Random memory page errors causing crashing

[DeanH]

Some of my users have been reporting random crashes. The program, written entirely in PureBasic for Windows, randomly shuts down and vanishes off the screen. In only one case, the program was still in the Task Manager but not visible. In all other cases the exe has shut down. More often than not, it happens after the program has been un-minimized but not always. It can happen at any point at any time. It can happen in the middle of typing something. I originally thought it might have something to do with input as it often happened when data was enter into a StringGadget but this behaviour has also happened on other statements, even LoadImage and DatabaseUpdtae. The latest was End.

I set up OnErrorCall to trap for errors. The only thing appearing is a Memory Page Error and that does not always appear. The exe can still shut down without reporting an error. The latest error took place on a MessageRequester line (below).

Code: Select all

;ErrorMessage$ is a string containing the ErrorMessage(), ErrorLine() and ErrorFile()
If MessageRequester("Error", ErrorMessage$,#PB_MessageRequester_YesNo|#PB_MessageRequester_Error)>#PB_MessageRequester_Yes
	End
EndIf

The error was reported on the End statement!

I have been investigating this for months. I have tried different compiler versions of PB, including 6.

I do not think it is a PureBasic problem but some type of interference. According to the users the folder or mapped drive is excluded from Defender checking.

This is only happening to a fraction of my users but there are several. I am stumped and it is driving these users crazy. Any ideas?

[Caronte3D]

When I have randomly shuts down and vanishes off the screen, offten is because one of two reasons:
1.- Read or write global variables on a threaded program without mutex
2.- Invalid Memory read or write

[DeanH]

Caronte3D, can you please clarify. The programs (there are several modules) were compiled unthreaded. I tried threaded but it made no difference. Can you explain the mutex aspect and what I could do. There are a great deal of global variables, arrays, lists and maps.

It is also odd this behaviour only started mid last year after over 10 years.

As to memory read/write, I checked for that. I do very little direct memory access. Just normal PB variable = assignment. I have also made sure libraries are not left open, files and databases (SQLite) are not left open. Callbacks are at a minimum.

I have tried the C-backend compiler but it keeps hanging on me during compilation.

[idle]

Do you have any threads? If you do are there any Global variables or resources that are written to over thread boundaries?

[Bitblazer]

It is also odd this behaviour only started mid last year after over 10 years.

• Try to find backup versions as late as possible which always worked.
• Make a diff to your current sources.
• Start adding everything one by one to find out where the problem started.
• Look at the last addition / change

There you have your bug.

Yes, it is a lot work. Work that should have happened a year ago
But it does give you answers. If you think it is a compiler bug, go to the purebasic museum and get the compiler version you used back then and see if that works.

For large projects, a backup history is very useful. At every major stable point of an important project, do a seperate backup and keep it in your project history. Storage is dirt cheap nowadays.

If i had to do this, i would create two project folders, get good music, snacks and use WinMerge.

[Caronte3D]

Caronte3D, can you please clarify.

If you program doesn´t use threads the mutex is not used so not problem accessing global variables at all.

Good luck, that type of bugs that occurs randomly being a pain

[Caronte3D]

Some times using a "OnError" procedure helps a lot to find those bugs because you see the line number where it occurs

[DeanH]

Thank you for the suggestions. I have tried all of those.

I have been using OnError. That's how I saw the Memory Page Error message. It reports the problem at seemingly random points. No consistency at all. Sometimes the error isn't even reported. The exe just shuts down. One user reported two instances of it happening yesterday.

I keep backups of various versions going back into the 90's. Current version is 10.8.1 and I have versions right back to 10.1. Database has evolved, though, so there's a limit as to how far back I can try. I put one of the slightly "older" versions (10.7.5) into a site but it, too, showed the same random vanishing behaviouir. I also tried compiling using an older version of PB, but that's limited as I have used some of the more recent functions like HttpRequest. No difference between using PB 5.73 and 6.00.

I am thinking some type of background task is causing this but have no idea what it might be. The schools involved claim their anti-virus systems have excluded the mapped drive. They're all using workstations, and I did try it on a local C: drive, which reduced the frequence but did not elmininate it.

[Bitblazer]

This is only happening to a fraction of my users but there are several. I am stumped and it is driving these users crazy. Any ideas?

I have been using OnError. That's how I saw the Memory Page Error message. It reports the problem at seemingly random points. No consistency at all. Sometimes the error isn't even reported. The exe just shuts down. One user reported two instances of it happening yesterday.

Did you already try to run your project with the purifier settings set to very restrictive checks? Because a tricky memory heap corruption by under/overwriting could do that.

[juergenkulow]

Do you still have the ErrorAddress() from the memory page error, the assembler instruction before ErrorAddress(), the complete stack ending at rsp or esp and the contents of the important registers and important variables like event?(LogError.txt) I prefer OnErrorGoto instead of OnErrorCall or the VS2008 debugger.

[BarryG]

Dean, these types of errors are annoying to find, I know. See my experience here -> https://www.purebasic.fr/english/viewto ... 25#p541025

I've also found other mysterious crashes by using a dedicated error handler procedure with ErrorLine support turned on. It was based on this -> https://www.purebasic.fr/english/viewto ... 92#p491192

Use that error handler code - it really helped me.

[hoerbie]

Just one idea, when reading

functions like HttpRequest

and

were compiled unthreaded

do you use the async mode of HttpRequest?
I've had a lot of crashes, until I put the async HttpRequest commands in a thread and compiling thread safe. Using async HttpRequest around/between WaitWindowEvent(timeout) had crashes in real life on different and simple code lines, but never crashed when slowly going through the program with the debugger step by step.

[TassyJim]

It reminds me of the problems I had when I started to compile in 64bit.
I had used longs where I should have been using integers.

Jim

[DeanH]

Thanks everyone.

I ran into the integer issue, too, when switching to 64-bit a few years ago. I went through the source code and took care of that. I had to go to 64-bit due to constant anti-virus false positives, especially after PureLocker turned up. I have tried both 32 and 64 bit in these sites but both behaved the same.

Have not had this problem with any HttpRequest. They are all separate from WindowEvent or WaitWindowEvent loops.

I'll try the purifier trick. I did try threadsafe but that made no difference. Still happened.

This morning I asked a user to revert back to a version from mid 2020. Back then this did not happen. But it happened this morning to them. I am now feeling the problem is caused by a background task. Unfortunately I do not have a clue as to what that might be. It happens mostly when a minimized window has been brought back up and the user tries to do something. But it can also happen without minimizing, too (check that).

Thank you everyone so far for the suggestions. This problem remains an annoying mystery.

[skywalk]

Are there any hardware similarities between the users that crash?
What windows version/build do you bless?
Do you have your own log you can use to dump time, procedure calls, etc?