nvme_scan_topology() failed when /sys/class/nvme or /sys/class/nvme-subsystem didn't exist (i.e., no NVMe devices present), causing nvme_scan() to propagate the error and the Python global_ctx() constructor to return NULL — breaking the python-create-ctrl-object and python-sigsegv-during-gc tests.

Root cause

nvme_scan_ctrls() and nvme_scan_subsystems() wrap scandir(), which returns -1 with errno=ENOENT when the target directory doesn't exist. The prior code unconditionally treated any negative return as fatal:

ctrls.num = nvme_scan_ctrls(&ctrls.ents);
if (ctrls.num < 0) {
    nvme_msg(ctx, LOG_DEBUG, "failed to scan ctrls: %s\n",
             nvme_strerror(-ctrls.num));  // wrong: -(-1) = 1 = EPERM
    return ctrls.num;                    // propagates -1 as fatal error
}

Fix

• Capture errno immediately after each failing scandir call
• Treat ENOENT as zero entries — a missing sysfs directory is a valid "no devices" state, not an error
• Return an error only for unexpected errno values
• Fix the error message to use strerror(err) instead of nvme_strerror(-ctrls.num) which was deriving a nonsensical error code

ctrls.num = nvme_scan_ctrls(&ctrls.ents);
if (ctrls.num < 0) {
    int err = errno;
    if (err != ENOENT) {
        nvme_msg(ctx, LOG_DEBUG, "failed to scan ctrls: %s\n", strerror(err));
        return -err;
    }
    ctrls.num = 0;
}

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.