On 02/17/2017 at 17:59 Leo Famulari writes: > On Fri, Feb 17, 2017 at 01:35:44PM -0500, myglc2 wrote: >> Hi Leo, >> >> I think that what you have is great. With this in my system config ... >> >> (agetty-service (agetty-configuration >> (tty "ttyS1") >> (baud-rate "115200"))) >> >> ... it works painlessly on a headless GuixSD server over IPMI. I think >> you can put a brief example in the doc, refer the user to the code and >> the agetty man page for more info, and declare victory. > > Awesome! > > Veering off-topic, I wonder what created that device node /dev/ttyS1? I > have 32 of them, but (gnu build linux-boot) appears to only create the > first one: > > https://git.savannah.gnu.org/cgit/guix.git/tree/gnu/build/linux-boot.scm#n160
I found the serial line on my server by doing ... g1@g1 ~$ dmesg | grep tty [ 0.000000] console [tty0] enabled [ 2.192386] 00:07: ttyS1 at I/O 0x2f8 (irq = 3, base_baud = 115200) is a 16550A ... but linux-boot.scm only specifies ttyS0 ... 160 ;; Serial line. 161 (mknod (scope "dev/ttyS0") 'char-special #o660 162 (device-number 4 64)) ... but gnu/packages/linux-libre-4.9-i686.conf has ... CONFIG_SND_SERIAL_U16550=m ... so maybe the kernel probes for and finds the UART. >> But let me digress a bit on this topic. What if, in situations like >> this, Guix provided an easy way to export the "native config" generated >> by Guix? >> >> Then we could tell the user ... >> >> 1) If you want to know exactly what we are doing, export the native >> config and read the native doc. >> >> 2) If you want features we don't support, export the native config, read >> the doc, modify it, and feed it into the "native config hatch." >> >> With this approach, we could implement and document only "key features" >> with a clear conscience. When we haven't implemented a feature the user >> needs they will be no worse off that they were before. In fact, they >> will usually be ahead, because Guix has taken care of the general >> requirements and provided a sound starting point for a native config. > > An interesting idea! I guess the implementation would vary based on > services that use a configuration file (like nginx) and those that are > configured on the command-line (like agetty). But I'm still finding my > way around services, in general, so I'm not sure. Agreed. Ideally there would be a strategy and reusable modules that supports either approach, or a mix. Then maybe this would lighten the effort required for each service and help regularize these interfaces.
