4 ms·
As long as it builds as a module, uses only the standard filesystem apis and doesn't have code changes outside that, code review is far less important IMO...
by londons_explore 6y ago
As long as it builds as a module, uses only the standard filesystem apis and doesn't have code changes outside that, code review is far less important IMO...
- huhtenberg 6y agoAt the risk of stating the obvious, this is an exceptionally short-sighted approach. Committing poorly understood code is not that different from using binary blobs.
- lumost 6y agoThe linux kernel is successful partly because by and large what's in the kernel works and doesn't generally break on new revisions. Accepting a new driver which may in fact not work, breaks in ambiguous ways, interacts with other components poorly, or otherwise generates headaches for the kernel could be more trouble than it's worth. Ultimately this driver will eventually need to be modified by others, and ensuing it doesn't get a reputation as a nightmare right out of the gate is also worthwhile.
- aronpye 6y agoSo you’d willingly allow tens of thousands of lines of code which you don’t understand or really know what it does into a kernel which is used by literally billions of devices?
- cratermoon 6y ago> uses only the standard filesystem apis and doesn't have code changes outside that That's literally why having to review 27K lines of code is a pain. You're not seriously going to trust, sight unseen, that there's nothing in the code that slips in a backdoor or catastrophic data-loss bug, are you?
- segmondy 6y agoI wish you weren't getting downvoted, but what you said is absolutely correct. If it's a module and separate from core, who cares. If it's modifying core kernel or mixed up with kernel space code, then nope.
- wtallis 6y agoLinux isn't a microkernel; a filesystem accepted into the kernel source tree will run as kernel space code whether or not it's compiled as a loadable module. So all 27k lines need to be audited for security purposes and to ensure they're interfacing with the rest of the kernel only in the approved ways, because there aren't a lot of technological barriers to the filesystem misbehaving. But more important than that is the maintenance burden. NTFS will be around for a long time, but it's also a moving target because Microsoft hasn't replaced it yet. Kernel developers have to keep in mind how this code will look in a few decades, after all the original developers are retired. If it's written in a very different style from other Linux filesystems, will there be anyone left who both knows enough about the workings of the Linux IO stack 20 years hence, and understands Paragon's code conventions?