163423 - NativeDeleteKey needs to deal with non-existent keys gracefully

Status: VERIFIED FIXED

(Core Graveyard :: Installer: XPInstall Engine, defect, P1)

Description

[Syd Logan]

NativeDeleteKey can be called after the key has been deleted, which will lead to 
an error. For example, installer script schedules same key twice for deletion. 
The first will succeed during performInstall, the second will fail. It should be 
more tolerant of this (but debug runtime should flag this error with an assert).

Patch on its way.

Comment 1

[Syd Logan]

Attachment: patch to fix this

Comment 2

[Syd Logan]

Attachment: Patch with debug asserts added

Forgot to add a debug assert, plus some nitpicking about how to code
conditional statements.

Comment 3

[Daniel Veditz [:dveditz]]

Comment on attachment 95839 [details] [diff] [review]
Patch with debug asserts added

sr=dveditz  (explicitly testing against PR_TRUE doesn't match the style of the
file, though)

Comment 4

[Joe Hewitt (gone)]

Comment on attachment 95839 [details] [diff] [review]
Patch with debug asserts added

r=hewitt

Comment 5

[Syd Logan]

fix landed in trunk, should be in the trunk 8/19/2002 builds.

Comment 6

[Jaime Rodriguez, Jr.]

Marking as Resolved/Fixed per comment Comment #5 From [e-mail].

paw: who can verify this fix on the trunk for us?

Comment 7

[Paul Wyskoczka]

Jimmy, who is the QA contact

Comment 8

[scottputterman]

adding adt1.0.1+

Comment 9

[Jimmy Lee]

Build: 2002-08-19-08-trunk(WIN)

This is not working for me.  Marking reopen.  The install.log returns:
-------------------------------------------------------------------------------
http://jimbob/jars/a_winreg_deletekey.xpi  --  08/19/2002 15:38:13
-------------------------------------------------------------------------------

     Acceptance: a_winreg_deletekey (version 1.2.3.4)
     ------------------------------

     ** ERROR (-242): Delete Registry Key: HKEY_LOCAL_MACHINE\Software\WinReg
Test Key\subkey1\subkey2 []

     Install cancelled by script  --  08/19/2002 15:38:13

I am expecting something like "Trying to delete a key that doesn't exist
anymore. Possible causes include calling deleteKey for the same key multiple times."

The test I am running is http://jimbob/jars/a_winreg_deletekey.xpi

Install script:
initInstall("Acceptance: a_winreg_deletekey", "a_winreg_deletekey", "1.2.3.4", 0); 

var winreg = getWinRegistry();

winreg.setRootKey(winreg.HKEY_LOCAL_MACHINE);
winreg.deleteKey("Software\\WinReg Test Key\\subkey1\\subkey2");

if (0 == getLastError())
	performInstall();
else
	cancelInstall();

Comment 10

[Jimmy Lee]

I should add that -242 does sound correct though.

KEY_DOES_NOT_EXIST          = -242

Comment 11

[Judson Valeski]

 please checkin to the MOZILLA_1_0_BRANCH branch. once there, remove the
"mozilla1.0.1+" keyword and add the "fixed1.0.1" keyword. 

Comment 12

[Syd Logan]

that is a different error, we are seeing an error '2' in the native installs.
The -242 you are seeing is probably valid and is due to a check in the prep
stage. The 2 error is something totally different. Please file another bug if
you think this is wrong behavior to see -242 (seems like right behavior to me),
we are trying to resolve a specific bug in the native stub installer.

Comment 13

[Jimmy Lee]

-242 is correct to me.  However, I can use some guidance to identify where this
Error "2" appears from the native installer.  Thanks.

Comment 14

[Syd Logan]

Well, write a script that tries to delete the same key twice. I'm not sure how
far up the chain the -2 will be propogated in your test scaffolding, but in the
stub installer, it will generate a '2' error. Perhaps modifying an install.js
executed in one of the stub or blob install xpis will do the trick.

Comment 15

[Jaime Rodriguez, Jr.]

jimmylee: can you pls verify this as fixed on the trunk and 1.0 branch, then
replace the "fixed1.0.1" keyword with "verified1.0.1"?

Comment 16

[Jimmy Lee]

Build: 2002-08-21-09-trunk(WIN), 2002-08-21-08-1.0(WIN)

Looks good on both TRUNK and BRANCH.  Thanks to Gary for providing the machine
that reliably duplicates the problem.

Marking Verified.  Adding keyword "verified1.0.1".