The program crashes upon startup on certain machines.

[firefly-insky]

Hello, we are encountering an issue on a Windows 10 machine where the program fails to launch, and this problem is 100‑percent reproducible.

1.3.3.1037-20260814-104516-11356.dmp

mimalloc.dll.pdb

[daanx]

I think this is due to free'ing a pointer that wasn't allocated by mimalloc. Can you try to run a debug build and see if you get warnings? (Or compile in release mode with -DMI_FREE_IS_CHECKED=ON and run with MIMALLOC_SHOW_ERRORS=1. )

If it is indeed an invalid pointer, you could try to free with mi_cfree (checked free) ? Or use redirection?

[firefly-insky]

I think this is due to free'ing a pointer that wasn't allocated by mimalloc. Can you try to run a debug build and see if you get warnings? (Or compile in release mode with and run with . )-DMI_FREE_IS_CHECKED=ON``MIMALLOC_SHOW_ERRORS=1

If it is indeed an invalid pointer, you could try to free with (checked free) ? Or use redirection?mi_cfree

Thanks for your reply. The issue only occurs on one of our customer's machines. Due to information security restrictions, we are unable to perform debugging. As a temporary workaround, we have removed mimalloc, and no similar problems have been observed on other devices.

[daanx]

Ah, that is worrisome. How do you use mimalloc in the application? Do you use mimalloc-redirect.dll or is the application using mi_free directly? (through a new/delete override)?

The default for a shared library (mimalloc.dll) should be to have MI_FREE_IS_CHECKED=1 defined at compile time but that is not the case here (from the assembly above). What is the mimalloc version you are using? Checking pointers is a tiny bit slower but it will ignore any invalid pointers that are passed.

[firefly-insky]

Ah, that is worrisome. How do you use mimalloc in the application? Do you use mimalloc-redirect.dll or is the application using mi_free directly? (through a new/delete override)?

The default for a shared library (mimalloc.dll) should be to have MI_FREE_IS_CHECKED=1 defined at compile time but that is not the case here (from the assembly above). What is the mimalloc version you are using? Checking pointers is a tiny bit slower but it will ignore any invalid pointers that are passed.

We are using mimalloc v3.3.0. We override the global new and delete operators by including mimalloc-new-delete.h in main.cpp. Does this relate to the order of header file inclusion?

[firefly-insky]

A comparison with the normal scenario shows that _mi_page_map is significantly smaller.