148897 - LDAP auth dialog box has bizarre text (egcs 1.1.2 codegen bug)

Status: VERIFIED FIXED

(MailNews Core :: LDAP Integration, defect, P2)

Description

[Dan Mosedale (:dmosedale, :dmose)]

Configure an LDAP server to have a Bind DN currently only the 5/3/2002 and newer
trunk builds have this option).  In a nightly build, one sees:

"Please enter your password for authPromptText."

In a debug build, one sees the expected thing, where authPromptText is instead
the hostname of the LDAP server.

I'm fairly sure (though not yet 100%) this is not a packaging problem, because
if there were packing problems here, other strings would be busted too, I think.

Comment 1

[Michael Buckland]

Discussed in mail news bug meeting.  Decided to plus this bug.

Comment 2

[Dan Mosedale (:dmosedale, :dmose)]

OK, problem still exists in a debug tree built with egcs 1.1.2, so this is
neither a packaging nor an optimizer problem.  Looks like codegen.  Time to bust
out the debugger and disassemble...

Comment 3

[Dan Mosedale (:dmosedale, :dmose)]

Attachment: patch, v1

Fix incorrect array length calculation & add temporary variable to work around
egcs bug.

Comment 4

[Brian Ryner (not reading)]

Comment on attachment 88385 [details] [diff] [review]
patch, v1

r=bryner

Comment 5

[David :Bienvenu]

Comment on attachment 88385 [details] [diff] [review]
patch, v1

sr=bienvenu

Comment 6

[Dan Mosedale (:dmosedale, :dmose)]

This exact same patch (literally, the same patch file) also must be applied to
xpfe/components/autocomplete/src/nsLDAPAutoCompleteSession.cpp.  |patch -l|
should be used.

Comment 7

[Dan Mosedale (:dmosedale, :dmose)]

Fix checked into both of these files on the trunk.  Still needs to land on the
branch.

Comment 8

[jag (Peter Annema)]

Are you sure egcs is the problem here?

-        const PRUnichar *hostArray[1] = { NS_ConvertASCIItoUCS2(host).get() };

That line of code is so bogus. The temporary object there and therefore its
internal buffer may have gone away on the line following it, leaving a dangling
pointer in hostArray. The solution you've used is the correct one. That it
happened to work with other compilers is pure luck.

Comment 9

[Judson Valeski]

please checkin to the 1.0.1 branch. once there, remove the "mozilla1.0.1+"
keyword and add the "fixed1.0.1" keyword.

Comment 10

[Dan Mosedale (:dmosedale, :dmose)]

jag: the braces used to delimit an array-initializer also count as block scope?
 I didn't think that was the case, and I don't have my copy of Stroustrup handy
to check.  Ah well, either way the bug is fixed.

Comment 11

[Paul Wyskoczka]

dding adt1.0.1+ on behalf of the adt for checkin to the 1.0 branch.  Please get
drivers approval before checking in. When you check this into the branch, please
change the mozilla1.0.1+ keyword to fixed1.0.1

Comment 12

[jag (Peter Annema)]

It's nothing to do with block scopes. Take for example

  const char* foopy = nsCString("foopy").get();

  printf("%s\n", foopy); // foopy at this point may be a dangling pointer.

See http://lxr.mozilla.org/seamonkey/source/string/doc/string-guide.html

Comment 13

[Dan Mosedale (:dmosedale, :dmose)]

jag: Oh, right!  I see now; thanks.  I'll open another bug to fix the bogus
commentary that I've just added.  :-)

Committed to the branch.

Comment 14

[yulian chang]

Verified with 20020626 branch build on LINUX platform.