daniel Veditz [:dveditz]]

A host plus non-existant file returns a -228 now, right? That's what I worry 
about because links go stale and users shouldn't suffer a hang or crash in that 
case. If the problem is now limited to a strange host-only-URL situation the 
site author can debug that before publishing the page.

Comment 5

[Jimmy Lee]

A host (ie jimbob) and nonexistent file returns -207.  But when I try this with 
an ftp server instead of http, then nothing gets written to the install.log.

Comment 6

[Daniel Veditz [:dveditz]]

Linux only->pp

Updating summary to reflect my understanding that it's not simply a specified
.xpi going missing on the host (which could be relatively common) but rather a
URL that specifies no host file whatsoever. Assuming that's true then we should
minus this one.

Comment 7

[Daniel Veditz [:dveditz]]

Don has moved to another group. Install bugs -> syd

Comment 8

[Jimmy Lee]

Nominating.

Comment 9

[Syd Logan]

over to dprice

Comment 10

[chris hofmann]

dprice is on sabitcal. moving to next milestone. 

Comment 11

[Daniel Veditz [:dveditz]]

obscure case, nsbeta1-

Comment 12

[Syd Logan]

Resetting milestone of all nsbeta1-bugs, only nsbeta1+ bugs should have a target
milestone.

Comment 13

[Syd Logan]

Resetting milestone, only nsbeta1+ bugs can have a milestone on them, these are
niminated, but not yet plussed.

Comment 14

[Brant Gurganus]

By the definitions on <http://bugzilla.mozilla.org/bug_status.html#severity> and
<http://bugzilla.mozilla.org/enter_bug.cgi?format=guided>, crashing and dataloss
bugs are of critical or possibly higher severity.  Only changing open bugs to
minimize unnecessary spam.  Keywords to trigger this would be crash, topcrash,
topcrash+, zt4newcrash, dataloss.

Comment 15

[David Epstein]

bug needs to be reassigned. Don't have a Linux platform to test this out now to
see if it's still crashing.

Comment 16

[Matthew Cline]

Both Firefox 2.0.0.11 and the latest Minefield display "false" without crashing; don't know if this means the crash is fixed or if the test needs to be updated.

Comment 17

[Daniel Veditz [:dveditz]]

"false" means you tried the test on a non-whitelisted site (this bug pre-dates the whitelist). If you append "void(0);" to the end of the testcase it will prevent the page navigation which makes the error clearer, but if you're not getting the confirmation dialog then you didn't get to step 2.

You're trying this on Linux? It never crashed on windows.

Comment 18

[Daniel Veditz [:dveditz]]

The xpinstall script engine has been removed from the trunk, bugs in it are obsolete.