On 6/15/20 7:32 AM, Matthias Schiffer wrote:
I just think of ar71xx and ath79, where we have the same device but
different targets. Of course, the name won't be exactly equal, as ath79 will
have e.g. tplink_ prefix and ar71xx won't.
Isn't ar71xx removed from master builds? It's neither at snapshot
https://downloads.openwrt.org/snapshots/targets/ nor planed to be re-
added to any upcoming release, is it?
Yes, but it's just an example for a similar situation which might arise in the 
future. Then, we even might not have the current situation with the different 
device names, but may end up with completely identical names except for the 
target.

For bcm63xx, we have two subtargets that build the same devices.
If we look at PR#2957, we might have a now bmips target at some point
that features the same devices as bcm63xx.
Can you please explain why that's the case? Why do we offer different
images for the same device? I understand that for ar71xx -> ath79 within a
I don't have any idea why this situation at bcm63xx exists; I just got aware of 
it at some point. Maybe Noltari or KanjiMonster can help ...

transfer period but it's never the scope to offer different "flavors"
long term, is it?
This won't necessarily break anything, as images will still be in different
folders (at least in /bin).
I would be at least confusing and reverts the "unique profile name" idea.
However, we couldn't tell the difference between ar71xx/ath79 or similar
from the image name (easily) after this change, or whether it's generic or
smp for bcm63xx. For my personal taste, this drawback is bigger that the gain
we will get from removing the target/subtarget part.
Again, this sounds like a undesirable state where we not only build but also
maintain multiple images for the save device. Wouldn't it be possible to add
the target to all those "legacy images", however remove it wherever a target
uses device tree and images.mk aka has long term support?
Well, just look at the situation in 19.07. There we have both ar71xx and ath79 
for the same devices, and even if we wanted, it would actually be quite hard to 
really filter out the ath79 devices in ar71xx. I really don't think removing 
the target from image names will pay out in the future.

So, unless there is overwhelming support, I tend to NAK this.
:(
A compromise could be found by just removing the subtarget, but keeping the 
target in file names. This would mostly solve your problem with the generic 
names (at least there would be less duplicate info), but there would be 
significantly less situations where this was an impediment. Normally, no 
duplicate devices in a target exist, and if they are moved between subtargets, 
they are actually moved and not copied. The only remaining problem I can think 
of at the moment would be the bcm63xx situation, and maybe that one can be 
resolved at low cost.

Best

Adrian

What about x86-{generic,legacy,64,...}? These subtargets each define a
device just called "generic", with the image names only distinguished by
their subtarget name.
That's why I created this PR earlier, I should have created a patchset instead of splitting it on the ML and GitHub...
https://github.com/openwrt/openwrt/pull/3082

_______________________________________________
openwrt-devel mailing list
openwrt-devel@lists.openwrt.org
https://lists.openwrt.org/mailman/listinfo/openwrt-devel

Reply via email to