Re: Mysterious crash with NSTableView

[Kyle Sluder]

> On Aug 29, 2016, at 6:39 AM, Andreas Falkenhahn  
> wrote:

> 
>> On 29.08.2016 at 02:10 Kyle Sluder wrote:
>> 
>> Delegates are different because they are often messaged in response to
>> various exogenous events. Some of these events might happen transiently
>> during window teardown, which is usually a time of massive fluctuation
>> in an app’s object graph. This is why -windowDidClose: is a good time to
>> nil out delegate properties that point back and the window controller
>> which ultimately owns the control.
> 
> So does this mean that this is potentially dangerous?
> 
>[NSApp runModalForWindow:win];
>[tableView setDelegate:nil];
>[tableView setDataSource:nil];
>[win release];
> 
> Should I move the "set to nil" calls to -windowDidClose instead?

Since you hold a strong reference to the window until after you clear out the 
delegate/dataSource backpointers, you know that the window cannot be 
reallocated until at least that point (but could be delayed to any arbitrary 
point in the future). So this approach is safe from window teardown causing the 
table view to message a zombie delegate.

--Kyle Sluder

> 
> -- 
> Best regards,
> Andreas Falkenhahnmailto:andr...@falkenhahn.com
> 
___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Re: Mysterious crash with NSTableView

[Andreas Falkenhahn]

On 29.08.2016 at 02:10 Kyle Sluder wrote:

> Delegates are different because they are often messaged in response to
> various exogenous events. Some of these events might happen transiently
> during window teardown, which is usually a time of massive fluctuation
> in an app’s object graph. This is why -windowDidClose: is a good time to
> nil out delegate properties that point back and the window controller
> which ultimately owns the control.

So does this mean that this is potentially dangerous?

[NSApp runModalForWindow:win];
[tableView setDelegate:nil];
[tableView setDataSource:nil];
[win release];

Should I move the "set to nil" calls to -windowDidClose instead?

-- 
Best regards,
 Andreas Falkenhahnmailto:andr...@falkenhahn.com


___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Re: Mysterious crash with NSTableView

[Andreas Falkenhahn]

On 27.08.2016 at 23:54 Quincey Morris wrote:

> On Aug 27, 2016, at 13:55 , Andreas Falkenhahn  wrote:
>
> I still don't know whether setting the delegate to nil
> before release is a general rule or does it only apply to NSTableView?
>
> It’s not clear what delegate you mean by “button delegate”, but it
> doesn’t really matter. 

Oops, I meant the button target as Kyle has already pointed out.

> Incidentally, while it might be a noble goal to have your software
> support OS versions all the way back to 10.6 or 10.5

Just 10.6 for x86. 10.4 for PowerPC but the PowerPC version of my
app uses Carbon anyway.

> Lastly, I may have lost track of things during the series of
> threads we’ve had about memory management in your app, but I can’t
> now understand why you don’t adopt ARC. I guess I thought it was
> because you were maintaining existing code, which presumably did its
> manual memory management properly. But if you’re writing new code,
> or even updating old code, there is literally no downside (AFAIK) in
> switching to ARC. Isn’t it a positive upside in source files shared
> between Mac and other platforms of your app, to not need retain and
> release calls? Note also that ARC and manual memory management can
> be intermixed in the same target freely (though not in the same
> source file), provided the existing manual memory management does follow the 
> rules.
> Since you’re not 100% familiar with the rules, why not just switch to ARC?

The actual Cocoa code in my app is really just a few kilobytes. It's
not complicated at all so there is no pressure to simplify things by
using ARC. If this was a huge project with lots of new code to be
written, it would be a different matter, but it's really just a few
lines. So it's a nice exercise for understanding manual memory management
with Cocoa.

Of course, I might switch to ARC sooner or later but I first
like to learn and understand the old way of doing things... maybe this
also helps me to appreciate ARC more once I start using it ;)

-- 
Best regards,
 Andreas Falkenhahnmailto:andr...@falkenhahn.com


___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Re: Mysterious crash with NSTableView

[Kyle Sluder]

On Sat, Aug 27, 2016, at 03:55 PM, Andreas Falkenhahn wrote:
> On 27.08.2016 at 18:10 Alex Zavatone wrote:
> 
> > Buuut, to the mind of the person learning this or trying to fit all
> > the complexity of learning this in their head, it might be wise in
> > the docs to remind the programmer that they need to do this and why
> > with a, "just in case you are assuming that your code is ready to
> > function now, please remember that the objects that end up making a
> > working tableView require that the tableView's  dataSource object
> > and delegate object need to be assigned, generally to the hosting
> > view controller's instance of self. If this causes you to raise an
> > eyebrow in confusion, the reasons why you must do this are xxx and
> > .  Likewise upon deallocation of the tableView (if it gets
> > called), these objects will need to be set to nil after  but
> > before the tableView object is able to dispose of itself."
> 
> Huh? "Remember that you have to set this property to nil before you
> release the object" would already be sufficient...
> 
> By the way, I still don't know whether setting the delegate to nil
> before release is a general rule or does it only apply to NSTableView?
> Should I also set my button delegates to nil before the buttons are
> released?

Buttons don’t have delegates; they have targets.

Generally applications work fine without nilling out the targets of
their controls. This only works because controls almost exclusively
message their targets synchronously on the main in response to user
actions. There’s no opportunity for one of the target’s dependencies to
be deallocated in between the user triggering the target-action message
and the target receiving the action message.

Delegates are different because they are often messaged in response to
various exogenous events. Some of these events might happen transiently
during window teardown, which is usually a time of massive fluctuation
in an app’s object graph. This is why -windowDidClose: is a good time to
nil out delegate properties that point back and the window controller
which ultimately owns the control.

That said, it is totally possible for a control to delay-perform sending
an event, and even if it takes pains to make sure it doesn’t message the
_target_ if it’s been deallocated, one of the target’s _dependencies_
might have been deallocated. These cases are usually found after much
swearing and trial-by-fire. Such is the difference between theory and
practice.

--Kyle Sluder

> 
> -- 
> Best regards,
>  Andreas Falkenhahn   
>  mailto:andr...@falkenhahn.com

___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Re: Mysterious crash with NSTableView

[Quincey Morris]

On Aug 27, 2016, at 13:55 , Andreas Falkenhahn  wrote:
> 
> I still don't know whether setting the delegate to nil
> before release is a general rule or does it only apply to NSTableView?

It’s not clear what delegate you mean by “button delegate”, but it doesn’t 
really matter. Delegates are not a special case of the rules, just a case where 
you need to pay attention to what the rules imply.

A “delegate” property is “strong”, “zeroing weak” or “unsafe”. (The terminology 
or syntax keyword for each of these has varied over time and context.) If it’s 
strong or zeroing-weak, there’s no need to set the property to nil. If it’s 
unsafe, you generally should set it to nil before you allow the lifetime of the 
delegate object to end.

Therefore, you can look in the API for the property — definitively, in the SDK 
you’re building against — and determine which of the above ownership types the 
property uses.

Note that you will see some inconsistencies in ownership types in different 
parts of Cocoa. That’s because the delegate pattern was standardized somewhere 
around 10.5, but older APIs may have been different and cannot be changed 
without breaking source and/or run-time compatibility.

There are a few real exceptions, which are historical relics, and documented (I 
think) in the ARC transition guide:

— ARC-compatible ownership rules are supported back to (something like) 10.5, 
but zeroing weak pointers weren’t supported until (IIRC) 10.6.8. In the 
interim, zero weak properties were actually unsafe.

— A very few classes (NSWindowController was one of them) did not support the 
zeroing weak mechanism (I mean pointers to objects of these classes, not 
references within these classes to objects of other classes). I believe these 
have all been corrected now, but you might need to be aware of which OS version 
they changed in, if you’re supporting older deployment targets. Or, avoid 
depending on the zeroing part of the behavior.

Incidentally, while it might be a noble goal to have your software support OS 
versions all the way back to 10.6 or 10.5, there are good reasons for moving on 
to later deployment targets. OS X really has gotten better over the years, and 
way-back support starts to become more a source of bugs than a source of 
compatibility.

Lastly, I may have lost track of things during the series of threads we’ve had 
about memory management in your app, but I can’t now understand why you don’t 
adopt ARC. I guess I thought it was because you were maintaining existing code, 
which presumably did its manual memory management properly. But if you’re 
writing new code, or even updating old code, there is literally no downside 
(AFAIK) in switching to ARC. Isn’t it a positive upside in source files shared 
between Mac and other platforms of your app, to not need retain and release 
calls? Note also that ARC and manual memory management can be intermixed in the 
same target freely (though not in the same source file), provided the existing 
manual memory management does follow the rules.

Since you’re not 100% familiar with the rules, why not just switch to ARC?

___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Re: Mysterious crash with NSTableView

[Andreas Falkenhahn]

On 27.08.2016 at 18:10 Alex Zavatone wrote:

> Buuut, to the mind of the person learning this or trying to fit all
> the complexity of learning this in their head, it might be wise in
> the docs to remind the programmer that they need to do this and why
> with a, "just in case you are assuming that your code is ready to
> function now, please remember that the objects that end up making a
> working tableView require that the tableView's  dataSource object
> and delegate object need to be assigned, generally to the hosting
> view controller's instance of self. If this causes you to raise an
> eyebrow in confusion, the reasons why you must do this are xxx and
> .  Likewise upon deallocation of the tableView (if it gets
> called), these objects will need to be set to nil after  but
> before the tableView object is able to dispose of itself."

Huh? "Remember that you have to set this property to nil before you
release the object" would already be sufficient...

By the way, I still don't know whether setting the delegate to nil
before release is a general rule or does it only apply to NSTableView?
Should I also set my button delegates to nil before the buttons are
released?

-- 
Best regards,
 Andreas Falkenhahnmailto:andr...@falkenhahn.com

___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Re: Mysterious crash with NSTableView

[Alex Zavatone]

Sent from my iPad. Please pardon typos.

On Aug 26, 2016, at 10:42 AM, Andreas Falkenhahn  wrote:

> On 26.08.2016 at 17:35 Gary L. Wade wrote:
> 
>> Try clearing your table view's data source and delegate before
>> releasing their object. It appears the pointer gets reassigned to an
>> NSRectSet before your table view completely goes away but after its delegate 
>> and data source have.
> 
> Great, that solves it. Thanks a lot!
> 
> But once again, I think it's a crime that there is no mentioning of this in 
> the class
> documentation of "setDelegate" and "setDatasource" :( 
> 

Agrees as it SHOULD be understood by the programmer as "well, that is simply 
how part of Objective-C works so I should EXPECT to have to set them."

Buuut, to the mind of the person learning this or trying to fit all the 
complexity of learning this in their head, it might be wise in the docs to 
remind the programmer that they need to do this and why with a, "just in case 
you are assuming that your code is ready to function now, please remember that 
the objects that end up making a working tableView require that the tableView's 
 dataSource object and delegate object need to be assigned, generally to the 
hosting view controller's instance of self. If this causes you to raise an 
eyebrow in confusion, the reasons why you must do this are xxx and .  
Likewise upon deallocation of the tableView (if it gets called), these objects 
will need to be set to nil after  but before the tableView object is able 
to dispose of itself."

Please feel free to use this as a running start on an addition to the docs that 
actually explains to the programmer how they will most likely set up their 
tableView.

- Alex Zavatone


> -- 
> Best regards,
> Andreas Falkenhahnmailto:andr...@falkenhahn.com
> 
> ___
> 
> Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)
> 
> Please do not post admin requests or moderator comments to the list.
> Contact the moderators at cocoa-dev-admins(at)lists.apple.com
> 
> Help/Unsubscribe/Update your Subscription:
> https://lists.apple.com/mailman/options/cocoa-dev/zav%40mac.com
> 
> This email sent to z...@mac.com

___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Re: Mysterious crash with NSTableView

[Greg Parker]

> On Aug 26, 2016, at 10:46 PM, Doug Hill  wrote:
> 
>> On Aug 26, 2016, at 9:20 PM, Jeff Szuhay  wrote:
>> 
 On Aug 26, 2016, at 8:44 PM, Sandor Szatmari 
  wrote:
 
 However, in your case I wonder what the static analyzer in Xcode tells you 
 about the bug you see?
>>> 
>>> I believe Andreas mentioned he does not use Xcode as his product is cross 
>>> platform, but this is a good suggestion.
>> 
>> Any why not? 
>> 
>> Sure, build it without Xcode, but couldn’t you create a shell project where 
>> the product doesn’t really matter, then build and use the tools in Xcode?
> 
> I believe you can also invoke the analyzer via the command-line tools.

If your build system runs from the command line and honors the CC and CXX 
environment variables then you may be able to use the scan-build tool to run 
the clang static analyzer.
http://clang-analyzer.llvm.org/scan-build.html


-- 
Greg Parker gpar...@apple.com  Runtime Wrangler



___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Re: Mysterious crash with NSTableView

[Doug Hill]

> On Aug 26, 2016, at 9:20 PM, Jeff Szuhay  wrote:
> 
> 
>>> On Aug 26, 2016, at 8:44 PM, Sandor Szatmari  
>>> wrote:
>>> 
>>> 
>>> However, in your case I wonder what the static analyzer in Xcode tells you 
>>> about the bug you see?
>> 
>> I believe Andreas mentioned he does not use Xcode as his product is cross 
>> platform, but this is a good suggestion.
> 
> Any why not? 
> 
> Sure, build it without Xcode, but couldn’t you create a shell project where 
> the product doesn’t really matter, then build and use the tools in Xcode?

I believe you can also invoke the analyzer via the command-line tools.
___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Re: Mysterious crash with NSTableView

[Jeff Szuhay]

> On Aug 26, 2016, at 8:44 PM, Sandor Szatmari  
> wrote:
> 
>> 
>> However, in your case I wonder what the static analyzer in Xcode tells you 
>> about the bug you see?
> 
> I believe Andreas mentioned he does not use Xcode as his product is cross 
> platform, but this is a good suggestion.

Any why not? 

Sure, build it without Xcode, but couldn’t you create a shell project where the 
product doesn’t really matter, then build and use the tools in Xcode?

___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Re: Mysterious crash with NSTableView

[Sandor Szatmari]

Doug,

> On Aug 26, 2016, at 17:58, Doug Hill  wrote:
> 
>> On Aug 26, 2016, at 1:52 PM, Jens Alfke  wrote:
>> 
>>> On Aug 26, 2016, at 8:42 AM, Andreas Falkenhahn  
>>> wrote:
>>> 
>>> But once again, I think it's a crime that there is no mentioning of this in 
>>> the class
>>> documentation of "setDelegate" and "setDatasource" :(
>> 
>> In the Xcode 8 docs for NSTableView.dataSource, it does:
>>"Note that in versions of OS X prior to v10.12, the table view did not 
>> retain the data source in a managed memory environment."
>> 
>> I’m sure I won’t be the first person to suggest that you switch to ARC. If 
>> you find ref-counting confusing, ARC will help you a lot.
>> 
>> —Jens
> 
> As usual, Jens speaks truthfully about using ARC.
> 
> However, in your case I wonder what the static analyzer in Xcode tells you 
> about the bug you see?

I believe Andreas mentioned he does not use Xcode as his product is cross 
platform, but this is a good suggestion.

Andreas,

If you add a static analysis phase to your Makefile does it help highlight 
these issues?

http://clang-analyzer.llvm.org

Sandor Szatmari

> Can it catch the bug? In manual ref-counting, the analyzer has saved my skin 
> more than once. :)
> 
> Doug
> ___
> 
> Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)
> 
> Please do not post admin requests or moderator comments to the list.
> Contact the moderators at cocoa-dev-admins(at)lists.apple.com
> 
> Help/Unsubscribe/Update your Subscription:
> https://lists.apple.com/mailman/options/cocoa-dev/admin.szatmari.net%40gmail.com
> 
> This email sent to admin.szatmari@gmail.com

___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Re: Mysterious crash with NSTableView

[Doug Hill]

On Aug 26, 2016, at 1:52 PM, Jens Alfke  wrote:
> 
> On Aug 26, 2016, at 8:42 AM, Andreas Falkenhahn  
> wrote:
>> 
>> But once again, I think it's a crime that there is no mentioning of this in 
>> the class
>> documentation of "setDelegate" and "setDatasource" :( 
> 
> In the Xcode 8 docs for NSTableView.dataSource, it does:
>   "Note that in versions of OS X prior to v10.12, the table view did not 
> retain the data source in a managed memory environment."
> 
> I’m sure I won’t be the first person to suggest that you switch to ARC. If 
> you find ref-counting confusing, ARC will help you a lot.
> 
> —Jens

As usual, Jens speaks truthfully about using ARC.

However, in your case I wonder what the static analyzer in Xcode tells you 
about the bug you see? Can it catch the bug? In manual ref-counting, the 
analyzer has saved my skin more than once. :)

Doug
___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Re: Mysterious crash with NSTableView

[Jens Alfke]

> On Aug 26, 2016, at 8:42 AM, Andreas Falkenhahn  
> wrote:
> 
> But once again, I think it's a crime that there is no mentioning of this in 
> the class
> documentation of "setDelegate" and "setDatasource" :( 

In the Xcode 8 docs for NSTableView.dataSource, it does:
"Note that in versions of OS X prior to v10.12, the table view did not 
retain the data source in a managed memory environment."

I’m sure I won’t be the first person to suggest that you switch to ARC. If you 
find ref-counting confusing, ARC will help you a lot.

—Jens
___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Re: Mysterious crash with NSTableView

[Quincey Morris]

On Aug 26, 2016, at 11:39 , Andreas Falkenhahn  wrote:
> 
> From a logical point of view I'd expect this call
> 
>[win release];
> 
> to kill off the NSTableView and its associates completely. So I'd consider
> it safe to  …

Welcome to the latest episode of “It’s Deja Vu All Over Again!”. We used to 
have these sorts of discussions on this list all the time, until ARC, at which 
point they pretty much stopped.

Reasoning about retain counts is extremely difficult, and common sense will 
lead you astray.

In particular, you’re confusing “release” with “deallocate”, which is something 
we all did back in the days when we didn’t use ARC. The concept of releasing 
the “last” reference to an object, the “one” that’s “keeping it alive” is going 
to get you in all sorts of trouble.

>  I mean, who the heck is referencing those delegates after the call to [win 
> release]?

A careful reading of the documentation will tell you that windows are managed 
by a window manager process that’s separate from yours. It will keep windows 
alive until it gets a chance to remove them from its own data structures, which 
may not happen until the next iteration of your run loop.

But even if that weren’t true, the assumption that the table view object won’t 
outlive the window object is invalid.

___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Re: Mysterious crash with NSTableView

[Andreas Falkenhahn]

On 26.08.2016 at 20:12 Kyle Sluder wrote:

> On Fri, Aug 26, 2016, at 12:25 PM, Gary L. Wade wrote:
>> I'm talking about exchanging release for autorelease on your list
>> delegate, which happens after runModal finishes. Since you now do an
>> orderOut, your table view should not need its data source/delegate, so it
>> should be safe keeping them set and using an autorelease, but clearing
>> them is the safest.

> Neither autorelease nor release on its own is sufficient here; Andreas
> should clear out his delegates. 

So should I also clear the button delegates or is NSTableView an exception
here?

Side note:

What I find really weird in this whole thing is the fact that it crashed
although the delegates are clearly released *after* the window. I wouldn't
be surprised to get crashes if I released the delegates *before* the window.
That would be natural behaviour but from a logical point of view it's really
weird to see it crash with the delegate/data source being released *after*
the window, i.e. when one would expect NSTableView to have been fully 
deallocated
already.

From a logical point of view I'd expect this call

[win release];

to kill off the NSTableView and its associates completely. So I'd consider
it safe to release the delegate/data source after that, but apparently they
need to be cleared out before. It's good to know but from a logical point
of view I still think it's weird. I mean, who the heck is referencing those
delegates after the call to [win release]? Everything should be gone by
that call but there's probably some advanced wizardry going on that I'm
unaware of.

-- 
Best regards,
 Andreas Falkenhahnmailto:andr...@falkenhahn.com


___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Re: Mysterious crash with NSTableView

[Kyle Sluder]

On Fri, Aug 26, 2016, at 12:25 PM, Gary L. Wade wrote:
> I'm talking about exchanging release for autorelease on your list
> delegate, which happens after runModal finishes. Since you now do an
> orderOut, your table view should not need its data source/delegate, so it
> should be safe keeping them set and using an autorelease, but clearing
> them is the safest.

Neither autorelease nor release on its own is sufficient here; Andreas
should clear out his delegates. This is true either in ARC or non-ARC,
whether the delegate property is declared `unsafe_unretained` (née
`assign`) or `weak`.

You don’t know when an object will be deallocated. While it would be
unexpected (and arguably an error) for NSTableView to hand off its
delegate to something that will take a long-lived strong +1 reference to
it, it is perfectly legal for the delegate to wind up in the autorelease
pool for an indefinite period of time. Delegates are most likely not
written to handle being called in that limbo state where they’re still
alive and assigned to the delegate property, but the object that created
them has disavowed them by releasing its last strong reference to them.

`weak` prevents retain cycles. It does not ensure program correctness.

> 
> This kind of exercise in understanding the MRC rules is precisely why ARC
> was created. If you need to use MRC, you probably should do what lots of
> others did and draw lots of pictures in timeframes with tally counts on
> objects to understand what happens when. If you go with ARC, this
> clearing should happen for you—I'm pretty sure the data source and
> delegate are weak, but I don't have the headers in front of me right now.

Automatic clearing of zeroing weak references is dependent on the
*implementor* being compiled with ARC. Non-ARC clients will still see
another object’s zeroing weak ref go to nil when the last strong ref to
the referent is released.

Here’s some sample code:

/* WeakRefHolder.h
 * This file is transcluded twice: once with ARC enabled, once with ARC
 disabled.
 */
#import 

@interface WeakRefHolder : NSObject

@property(weak) id weakRef;

@end
/* WeakRefHolder.m
 * clang -c -o WeakRefHolder.o -fobjc-arc WeakRefHolder.m
 */

#if !__has_feature(objc_arc)
#error WeakRefHolder implementation must be compiled with ARC
#endif

#import "WeakRefHolder.h"

@implementation WeakRefHolder
// Automatically synthesize implementation of .weakRef property
@end

/* main.m
 * Compile: clang -o main -framework Foundation -fno-objc-arc main.m
 WeakRefHolder.o
 */

#if __has_feature(objc_arc)
#error Main file should not be compiled with ARC
#endif

#import "WeakRefHolder.h"
#import 

@interface SquealOnDealloc : NSObject
@end

@implementation SquealOnDealloc
- (void)dealloc {
printf("<%p> deallocating!\n", self);
[super dealloc];
}
@end

int main(int argc, char **argv)
{
WeakRefHolder *holder = [[WeakRefHolder alloc] init];
id referent;
@autoreleasepool {
printf(">> Pushing autorelease pool\n");
referent = [[SquealOnDealloc alloc] init];
printf("Assigning <%p> to .weakRef property\n",
referent);
holder.weakRef = [referent autorelease];
printf("<< Popping autorelease pool\n");
}

printf(".weakRef = %p\n", holder.weakRef);
return 0;
}

--Kyle Sluder

> --
> Gary L. Wade (Sent from my iPhone)
> http://www.garywade.com/
> 
> > On Aug 26, 2016, at 9:38 AM, Andreas Falkenhahn  
> > wrote:
> > 
> >> On 26.08.2016 at 17:52 Gary L. Wade wrote:
> >> 
> >> You would not see this if you hid or removed the table view first
> >> since it would not need its data source or delegate then. Try going
> >> with ARC or at least use autorelease on your delegate/data source.
> > 
> > I'm not using autorelease on the delegate/data source on purpose because
> > AFAIU autoreleased objects are (potentially) killed whenever the application
> > is in an event loop and this is the case with runModalForWindow(). AFAIU
> > if I used autorelease on the delegate/data source then both would be
> > killed in the run loop started by runModalForWindow() because both 
> > setDelegate()
> > and setDatasource() don't retain. So using autorelease doesn't make sense
> > to me here. Is this right or did I get anything wrong here?
> > 
> > -- 
> > Best regards,
> > Andreas Falkenhahnmailto:andr...@falkenhahn.com
> >

___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Re: Mysterious crash with NSTableView

[Gary L. Wade]

I'm talking about exchanging release for autorelease on your list delegate, 
which happens after runModal finishes. Since you now do an orderOut, your table 
view should not need its data source/delegate, so it should be safe keeping 
them set and using an autorelease, but clearing them is the safest.

This kind of exercise in understanding the MRC rules is precisely why ARC was 
created. If you need to use MRC, you probably should do what lots of others did 
and draw lots of pictures in timeframes with tally counts on objects to 
understand what happens when. If you go with ARC, this clearing should happen 
for you—I'm pretty sure the data source and delegate are weak, but I don't have 
the headers in front of me right now.
--
Gary L. Wade (Sent from my iPhone)
http://www.garywade.com/

> On Aug 26, 2016, at 9:38 AM, Andreas Falkenhahn  
> wrote:
> 
>> On 26.08.2016 at 17:52 Gary L. Wade wrote:
>> 
>> You would not see this if you hid or removed the table view first
>> since it would not need its data source or delegate then. Try going
>> with ARC or at least use autorelease on your delegate/data source.
> 
> I'm not using autorelease on the delegate/data source on purpose because
> AFAIU autoreleased objects are (potentially) killed whenever the application
> is in an event loop and this is the case with runModalForWindow(). AFAIU
> if I used autorelease on the delegate/data source then both would be
> killed in the run loop started by runModalForWindow() because both 
> setDelegate()
> and setDatasource() don't retain. So using autorelease doesn't make sense
> to me here. Is this right or did I get anything wrong here?
> 
> -- 
> Best regards,
> Andreas Falkenhahnmailto:andr...@falkenhahn.com
> 


___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Re: Mysterious crash with NSTableView

[Andreas Falkenhahn]

On 26.08.2016 at 17:52 Gary L. Wade wrote:

> You would not see this if you hid or removed the table view first
> since it would not need its data source or delegate then. Try going
> with ARC or at least use autorelease on your delegate/data source.

I'm not using autorelease on the delegate/data source on purpose because
AFAIU autoreleased objects are (potentially) killed whenever the application
is in an event loop and this is the case with runModalForWindow(). AFAIU
if I used autorelease on the delegate/data source then both would be
killed in the run loop started by runModalForWindow() because both setDelegate()
and setDatasource() don't retain. So using autorelease doesn't make sense
to me here. Is this right or did I get anything wrong here?

-- 
Best regards,
 Andreas Falkenhahnmailto:andr...@falkenhahn.com

___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Re: Mysterious crash with NSTableView

[Gary L. Wade]

You would not see this if you hid or removed the table view first since it 
would not need its data source or delegate then. Try going with ARC or at least 
use autorelease on your delegate/data source.
--
Gary L. Wade (Sent from my iPad)
http://www.garywade.com/

> On Aug 26, 2016, at 8:42 AM, Andreas Falkenhahn  
> wrote:
> 
>> On 26.08.2016 at 17:35 Gary L. Wade wrote:
>> 
>> Try clearing your table view's data source and delegate before
>> releasing their object. It appears the pointer gets reassigned to an
>> NSRectSet before your table view completely goes away but after its delegate 
>> and data source have.
> 
> Great, that solves it. Thanks a lot!
> 
> But once again, I think it's a crime that there is no mentioning of this in 
> the class
> documentation of "setDelegate" and "setDatasource" :( 
> 
> -- 
> Best regards,
> Andreas Falkenhahnmailto:andr...@falkenhahn.com
> 


___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Re: Mysterious crash with NSTableView

[Andreas Falkenhahn]

On 26.08.2016 at 17:35 Gary L. Wade wrote:

> Try clearing your table view's data source and delegate before
> releasing their object. It appears the pointer gets reassigned to an
> NSRectSet before your table view completely goes away but after its delegate 
> and data source have.

Great, that solves it. Thanks a lot!

But once again, I think it's a crime that there is no mentioning of this in the 
class
documentation of "setDelegate" and "setDatasource" :( 

-- 
Best regards,
 Andreas Falkenhahnmailto:andr...@falkenhahn.com

___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Re: Mysterious crash with NSTableView

[Gary L. Wade]

Try clearing your table view's data source and delegate before releasing their 
object. It appears the pointer gets reassigned to an NSRectSet before your 
table view completely goes away but after its delegate and data source have.
--
Gary L. Wade (Sent from my iPad)
http://www.garywade.com/

> On Aug 26, 2016, at 8:15 AM, Andreas Falkenhahn  
> wrote:
> 
> Does anybody have an idea why the following code causes a "Segmentation 
> fault" and thus
> an immediate program termination? Interestingly, the code runs just fine and 
> the
> NSTableView appears correctly and is functional. The "Segmentation fault" 
> occurs when
> buttonPressed() is left and control returns to the main run loop. Thus, I 
> suspect that
> the error is probably related to autoreleasing, e.g. releasing a resource 
> twice or
> something. But I don't really see anything that's wrong with my code. 
> Sometimes I also
> get this message instead of the segmentation fault:
> 
>   [NSRectSet tableView:objectValueForTableColumn:row:]:unrecognized 
> selector sent to instance 0x100153060
> 
> But most of the time it just crashes with a segmentation fault. The code 
> itself is
> really simple and straightforward, here it is:
> 
> - (void)buttonPressed
> {
>NSRect rect = NSMakeRect(100, 100, 320 + 2 * 20, 78 + 200 + 20);
>
>NSWindow *win = [[NSWindow alloc] initWithContentRect:rect 
> styleMask:NSTitledWindowMask backing:NSBackingStoreBuffered defer:NO];
> 
>id listDelegate = [[MyListDelegate alloc] init];
> 
>NSScrollView *scrollview = [[NSScrollView alloc] 
> initWithFrame:NSMakeRect(20, 52, rect.size.width - 2 * 20, 200)];
>NSTableView *tableview = [[NSTableView alloc] initWithFrame:NSMakeRect(0, 
> 0, rect.size.width - 2 * 20 - 16, 200)];
>NSTableColumn *column = [[NSTableColumn alloc] 
> initWithIdentifier:@"Column"];
>
>[tableview addTableColumn:column];
>[tableview setDelegate:listDelegate];
>[tableview setDataSource:listDelegate];
>[scrollview setDocumentView:tableview];
>[[win contentView] addSubview:scrollview];
>[scrollview release];
>[tableview release];
>[column release];
> 
>id buttonDelegateOK = [[MyButtonDelegate alloc] init];
>NSButton *ok = [[NSButton alloc] initWithFrame:NSMakeRect(rect.size.width 
> - 82 - 14, 12, 82, 32)];
>[ok setTitle:@"OK"];
>[ok setButtonType:NSMomentaryPushInButton];
>[ok setBezelStyle:NSRoundedBezelStyle];
>[ok setKeyEquivalent:@"\r"];
>[ok setTarget:buttonDelegateOK];
>[ok setAction:@selector(buttonPressed)];
>[[win contentView] addSubview:ok];
>[ok release];
>
>[NSApp runModalForWindow:win];
>
>[win orderOut:nil];
>[win release];
>
>[listDelegate release];
>[buttonDelegateOK release];
> } 
> 
> ---
> 
> The list delegate is also minimal and looks like this:
> 
> @interface MyListDelegate : NSObject
> @end
> 
> @implementation MyListDelegate
> - (int)numberOfRowsInTableView:(NSTableView *)_tableView
> {
>return 3;
> }
> 
> - (id)tableView:(NSTableView *)_tableView 
> objectValueForTableColumn:(NSTableColumn *) tableColumn row:(int)row
> {
>return @"Foobar";
> }
> @end
> 
> ---
> 
> I've already spent several hours trying to find out why this particular code 
> results
> in a "Segmentation fault" but I just don't see it.
> 
> Can anybody help? Thanks in advance!
> 
> Full minimal demo program is attached for reference.
> 
> -- 
> Best regards,
> Andreas Falkenhahn  mailto:andr...@falkenhahn.com
> 


___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com

Mysterious crash with NSTableView

[Andreas Falkenhahn]

Does anybody have an idea why the following code causes a "Segmentation fault" 
and thus
an immediate program termination? Interestingly, the code runs just fine and the
NSTableView appears correctly and is functional. The "Segmentation fault" 
occurs when
buttonPressed() is left and control returns to the main run loop. Thus, I 
suspect that
the error is probably related to autoreleasing, e.g. releasing a resource twice 
or
something. But I don't really see anything that's wrong with my code. Sometimes 
I also
get this message instead of the segmentation fault:

   [NSRectSet tableView:objectValueForTableColumn:row:]:unrecognized 
selector sent to instance 0x100153060

But most of the time it just crashes with a segmentation fault. The code itself 
is
really simple and straightforward, here it is:

- (void)buttonPressed
{   
NSRect rect = NSMakeRect(100, 100, 320 + 2 * 20, 78 + 200 + 20);

NSWindow *win = [[NSWindow alloc] initWithContentRect:rect 
styleMask:NSTitledWindowMask backing:NSBackingStoreBuffered defer:NO];

id listDelegate = [[MyListDelegate alloc] init];

NSScrollView *scrollview = [[NSScrollView alloc] 
initWithFrame:NSMakeRect(20, 52, rect.size.width - 2 * 20, 200)];
NSTableView *tableview = [[NSTableView alloc] 
initWithFrame:NSMakeRect(0, 0, rect.size.width - 2 * 20 - 16, 200)];
NSTableColumn *column = [[NSTableColumn alloc] 
initWithIdentifier:@"Column"];

[tableview addTableColumn:column];
[tableview setDelegate:listDelegate];
[tableview setDataSource:listDelegate];
[scrollview setDocumentView:tableview];
[[win contentView] addSubview:scrollview];
[scrollview release];
[tableview release];
[column release];

id buttonDelegateOK = [[MyButtonDelegate alloc] init];
NSButton *ok = [[NSButton alloc] 
initWithFrame:NSMakeRect(rect.size.width - 82 - 14, 12, 82, 32)];
[ok setTitle:@"OK"];
[ok setButtonType:NSMomentaryPushInButton];
[ok setBezelStyle:NSRoundedBezelStyle];
[ok setKeyEquivalent:@"\r"];
[ok setTarget:buttonDelegateOK];
[ok setAction:@selector(buttonPressed)];
[[win contentView] addSubview:ok];
[ok release];

[NSApp runModalForWindow:win];

[win orderOut:nil];
[win release];

[listDelegate release];
[buttonDelegateOK release];
} 

---

The list delegate is also minimal and looks like this:

@interface MyListDelegate : NSObject
@end

@implementation MyListDelegate
- (int)numberOfRowsInTableView:(NSTableView *)_tableView
{
return 3;
}

- (id)tableView:(NSTableView *)_tableView 
objectValueForTableColumn:(NSTableColumn *) tableColumn row:(int)row
{
return @"Foobar";
}   
@end

---

I've already spent several hours trying to find out why this particular code 
results
in a "Segmentation fault" but I just don't see it.

Can anybody help? Thanks in advance!

Full minimal demo program is attached for reference.

-- 
Best regards,
 Andreas Falkenhahn  mailto:andr...@falkenhahn.com

main.m
Description: Binary data
___

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to arch...@mail-archive.com