Guard evdi_open_attached_to() against a NULL parent device

[scotttromley]

Problem

evdi_open_attached_to_fixed() documents NULL as a valid argument meaning "open a
generic device", and implements it:

if (sysfs_parent_device == NULL)
        device_index = get_generic_device();

But the deprecated evdi_open_attached_to() wrapper calls strlen() on the pointer
unconditionally:

return evdi_open_attached_to_fixed(sysfs_parent_device, strlen(sysfs_parent_device));

so passing NULL faults before that check can ever run.

This was reported in #582, which was closed as completed with no comments, but the
wrapper was never changed — the fault is still present on master as of 21a6390.

How I hit it

Silicon Motion ships SMIUSBDisplayManager, the userspace driver for their SM768 USB
display adapters, on top of evdi. It calls evdi_open_attached_to(NULL) to claim one
of the generic devices created by initial_device_count, and dies immediately:

SIGSEGV si_code=SEGV_MAPERR si_addr=NULL
#1 evdi_open_attached_to (libevdi.so.1 + 0x4523)    thread: pnpThreadfunc

strace shows the fault landing straight after a successful 64 MB framebuffer mmap
with all four firmware blobs already open, so the caller is otherwise healthy — it is
purely the wrapper.

Because that daemon is a closed-source binary, the workaround suggested in #582
(calling evdi_open_attached_to_fixed(NULL, 0) instead) is not available to affected
users. The wrapper itself has to accept NULL.

Fix

Pass a length of 0 when the pointer is NULL, so the documented behaviour is reachable
through the old entry point as well as the new one. Formatted per the repo's
.clang-format (note the original single-line call was 90 columns and not
clang-format clean).

Testing

• make -C library builds clean.
• evdi_open_attached_to(NULL) now returns a valid handle instead of faulting, and
  evdi_close() on it succeeds. libevdi logs Opened /dev/dri/card3 as slave drm device / Marking /dev/dri/card3 as unused.
• End to end: with this fix SMIUSBDisplayManager starts and stays up, and the
  attached panel comes up as a working fourth display at 2560x1440@60. Tested on
  kernel 7.2.2 (clang + ThinLTO), evdi 1.15.0, Hyprland 0.56.2 / aquamarine 0.14.0.
• No behaviour change for non-NULL callers: same pointer, same length.

🤖 Generated with Claude Code