In lots of places, depending
On virtual terminals and real terminals, the TERM environment variable is set by the program that chains to login, and is inherited all of the way along to the interactive shell that executes once one has logged on. Where, precisely, this happens varies from system to system, and according to the kind of terminal.
real terminals
Real, serial, terminals can vary in type, according to what's at the other end of the wire. So conventionally the getty program is invoked with an argument that specifies the terminal type, or is passed the TERM program from a service manager's service configuration data.
systemd's variability
The serial-getty@.service service unit file, or drop-in files that apply thereto, is where to change the terminal type for real terminals on systemd systems. Note that such a change applies to all terminal login services that employ this service unit template. (To change it for only individual terminals, one has to manually instantiate the template, or add drop-ins that only apply to instantiations.)
systemd has had at least four mechanisms during its lifetime for picking up the value of the TERM environment variable. At the time of first writing this answer, as can be seen, there was an Environment=TERM=something line in the template service unit files. At other times, the types linux and vt102 were hard-wired into the getty and serial-getty service unit files respectively. More recently, the environment variable has been inherited from process #1, which has set it in various ways.
As of 2020, the way that systemd decides what terminal type to specify in a service's TERM environment variable is quite complex, and not documented at all. The way to change it remains a drop-in configuration file with Environment=TERM=something. But where the default value originates from is quite variable. Subject to some fairly complex to explain rules that involve the TTYPath= settings of individual service units, it can be one of three values: a hardwired linux, a hardwired vt220 (no longer vt102), or the value of the TERM environment variable that process #1 inherited, usually from the kernel/bootstrap loader.
(Ironically, the getttyent() mechanism still exists in the GNU C library, and systemd could have re-used the /etc/ttys mechanism.)
kernel virtual terminals
Kernel virtual terminals, as you have noted, have a fixed type. Unlike NetBSD, which can vary the kernel virtual terminal type on the fly, Linux and the other BSDs have a single fixed terminal type implemented in the kernel's built-in terminal emulation program. On Linux, that type matches linux from the terminfo database. (FreeBSD's kernel terminal emulation since version 9 has been teken. Prior to version 9 it was cons25 OpenBSD's is pccon.)
- On systems using
mingetty or vc-get-tty (from the nosh package) the program "knows" that it can only be talking to a virtual terminal, and they hardwire the "known" virtual terminal types appropriate to the operating system that the program was compiled for.
- On systemd systems, one used to be able to see this in the
/usr/lib/systemd/system/getty@.service unit file (/lib/systemd/system/getty@.service on un-merged systems), which read Environment=TERM=linux
setting the TERM variable in the environment passed to agetty.
For kernel virtual terminals, one does not change the terminal type. The terminal emulator program in the kernel doesn't change, after all. It is incorrect to change the type. In particular, this will screw up cursor/editing key CSI sequence recognition. The linux CSI sequences sent by the Linux kernel terminal emulator are different to the xterm or vt100 CSI sequences sent by GUI terminal emulator programs in DEC VT mode. (In fact, they are highly idiosyncratic and non-standard, and different both to all real terminals that I know of, and to pretty much all other software terminal emulators apart from the one built into Linux.)
GUI terminal emulators
Your GUI terminal emulator is one of many programs, from the SSH dæmon to screen, that uses pseudo-terminals. What the terminal type is depends from what terminal emulator program is running on the master side of the pseudo-terminal, and how it is configured. Most GUI terminal emulators will start the program on the slave side with a TERM variable whose value matches their terminal emulation on the master side. Programs like the SSH server will attempt to "pass through" the terminal type that is on the client end of the connection. Usually there is some menu or configuration option to choose amongst terminal emulations.
The gripping hand
The right way to detect colour capability is not to hardwire a list of terminal types in your script. There are an awful lot of terminal types that support colour.
The right way is to look at what termcap/terminfo says about your terminal type.
colour=0
if tput Co > /dev/null 2>&1
then
test "`tput Co`" -gt 2 && colour=1
elif tput colors > /dev/null 2>&1
then
test "`tput colors`" -gt 2 && colour=1
fi
Further reading
- Jonathan de Boyne Pollard (2018).
TERM. nosh Guide. Softwares.
The "emulated consoles" are system-specific, as you understand with your reference to fbdev. Furthermore, they are generally integrated into the kernel, so trying to modify them directly would require modifying the kernel, which is certainly more 'overkill' than putting up with X.
You could use fbdev directly, or write something based on SDL like Thomas Dickey suggests, but as he said, none of the necessary work has been done, and the amount of work that represents should not be underestimated.
But while most of the common desktop environments are fairly heavy-weight, X itself is not (by any remotely modern standards). If you have an app that is primarily text-based, but you need to put up an image every now and then, consider using X without a desktop environment. How? Use startx or a custom script to start the X server and an xterm (or your terminal emulator of choice), full-screen, with no window manager (or find one of the alternative ultra-lightweight WMs). Hide the mouse cursor if you want. Then, from within your text-mode program, you can start and stop lightweight image-displaying X clients at specified locations on the screen, and remove them under program control. You can resize the xterm or just accommodate the fact that part of your text screen will be obscured. Of course, there are an unlimited number of variations on this theme, but you get the idea.
I used to use X on Unix machines with under 4Mb of RAM (not Gb), so it's not necessary for X to be bloated. As a beneficial side-effect, you gain platform portability (you can potentially target anything with an X server) and the ability to run remotely.
Best Answer
I guess you could run your full-screen program in tmux or Screen pane directly, without additional shell session (shell is just another program).
Another way, which I prefer, is to use tiling/stacking window manager like i3 and terminal program urxvt. The latter has very fast daemon/client structure, which allows opening new windows instantly, so you could run any program in new window this way:
This needs to be in a script or a function, really.
New window will take one half, one third, or so on of the screen in default tiling mode. Combined modes are also possible in these WMs.