The Trunk: SMLoader-ul.83.mcz

Levente Uzonyi uploaded a new version of SMLoader to project The Trunk:
http://source.squeak.org/trunk/SMLoader-ul.83.mcz

==================== Summary ====================

Name: SMLoader-ul.83
Author: ul
Time: 14 August 2015, 7:02:58.782 pm
UUID: 4898c13c-15f3-4ed8-867a-df8d79905124
Ancestors: SMLoader-mt.82

#includesSubString: -> #includesSubstring:

=============== Diff against SMLoader-mt.82 ===============

Item was changed:
  ----- Method: SMLoader class>>initialize (in category 'class initialization') -----
  initialize
  "Hook us up in the world menu."
  "self initialize"
  Smalltalk
  at: #ToolBuilder
  ifAbsent: [self registerInFlapsRegistry.
  (Preferences windowColorFor: #SMLoader) = Color white
  ifTrue: ["not set"
  Preferences
  setWindowColorFor: #SMLoader
  to: (Color colorFrom: self windowColorSpecification brightColor)].
  (TheWorldMenu respondsTo: #registerOpenCommand:)
  ifTrue: [| oldCmds |
+ oldCmds := TheWorldMenu registry select: [:cmd | cmd first includesSubstring: 'Package Loader'].
- oldCmds := TheWorldMenu registry select: [:cmd | cmd first includesSubString: 'Package Loader'].
  oldCmds do: [:cmd | TheWorldMenu unregisterOpenCommand: cmd first].
  TheWorldMenu registerOpenCommand: {self openMenuString. {self. #open}}]].
  DefaultFilters := OrderedCollection new.
  DefaultCategoriesToFilterIds := OrderedCollection new!

Item was changed:
  ----- Method: SMLoaderPlus class>>initialize (in category 'class initialization') -----
  initialize
  "Hook us up in the world menu."
  "self initialize"
  Smalltalk
  at: #ToolBuilder
  ifPresent:
  [ : tb | self registerInFlapsRegistry.
  (Preferences windowColorFor: #SMLoader) = Color white "note set" ifTrue:
  [
  #(#SMLoader #SMReleaseBrowser ) do:
  [ : each |
  Preferences
  setWindowColorFor: each
  to: (Color colorFrom: self windowColorSpecification brightColor) ] ].
  (TheWorldMenu respondsTo: #registerOpenCommand:) ifTrue:
  [ | oldCmds |
  oldCmds := TheWorldMenu registry select:
+ [ : cmd | cmd first includesSubstring: 'Package Loader' ].
- [ : cmd | cmd first includesSubString: 'Package Loader' ].
  oldCmds do:
  [ : cmd | TheWorldMenu unregisterOpenCommand: cmd first ].
  TheWorldMenu registerOpenCommand: {self openMenuString.
  {self.
  #open}} ] ].
  DefaultFilters := OrderedCollection new.
  DefaultCategoriesToFilterIds := OrderedCollection new!

On 14.08.2015, at 20:33, [hidden email] wrote:

Hi. The Trunk says:

http://build.squeak.org/job/Trunk/default/10/

2015-08-14T22:46:37.575+01:00: Installing Tools-ul.630
2015-08-14T22:46:37.645+01:00: Installing Protocols-ul.49
2015-08-14T22:46:37.71+01:00: Installing SMLoader-ul.83
vvvvvvvvvvvvvvvvvv MessageNotUnderstood: ByteString>>includesSubstring: vvvvvvvvvvvvvvvvvv
The time is 2015-08-14T22:46:37.743+01:00
ByteString(Object)>>doesNotUnderstand: #includesSubstring:
[] in [] in SMLoaderPlus class>>initialize
OrderedCollection>>select:


That's Strange…


Best regards
        -Tobias

Hi Tobias,

Thanks for pointing it out. I thought Bert made some changes which ensure
that such dependencies don't require a separate update map. I tested the
update process on my machine, and it worked for me.
Anyway, I've modified the last update map to load Collections-ul.641
before other packages related to this change.
Btw, am I right that Jenkins is still using a non-Spur image to run the
tests?

Levente

On Sat, 15 Aug 2015, Tobias Pape wrote:

Hi Levente,

On 15.08.2015, at 11:16, Levente Uzonyi <[hidden email]> wrote:

> Hi Tobias,
>
> Thanks for pointing it out. I thought Bert made some changes which ensure that such dependencies don't require a separate update map. I tested the update process on my machine, and it worked for me.
> Anyway, I've modified the last update map to load Collections-ul.641 before other packages related to this change.
> Btw, am I right that Jenkins is still using a non-Spur image to run the tests?

The traditional Job is non-Spur:

        http://build.squeak.org/job/SqueakTrunk/

The 'old' Spur job

        ~~http://build.squeak.org/job/SqueakTrunkOnSpur/~~

is gone (it didn't work most of the time).

There's a new Trunk job that is Spur-based:

        http://build.squeak.org/job/Trunk/

But it still crashes the VM in some image segment code. See
related messages with Eliot.

Best regards
        -Tobias

Hi All,

Today I learned that ImageSegmentTest unloads and reloads the "non-active"
UIManagers from the image. Since ImageSegments are broken in Spur, the
reload will (may?) be partial. In my image this results in "half-loaded"
methods: the methods don't get reinstalled into the MethodDictionary, but
their selectors are loaded. It usually happens to MVCUIManager >>
#dialogStringFromQuery:withTitle:, and sometimes happens to
DummyUIManager's methods as well.

The result is a broken MethodDictionary, which can easily be the cause of
the VM crashes, and a broken Symbol table, because the selector's name is
a non-interned Symbol.

I think we should disable ImageSegmentTest >>
#testImageSegmentsShouldBeWritableToaFile until ImageSegments get fixed
in Spur, and/or modify the test to unload/reload transient classes
instead of ones which may be used in the future.

Levente

On Sat, 15 Aug 2015, Tobias Pape wrote:

Hi,

On 21.08.2015, at 01:53, Levente Uzonyi <[hidden email]> wrote:

> Hi All,
>
> Today I learned that ImageSegmentTest unloads and reloads the "non-active" UIManagers from the image. Since ImageSegments are broken in Spur, the reload will (may?) be partial. In my image this results in "half-loaded" methods: the methods don't get reinstalled into the MethodDictionary, but their selectors are loaded. It usually happens to MVCUIManager >> #dialogStringFromQuery:withTitle:, and sometimes happens to DummyUIManager's methods as well.
>
> The result is a broken MethodDictionary, which can easily be the cause of the VM crashes, and a broken Symbol table, because the selector's name is a non-interned Symbol.
>
> I think we should disable ImageSegmentTest >> #testImageSegmentsShouldBeWritableToaFile until ImageSegments get fixed in Spur, and/or modify the test to unload/reload transient classes instead of ones which may be used in the future.

Thank you for tracking this down!
I really wondered why the methods disappeared.

Best regards
        -Tobias

On 15.08.2015, at 11:16, Levente Uzonyi <[hidden email]> wrote:
>
> Hi Tobias,
>
> Thanks for pointing it out. I thought Bert made some changes which ensure that such dependencies don't require a separate update map. I tested the update process on my machine, and it worked for me.

What I did is that if a method/class is moved between packages, it will do that independent of the order of the packages.

What happened here I don’t quite understand, surely Collections would come before SMLoader in the update map?

- Bert -