Phil Charles tried to install the new dpkg (dpkg_1.9.17_hurd-i386.deb) but experienced problems with update-rc.d. Marcus Brinkmann noted that Wichert Akkerman had moved update-rc.d from dpkg to sysvinit (which GNU/Hurd does not use). Marcus' solution was to put a copy of update-rc.d into the Hurd package.
Joachim Nilsson was first to note the problems with this: apt-get today first upgraded dpkg to 1.9.17 and
then wanted to upgrade the Hurd package. I got a basic catch-22
situation since the new dpkg wanted update-rc.d in the Hurd, but the
Hurd was not going to be installed.
Joachim suggesed this work around:
Alternatively, Marcus said that it should work OK if the Hurd is upgraded first (using --force-overwrite).
Kapil Hari Paranjape asked if anyone with an i815-based had got X working on GNU/Hurd. The answer was no, because the XFree86 X server needs the agpgart (Linux) kernel module. Marcus Brinkman mused briefly about how to add such (framebuffer) support to Mach/OSKit for the Hurd.
Sounds like a project.
Tristan had problems removing his CD. The usual method (settrans -afg /cdrom, then settrans -afg /dev/hd2) did not work. After getting round to mentioning his gnumach version, Marcus pointed out that he needed the cdromlock patch from Alexey Dejneka (who had fixed OSKit).
Last week Daniel Wagner wrote a program to diagnose his misbehaving
(oskit-mach-based) network. This week Daniel writes: my test program seems to work on the gnumach. On
oskit-mach also. So it is clear, that the receipt of a packet is the
problem. I'm trying to figure out how to write the receive
part.
Packet reception, according to Roland McGrath, is more complex because it is a callback message.
Later, Daniel wrote:
No packets are received because net_bh in oskit-20010214/linux/dev/net_glue.c is never called.
netif_rx (in the same file) sets as a last command the NET_BH flag, in order to inform the interupthandler that net_bh should be called. That's my understanding of the things.
Roland thought that the oskit code that Linux bh maps onto had changed and advised Daniel:
You need to override the osenv_softirq_* functions with a new implementation in oskit-mach/oskit/ that works by calling setsoftclock() and have softclock_oskit check the software interrupt state and call its handlers. For disable/enable softintr you can use splsoftclock/splx similar to osenv_intr_* (in oskit-mach/oskit/osenv_synch.c)--actually nothing in the oskit calls those entry points.
You don't really have to make these osenv callback functions fully general to handle any possible call according to their interface. In oskit code code we are using for oskit-mach, there is only this one called in the linux net driver glue code and so handling what that needs is sufficient.
Marcus Brinkmann said that he needed someone to incorporate the CMU Mach manual pages into a GNU Mach reference manual. Douglas Hanks volunteered.
Gergely Nagy found that he could not open more than 16 aterms and investigated further. The problem was that not enough ptys were being created at install-time by MAKEDEV that matched what getpt was looking for (i.e. only /dev/ptyp[0-f]).
Roland McGrath altered the Hurd's MAKEDEV script to make pty?[0-9a-v] to satisfy getpt. Roland and Thomas Bushnell agreed that it would be a good idea if MAKEDEV is run in postinstall after a Hurd upgrade.