A seemingly simple question "is this computer running on mains power" turns
out to be really complex to answer.  Here's a simplified overview.

An electrical connection doesn't imply a power sensor, for this you need
appropriate circuitry.  Thus, most desktops and servers don't have any sense
of where their power comes from.  Here, assuming this means AC is quite
reasonable -- but on_ac_power doesn't make this decision for you, so you
know you should avoid displaying an "AC plug" icon, etc.

A typical laptop has one mains plug and one battery.  As running out of
battery causes data loss, all laptops have appropriate sensors (but such
sensors might still lack supported drivers).

High-end servers have multiple mains connectors.  Some laptops have multiple
batteries.  Phones and tablets may have different charging modes ("trickle",
normal, fast).

But in some cases, all hell goes loose.  For example, Allwinner A64 SoC
knows about three sources: AC, USB, battery, yet particular boards using it
don't match those.  Pine64 for one has "ac" on an USB type B plug, a
"battery" that watches a battery attachment, USB OTG type A that can power
the board yet lacks PMIC circuitry to signal that it's connected (thus "usb"
will always return online=0), and can be also powered via GPIO pins on Euler
and Exp connectors, which don't tell the kernel anything at all.  At this
time, mainline kernel lacking proper drivers to tell us this doesn't feel
that bad...


Available interfaces:

Modern
======

In /sys/class/power_supply/ you get one subdirectory per power source, with
fairly straightforward data.  The main thing to look at is "online" which
tells us whether that source is connected.  Type "type" can be one of:

        "Unknown", "Battery", "UPS", "Mains", "USB",
        "USB_DCP", "USB_CDP", "USB_ACA", "USB_C",
        "USB_PD", "USB_PD_DRP", "BrickID"

"Mains" can also mean any other known but non-specific connector.  It seems
that usually USB describes itself this way.

The difference between those many USB* variants seems to be irrelevant for
our purposes.  I have yet to see any that would provide any data about where
the power actually comes from.  Thus, we can't tell if we're on an USB
charger vs a powerbank.  But that's no different from AC coming from a
disconnected UPS...

"BrickID" is an Apple specific charger that I don't know anything about.

The vast majority of UPSes don't show up as "UPS" which stands only for some
specific type of GPIO control connector (which I don't know).  Usually, we'd
need to ask an userspace daemon like nut.

"Wireless" is not reported by any driver in the mainline kernel, but is used
by some out-of-tree devices.


Legacy
======

ACPI: /proc/acpi/.  Not interesting for us as all(?) drivers also provide a
modern sysfs interface.

APM: /proc/apm.  Present only on some very old 32-bit x86 machines; the
kernel issues real mode calls to the BIOS.  In Debian, i386 kernels still
ship APM modules, although to enable them you need to manually specify
"acpi=off apm=on" on the kernel's command line.

PMU: /proc/pmu/, specific to Apple Powerbooks.  The driver does _not_
provide data in /sys/.


Userspace
=========

Most UPSes are not recognized by the kernel, and talk to an userspace
program such as nut.  on_ac_power doesn't currently query nut nor any
other daemon of this kind.

Note that the UPS provides its stored electricity as AC or DC that will
appear to the kernel as mains power.  Thus, the UPS' answer needs to
override sysfs data.