On Sun, 17 Aug 2025 at 12:15:00 +1000, James Tocknell wrote:
I think armhf CPUs are still being manufactured [...]
I think their main issue is the amount of RAM and number
and frequency of the cores
A more fundamental issue is the size of the virtual address space, and
that is unsolvable. On a 32-bit architecture, there is no way that any
single process can possibly have access to more than 4 GiB of virtual
memory[1] because its pointers are simply not large enough to address
more, even if the machine physically has more than that (merely adding
RAM or swap won't help) - and for increasingly many packages, the
compiler's working set won't fit in 4 GiB any more. Disabling debug
symbols can kick this can further down the road for a while, at the cost
of making architecture-specific bugs much harder to diagnose and fix.
If the 32-bit ports are going to survive, then the affected packages
either need to be cross-compiled from a 64-bit architecture (which is
not something that Debian has traditionally done and not something that
our autobuilder network really supports), or dropped from the 32-bit
ports (which means also dropping everything that depends or
build-depends on them, recursively).
i386 since trixie is gradually becoming a partial architecture, where
not all packages are expected to be built (for example no kernel), and I
think that tendency will increase over time. armhf might well go through
a similar process.
smcv
[1] in practice less that that, because the kernel needs some of the
address space for its own use
