Hi,

I'm writing a Hollywood Genealogy database and at random times the most recent Linux Hosted Aros 32 bit version freezes. I wrote a routine which simply displays every person from 1 to 7000. It may freeze on person 55 or person 1125 or whatever random person but has never made it through all people. There is no Hollywood crash or error, AROS just freezes. I ran the exact program with the same sqlite3 version on Amiga OS3 and Native AROS 32 bit via virtualbox and have had no problem at all. How do I determine why it freezes to provide information for debugging purposes and would that issue still be fixed in a new version of Linux Hosted AROS 32 bit? Thanks.

NathanH

A few comments from me:

1) Does anything appear on the linux console from which AROS was started?
2) What do you mean by simply display? Are you opening a window for each person, printing the data to shell, etc..

Best if you recorded a video showing the listing process up to point of freezing while showing the linux console as well.

My first impression is that it is some sort of Intuition lock up or graphics driver lookup. I noticed you have some programs running in the backgroun - user menus, clock and dock bar. Please disable anything you are autostarting and see if that helps in any way,

@deadwood,

I took everything out of the WBStartup folder so nothing extra is running, opened a shell from the menu to start the program, and get the same result. It freezes on random entries between (now) 6 and in the 1000s. Last one was person I432.

NathanH

If you can prepare a fake database with smallest number of entries that causes lockup and share the database and program with me, I might be able to find out what is going on through debugger.

Try C:tasklist to see if there's any task with stack size smaller than 40960.

@aros-sg,

Thanks. I ran C:Tasklist and all of the running programs (including the one in question) had stacks of at least 40960.

NathanH

@deadwood,

My wife suggested getting a fake ged file from ChatGPT and after many versions, I obtained a good dataset to import into my genealogy app. The genes.gid file tells the app what file to load at startup. ChatGPT.genez is a zipped sqlite database with data that was imported from a ged file created by ChatGPT. From shell, CD into the directory and type run genes_aros. I run a routine on opening which displays each person record in turn. Random AROS Linux Hosted freezes occur on random people.

Hollywood threw an error at one point for an edge-case person and after fixing that I wanted to see if there were any other people with edge-cases that might throw a Hollywood error. That's why I originally wrote this subroutine. It then exposed this Hosted Freezing issue. As noted earlier, the freeze doesn't happen on AROSOne via Virtualbox nor on AmigaOS3.

You might want to replace your sqlite installation with version 3.34a (AROS Archives or Aminet) as I've found it to be much faster and runs without requestors when invoked from the CLI.

NathanH

When listing items, you are changing shape of the cursor to "Zzz" and back. I had lock up twice, each time inside linux code supporting that function. Try disabling changing the cursor and see if that helps with stability.

In arch/all-hosted/hidd/x11/x11gfx_hiddclass.c add LOCK_X11 and UNLOCK_X11 around XcursorImageCreate() and XcursorImageDestroy().

Btw, with debug version of AROS hosted one can do this without changing/recompiling anything. When running AROS under gdb one can do:



Whenever the breakpoint is hit it will execute the commands "silent", "return", "cont".

@deadwood and aros-g,

Thanks guys for taking the time and effort to look at this! Using Hollywood scripting in a sandbox, I imagine I'm about as far from calling OS library functions as is possible. I don't call any functions that explicitly change the cursor to or from busy. That automatically happens in MUIRoyale when you sleep or unsleep windows. I put windows to sleep whenever I query sqlite (as it takes a bit of time, esp. on OS39) so that the user doesn't try to change things until the system is in a known state. I appreciate the fact that I can do things in ways that are less likely to expose problems in the underlying system and will keep that in mind. There are probably places where I can avoid unsleeping until collectively more sqlite calls are made to help with stability; frankly, I had no clue that it would matter so much. Are there other gotchas between between AROS and AROS hosted that are documented or documentable?

Thanks again for looking at it. I really love the model/expediency of AROS Hosted but will likely need to be more cautious when using it. Thanks.

NathanH

@NathanH

Looks like fix from @aros-sg solved your problem. I ran Genes until it reached ID ~10000 and didn't have a lock up anymore. See the driver attached (to be placed in DEVSrivers).

@deadwood and @aros-sg,

It worked fine with my live data as well! Thank you for continuing to update the Linux Hosted 32-bit environment. My AROS usage is mainly for Hollywood development for OS3/OS4/AROS and the Linux Hosted environment excels in this area. Until Hollywood goes to 64-bit AROS this is my main environment. I was worried that all development was on 64-bit now so am very excited and thankful that 32-bit is still being updated. Thanks again!

NathanH

No worries. If there is a need, fixes can always be backported from 64-bit to 32-bit AROS :-)