From mboxrd@z Thu Jan 1 00:00:00 1970 From: b4@gewt.net (Cory Smelosky) Date: Tue, 21 Feb 2017 08:38:49 -0800 Subject: [TUHS] Mach for i386 / Mt Xinu or other In-Reply-To: <094c861a-e7ef-3e79-9d70-ada7e579d127@aueb.gr> References: <635c06b4-0048-4951-95ca-283c64c30fed@SG2APC01FT017.eop-APC01.prod.protection.outlook.com> <1ba0f584-6478-4332-bcae-63ac6cedf2f6@SG2APC01FT041.eop-APC01.prod.protection.outlook.com> <20170219154432.GA19243@mcvoy.com> <58ab3214.+jRaJEWVki5gYHFz%schily@schily.net> <20170220222457.GB3163@mcvoy.com> <58ac16ca.V0zEZijwK0rh0Cyr%schily@schily.net> <1487684861.1580775.887937752.698B8B60@webmail.messagingengine.com> <58ac5a3a.geKODIzKbcVWDsIS%schily@schily.net> <094c861a-e7ef-3e79-9d70-ada7e579d127@aueb.gr> Message-ID: <58AC6D19.9010100@gewt.net> Diomidis Spinellis wrote: > On 21/02/2017 17:18, Joerg Schilling wrote: >>> Requiring "pass through SCSI" for a CD burner violates the UNIX >>> philosophy that everything is a file (which implies that reading and >>> writing data be implemented, where possible, through the read and write >>> system calls rather than through special interfaces specific to a device >>> type). >> >> You are incorrectly informed: >> >> Writing CDs is a highly complex task. No known kernel is able to do that >> internally. >> >> So the only useful method is to use SCSI pass through. > > This is an interesting point. However, controlling the C/A/T > phototypesetter 20 years before writable CD-ROM appeared, was also a > very complex task, yet it was accomplished through a simple device file. > Controlling a voice modem is also complex, time-critical, and requires > bidirectional communication. Again, voice modems were controlled through > a character device file. How would you handle the different writing methods? Separate device files or an ioctl on a generic device node? (not arguing - genuinely curious about your theory) > > One can envisage a CD-ROM device driver abstracting the commands > required for writing CD-ROMs into a text-based interface made available > though a character device. These precedents demonstrate that the SCSI > pass through was an unneeded architecture-violating shortcut. > > Arguably, the same can also be claimed for the networking system calls. > > Diomidis >