Threads and CriticalSections

[Marcel Kollenaar]

Hi,

I hope there are some thread folks around here.

Imagine: there are two threads running, they run independend from each other and there is no way to tell which of them is serviced first. They have sometimes to access the same data structure but it not allowed that they do that at the same time. One thread has to wait at the other to change the structure. Is there a A5 mechanism for mutual exclusion which cost no CPU cycles?

Synchronizing with manual events is already established.

Marcel

[Jos Schaars]

RE: Threads and CriticalSections

Marcel,

Sounds somewhat far fetched, are you going where no one has gone before and exploring theorectical boundaries or do you see any pratical aplyances for such a scenario?

Jos

[Marcel Kollenaar]

RE: Threads and CriticalSections

Hi Jos,

"Sounds somewhat far fetched, ..."

It depends, A5v6 introduced Threads togheter with the WAS. So why not use it when you need it.

"are you going where no one has gone before..."

Yes, within Alpha this is some new playground .. discovering the edge, comes pretty near.

"and exploring theorectical boundaries or do you see any pratical aplyances for such a scenario?"

Theoretical, yes, read a lot about WIN32 processes and threads. I'm busy with some slow processes and I can use threads that do the dirty work for me. Another positive side effect is - when correct implemented - your user interface is responding better at the moment the slow process runs.

Sometimes I put a question here in the hope another A5 programmer has the same fire burning on this particular subject.

Marcel

[csda1]

RE: Threads and CriticalSections

Marcel,

What you are asking for is the ability to perform an "Atomic" operation. Atomic implies indivisibility and irreducibility.

The ability to execute a piece of code that can not be interrupted so that a control can be implemented is crucial to sharing resources (e.g. records, tables, variables etc.).

If you don't have this ability, it is possible to hang waiting for a required resource to be available or to corrupt a resource.

The easiest way to implement these basic functions requires a set and test function, e.g. decrement flag and compare to 0 instruction.

Does A5 have such capability, either high or low level? It must, but may not be exposed to the user. One way that it can definitely be accomplished is using a record of a table and attempting to get a lock by attempting to change it. If you get the record lock, then you can procede with operation until you release the lock by a save or cancel.

But this is a real waste of system resources to get around a limitation, so a search for another way would be ideal. I've experimented with Batch_begin and Batch_end, but I don't think that guarantees exclusive control.

Regards,

Ira

[Marcel Kollenaar]

RE: Threads and CriticalSections

Hi Ira,

Thank you for your insight.

My question has simply to do with multithreading. A5v6 can do this so I want to use it because I need it in a my solution, that's why I upgraded.

Sure I can build this completely in PureBasic or C or whatever and put it in a DLL. But creating multithreading from within a DLL is also another difficult issue. So beaten by the dog or cat, I don't know what's better. But it is a challenge to build it completely in Xbasic.

The problem is that there are some issues that aren't covert in A5. I have sometimes the feeling that A5 provide me a brush and the paint but ommitted the canvas. And painting my backdoor is already done.

Xbasic is a tremendous strong language from a certain point of view. I get the feeling that what I want isn't possible in Xbasic or I have to simplify it at a level that even Donald Duck can use it, and that's not my goal.

At the moment I put A5 on stack and take a break.

Marcel

[csda1]

RE: Threads and CriticalSections

Marcel,

That's what I am talking about. You want a resource flag that you can tell when your data structure is available for you to use without the other task being able to interrupt the task as it does an "atomic" operation.

There is nothing that would be free in terms of CPU cycles, but it can be minimal. The simplest way might be to use global variables as flags, as long as the setting and testing of them can occur in an atomic operation. If not, there is always a small window (although very small) of time where the testing of the flag can think it's OK to proceed, but in fact the other grabbed the resource.

An exclusive open or a record lock in change mode of a shared file or table are two possible ways to do this for sure, but it is real inefficient since it is using the operating system to access a them in order to serve the function of a flag. User code that does Autoincrementing code is an example of trying to do this. There have been several examples of this on the board, some better, some worse, and some loaded with bugs.


As for your statement;
"i"I have sometimes the feeling that A5 provide me a brush and the paint but ommitted the canvas. And painting my backdoor is already done. "/i"

I totally agree. There are many critical functions that I have asked for in the past that, while somewhat esoteric and would be used by relatively few users, they would be enabling technology allowing others to build on that.

I know I asked for atomic database operations (could be as simple as a LOCK and UNLOCK that would prevent code from being interupted by other database processes on the network) as far back as Alpha FOUR version 3 (although it could have been version 2.1)

Regards,

Ira