AmiNetXDuo 0.23.0 (NetX Duo 6.5.1, ThreadX 6.5.1)

AmiNetXDuo 0.23.0 (NetX Duo 6.5.1, ThreadX 6.5.1)

Drop AmiNetXDuo-0.23.0.lha on a Workbench
window and run Install-AmiNetXDuo.

• httpd -C serves this machine's display at /console. This is work in progress, and it serves the chipset's own planar screens only -- a screen on a graphics card is not shown. The browser shows the frontmost screen, so a Preferences editor that opens its own screen appears, and a change of resolution, depth or overscan is carried through. The mouse and keyboard work. A screen where nothing changes costs five bytes a frame; an idle 640x256 four-colour Workbench runs at about 33 frames a second on an A1200 and about four times that on an A3000. There is no password, and anyone who can reach the port has the screen, the keyboard, the mouse and a button that reboots the machine
• The console can save a PNG of the screen, and record what it receives to a file that carries the time each frame was taken and where the pointer was
• PCMCIA cards that this driver refused now attach. The socket was never put into I/O mode before the configuration register was written, so on a write-protected socket the write was swallowed; the chip's reset register was addressed in the wrong Gayle window; a card whose CIS carries no CISTPL_FUNCID was rejected outright; and the configuration register was written at the doubled address first. cnet.device, whose source is public domain, is what each of these was checked against
• anxnet.device no longer holds the PCMCIA slot on a machine where it found no card. Its card.resource handle was queued whether or not the slot could be owned and was never taken back out, so opening the driver once could lock cnet.device and every other PCMCIA driver out until the machine was rebooted
• PCMCIA cards that only answer 16-bit reads on their odd-numbered registers now work. Detection probes for it and switches, so there is no second binary and nothing for you to choose
• A card whose address PROM is blank now takes its station address from the CIS if the card carries one there, and otherwise derives a stable one from this machine rather than refusing. The derived address is locally administered, is the same on every boot, and HARDWAREADDRESS in the interface file still overrides it
• CheckNetDevice says what anxnet.device found: every card it looked at, the step each one reached, and why any was refused. It works when no card came up at all, which is when it is wanted, and it needs no special build
• The cross toolchain is published for macOS as well as Linux, so the tree can be built on either without a container. tools/fetch-toolchain.sh picks the one for your machine and checks it; tools/build-toolchain.sh reproduces it from source, pinned to exact commits rather than branch names
• A 64-bit divide by a value of 2^32 or more no longer takes the machine down. The three 64-bit shift helpers were each compiled into a call to themselves and recursed until the stack was gone; nothing in the shipped libraries divides by that much, so it was reachable rather than reached