Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.
Comment: Migrated to Confluence 5.3

...

You can see that the CLW file contains a declaration of a LONG variable, declared outside the class but inside the class's source module. The one method in the class returns the address of this variable. 

Now hereHere's a small program to prove that SomeClassModuleData is the same in all instances of DemoClass:

...

You can see why I had a problem with multiple instances of AnyAsciiFileClass, What I needed was one file structure for each instance of the class. Only it isn't possible to declare a file layout inside a class. Files have to be declared at compile time; you can't create the on the fly at runtime (at least not without the Dynamic File Driver or DynaLib). 

...

Code Block
TestFile                            DCL_System_IO_AsciiFile
AnotherTestFile                     DCL_System_IO_AsciiFile
! etc

Pool management

I decided to treat this as a pool management problem. I needed a list of all of the available file layouts, and a way to mark a file as in use. Then in the constructor I would check the list and initialize the class instance with an available file layout. 

My first thought was to write this pool management code as a separate class. My second thought was that the code would be pretty simple, so I might as well just do it with a few procedures and a queue declared in the class source module. And my third thought, after about five minutes of coding, was that it would probably be best after all to write the code as a separate class, which is what I did.

Sometimes when I create a new class the name of the class is obvious; other times I agonize over my choice, often changing it later. I'm still mulling over this one. In any case, I decided to call it DCL_System_Pool, although it's not really an object pool, more of a reference number pool. But it can server as a basis for more involved pool implementations, so for now DCL_System_Pool it is. If you're reading this years later and you don't see a class by that name in the DCL, keep looking. It's probably there somewhere. 

...

The logic is pretty easy to follow; when you get a pool element it's marked as in use; when you're done with it and release it the flag is cleared. There's also some support for customizing error handling. And to To make the class thread safe I've added a critical section.

Clearly even if this is the only place I ever use the class, having it in its own INC and CLW makes the class easier to work with , and more visible within the DCL library

Inside DCL_System_IO_AsciiFileManager I added five file layouts:

...

Code Block
DCL_System_IO_AsciiFile.Construct        PROCEDURE()
    CODE
	self.Errors &= new DCL_System_ErrorManager
	self.PoolItemNumber = AsciiFilePool.GetItemNumber()
	execute self.PoolItemNumber
		self.Init(AsciiFile1,AsciiFile1:Record,AsciiFileName1)
		self.Init(AsciiFile2,AsciiFile2:Record,AsciiFileName2)
		self.Init(AsciiFile3,AsciiFile3:Record,AsciiFileName3)
		self.Init(AsciiFile4,AsciiFile4:Record,AsciiFileName4)
		self.Init(AsciiFile5,AsciiFile5:Record,AsciiFileName5)
	else
		stop('Unable to initialize the ASCIIFile instance - if you continue the program will probably crash')
	end
	
DCL_System_IO_AsciiFile.Destruct         PROCEDURE()
	CODE
    SELF.CloseFile()
	AsciiFilePool.ReleaseItemNumber(self.PoolItemNumber)
	dispose(self.Errors)


I've updated the GitHub repository with these changes. If you previously used the ASCIIFileManager approach then you'll need to make two simple changes:

  • Remove the ASCIIFileManager call
  • Most importantly, remove the leading & from your DCL_System_IO_AsciiFile declaration to change it from a reference to an instance.

I wish I'd made this change sooner - the refactoring wasn't all that difficult, the class is much, much easier to use, and I've added a new pool tool to the kit in the event I need similar functionality somewhere else. 

Note
titleLate breaking weirdness

After I posted this article and a shorter one titled Class exporter now a pre-build task (DCL) I decided I needed to deal with a circular reference problem in the class exporter, which now uses DCL_System_IO_AsciiFile to read its list of class headers to parse from a text file.

Originally CreateDCLExportFile made use of DevRoadmapsClarion.DLL, but because CreateDCLExportFile is now a pre-build task for the class library there's a risk that a change to the class library could cause a runtime failure which would break the class library build. With the build broken, how do I then create the EXP so the build can succeed? Big problem.

But wait, that's no problem at all. I can do the same thing with CreateDCLExportFile that I did with ClarionTest - compile in the required classes so I don't need the DLL. Brilliant!

Except when I did that, CreateDCLExportFile GPFd.

I actually resorted to the debugger to see what was the problem. And I found that in an EXE my AsciiFilePool class was not being instantiated. Why, I can't imagine.

The fix was to change AsciiFilePool to a typed class and set up a reference:

 

Code Block
AsciiFilePoolType                       class(DCL_System_Pool),type
Construct                                   procedure
										end
AsciiFilePoolInstance                   &AsciiFilePoolType

The first line of my DCL_System_IO_AsciiFile.Construct method now has this line:

Code Block
if AsciiFilePoolInstance &= null then AsciiFilePoolInstance &= new AsciiFilePoolType.

This way I'm guaranteed to have a pool instance.

I find it quite strange that a module-level instance should need to be instantiated for EXEs but not DLLs, but there it is. If anyone can shed light on this please do so.


Info
titleWeirdness mystery solved!

After further investigation I discovered the reason for the GPF. It wasn't that the pool instance wasn't created, it was that in the EXE the pool instance was created after the file class instance. That meant the file class constructor fired before it had a pool instance to work with. In the DLL however the pool instance was created first.

I'm not sure why that should be the case; perhaps in the EXE anything in declared the module is only created when the module itself comes into scope (as when the class is instantiated), while in the DLL all static declarations are automatically instantiated on DLL loading. But that's just speculation on my part.

In any case, taking control of the pool object's instantiation solves the problem.