I remember that we had discussions about this when building the current
OpenWrt ToH, which ended in having the "Unsupported" field, where things can
be put that are not working. This information is more valuable than a
classification "limited support", where you anyway need to state somewhere
what limited means for this device.
"limited support" is possibly the wrong term, because it can be confused with
"partial support" where some parts fo the device are not usable by openwrt.
In the continum that I listed above, limited means that the standard image
distributed by LEDE will install and run, but that you are going to have severe
trouble installing anything after that (although with the imagebuilder approach,
you can change what's in the base image and probably install some other stuff)
This could be because opkg hits OOM, or because there are not enough flash
blocks left free to be usable. The exact definition can be discussed and
hammered out. It doesn't really matter which limit you run into, just that you
have run into a limit.
But it means the same thing for all devices, it's not something that varies in
meaning by device.
The idea is to identify devices that are still able to be installed with the
standard release image (including LUCI if that's in the what we define as the
standard release image) but that you can't expect to add anything else in
addition.
The devices are perfectly usable, just maxed out to start with.
The minimal support category is for the devices that you can no longer fit the
standard release packages onto. They can still be used with custom images, but
are very much in the 'experts only' category.
based on the discussions that I've been seeing, the criteria for hitting
'limited support' are
RAM: 32M
This is because as I understand things, opkg won't run reliably with 32M, but
is comfortable in 64M
FLASH: 4 flash blocks available (effectivly 1 flash block of storage after
overhead)
This is not X MB of flash because the amount of flash left after compiling
things is going to vary from device to device (instruction set density, flash
block size and what kernel modules are needed to support the hardware are three
significant things off the bat).
The analysis last night on the difference between having 5 flash blocks
available and having 4 is what I'm relying on. Per that discussion, 5 flash
blocks is still fairly usable, but at 4 you can still store configs on the
system, but not any noticable packages.
It may be that we want to set the limit a little higher, or that there is a way
to reduce the number of flash blocks that the filesystem keeps idle, so let's
discuss if this is the case or not.
Ideally, the build process would be able to generate a warning at the end of the
build if the particular image created is that tight. (and it may be that it's
tight/unusable in some filesystem types but not in others).
David Lang