Keepassx fails to compile on FreeBSD due to libusb implementation differences

[madpilot78]

On FreeBSD the libusb_hotplug_register_callback() function uses a pointer to a struct as an handle.

I'm changing the methods to use this as expected by the system library.

This now compiles fine but fails at runtime.

Not sure why this patch is not working and I must admit I do not understand what is being done using qintptr in

keepassxc/src/gui/osutils/DeviceListener.h

Line 41 in f093291

typedef qintptr Handle;

.

This has been noticed here: https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=277652

The failures with my patch are described here: https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=277652#c10

I do not have a yubikey so I cannot do testing myself.

Thanks in advance for any help!

Testing strategy

As said, with this changes the source compiles on FreeBSD, and the resulting binary works, except when exercising the modified code.

Type of change

• ✅ Bug fix (non-breaking change that fixes an issue)

[phoerious]

Instead of adding all the ifdefs, you should cast libusb_hotplug_callback_handle to qintptr instead. This should already be an opaque pointer type that needs no additional logic.

[madpilot78]

@phoerious Thanks for the feedback!

So, if I understand correctly, there is an existing bug in the code that uses ints where it should be using qintptr. This makes some sense to me, now I can understand why qintptr was used to start with.

I'll do some testing hoping I can make sense of all this.

Thanks again for your suggestions.

[phoerious]

On a normal 64 bit system it shouldn't make a difference, but with smaller word lengths it may.

[madpilot78]

@phoerious Problem is, the code as is fails to compile on FreeBSD. FreeBSD is using recent clang by default, which tends to be less permissive about this kind of casts.

Putting the correct types in the correct place should be enough to make it compile and work then.

[madpilot78]

Hi again,

I experimented a little but made little progress.

I uploaded what I have now, following suggestion to use qintptr.

This fails to compile. If I use classic C style casts in place of static_cast it does compile but fails at runtime.

I'm asking for help on how to fix this. My limited C++ knowledge has already reached its limit.

Thanks in advance.

[droidmonkey]

@phoerious what do you want to do here?

[madpilot78]

For completeness, these bugs have been created regarding FreeBSD:

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=277799

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=279556

But they apply to base system and resolution cannot be fast, most probably nothing can get released before the next major release at the earliest.

[phoerious]

I've pushed a commit that should make it work on both Linux and FreeBSD, though I cannot test the latter. If it works, we should rebase and squash the commits.

Do you know whether any other *BSD will be affected as well? I checked OpenBSD, but they don't seem to have a custom libusb in their tree.

[droidmonkey]

Looks reasonable to me

[madpilot78]

@phoerious Thanks a lot for your patch!

I've compiled and am running keepassxc with it locally, it works fine, but I am unable to test the USB functionality.

I've posted a patch against the ports tree including your patch so others can test on FreeBSD, I hope to have feedback in a few days at most.

[phoerious]

You should be able to test it to some degree even without a YubiKey. Open two tabs and switch between them. You should see a green progress bar appear for a short time below the password field. If that's the case and the app doesn't crash, it most likely works.

[madpilot78]

@phoerious

I performed a quick test you suggest by creating a new DB. Unluckily changing tabs causes a lockup of the program, I am unable to do anything else. I need to kill it with xkill.

[phoerious]

That is strange, though probably not related to this change. It may, however, be a problem with that libusb implementation.