> Over the years I've come to the conclusion that you must throw your new features into the face and shove good defaults down the throats of users, otherwise these feature are hardly used.
Agreed to an extent. I wouldn't call ignoring the systems colour preferences as "good defaults". But I do agree with your more general point.
> 8bit and 24bit colors are already opt-in and you can configure it. Your TERM and COLORTERM environmental variables determine if a (well behaved) terminal program will use those colors.
Unfortunately it's not that simple. Both the "256color" suffix to $TERM and $COLORTERM env vars report terminal compatibility -- not user preference.
Neither are standards either. It's a push to even call them a de facto standard because they're often not used by either terminal emulators nor application developer alike.
There is another env var that does report user preferences: $COLORFGBG, but that receives even less support than the former two vars.
There's also $NO_COLOR (or is $NOCOLOR?), which seems to be honored a little better. However the problem with this is it turns colour off completely. ie you cannot specify "use 4bit but not 8bit nor 24bit colour".
Like all things terminal detection wise, it's all a big fat steaming mess of inconsistencies. Which is why I think the best default is 3/4bit or no colour at all if $NO_COLOR / not a TTY. And have all the cool 24bit themes as an optional extra.
It's also the case that a lot of modern CLI tools will absolutely ignore your termcap and your $TERM env var and barf out whatever extension to ANSI is used by the currently most popular terminal emulators.
Unless it's a TUI app that I'm going to spend a lot of time in (e.g., Emacs), I do not want it theming itself. I want it to look at my termcap to see if it has ANSI color support, and if it does, then emit ANSI color codes. I use base16-shell to set my terminal colors and I want CLI and TUI apps to respect them.
> Unfortunately it's not that simple. Both the "256color" suffix to $TERM and $COLORTERM env vars report terminal compatibility -- not user preference.
Terminal compatibility that is expected to be used. If you only want the 16 base colors then this is the way to go. This is the most reliable and best supported way to get that in the terminal.
If you want some tools to still output 256 colors, you can change the TERM variable on the fly: "TERM=$TERM-256color infocmp -L".
You also should unset COLORTERM. It's for specifying that this terminal supports 24bit colors.
You’re basically just demonstrating my point that this gets very complicated very quickly and thus the polite way to handle this is tool authors to default to 4bit.
A lot of the complexity comes not only from technology that has been dragged along for almost 50 years now but also that the maintainer of ncurses/terminfo has been stubbornly refusing to add new capabilities to the terminfo library.
Using terminfo will give you at least a standard approach to what defaults to use. If you don't want to configure your terminal then you are asking others to do something you are not willing to do.
If we are supposed to go with the lowest common denominator without terminfo capabilities, we would even lose 4bit colors. Then only the basic 8 colors are safe because there a lot of terminals now that don't support the workaround of "bold plus basic color gives bright color" anymore.
That is a massive over simplification and you're pointing the fingers at the wrong culprits.
I've talked a lot recently about the problems with env vars for terminal detection and the problems with developers not understanding how to correctly query terminal capabilities. Heck, some don't even correctly support non-TTY outputs (ie pipes). And then there's the issue that terminfo is a C library so a lot of languages with average to no-C FFI support cannot use it without fork()ing -- which is slow. Plus terminfo is POSIX only which eliminates platforms like Windows -- which is increasingly a target now that Windows Terminal supports ANSI escape sequences.
It's a mess and blaming that on history and terminfo really misunderstands that the real problem is modern tools not respecting or having compatibility with that history, rather than the history breaking things.
> If we are supposed to go with the lowest common denominator without terminfo capabilities, we would even lose 4bit colors. Then only the basic 8 colors are safe because there a lot of terminals now that don't support the workaround of "bold plus basic color gives bright color" anymore.
You're missing the point. It's not about supporting the lowest common denominator -- though I'd love you to give some example of terminal emulators in common use that don't support 4-bit colours ;)
It's about supporting user defaults. Themes based on 8-bit and 24-bit colour pallets are defined by the developer. Whereas themes based on 3 and 4-bit colour pallets are defined by the terminal user.
Hence why I say applications should default to the terminal defaults (defined by the user)
There's times when applications should support 24-bit regardless of the user preferences. Such as image rendering in the terminal. But those are exceptions rather than the norm. And precisely why things like `256-color` and $COlORTERM need to be defining terminal support not user preference.
> And then there's the issue that terminfo is a C library so a lot of languages with average to no-C FFI support cannot use it without fork()ing -- which is slow. Plus terminfo is POSIX only which eliminates platforms like Windows -- which is increasingly a target now that Windows Terminal supports ANSI escape sequences.
export TERM=ms-terminal
Windows is POSIX. Windows NT always had a POSIX layer.
Terminfo is mostly a file based database collection. You don't need to link to the C library, you just must be able to read the terminal infos.
> You're missing the point. It's not about supporting the lowest common denominator -- though I'd love you to give some example of terminal emulators in common use that don't support 4-bit colours ;)
The question is not about support but about how to get them to be display. Which escape sequences produces colors 8-15? The answer is different from terminal to terminal and that's the problem.
1. That then breaks any applications that look for either `xterm` or `256-colors` in the TERM string. Unfortunately these days $TERM behaves more like a User-Agent string in HTTP-land so most terminal emulators are forced to pretend to be `xterm` even when they're not.
2. `export` is a Bourne Shell-ism. Powershell's syntax looks a little more like:
> Terminfo is mostly a file based database collection. You don't need to link to the C library, you just must be able to read the terminal infos.
mostly is doing a lot of heavy lifting there.
Also reading those databases without C is non-trivial as well. For example try manually parsing the output of /usr/share/terminfo/74/tmux-256color on macOS.
It's also worth noting here that there's two different database formats: terminfo and termcap. And different locations each platform (and users!) might configure those databases to reside.
Also tools like `tput` do more than just look up $TERM against terminfo. There are actually a few different places where terminal capabilities are hidden. Including ANSI escape codes, which was originally used in the hardware terminal days before $TERM existed and are still expected to be supported by modern terminal emulators.
Let's not forget that you also need to check if you're reading from, or writing to, a TTY. It might actually be a pipe, or even just a regular file. And while there is a standard C API for checking if a file descriptor is a TTY, `isatty()`, that itself is just a hack because there's no ABI in TTYs to return true/false if they are a TTY. So instead isatty() has to send ioctl calls to alter the state of the TTY, but a state that's only supported by TTYs (ie incompatible with pipes and regular files). It's a hack that's worked well for decades, but it's yet another class of edge cases that people take for granted.
> The question is not about support but about how to get them to be display. Which escape sequences produces colors 8-15? The answer is different from terminal to terminal and that's the problem.
What differs is the shade of colours that are produced with those escape codes. And that's exactly why I advocate that developers should stick to 3/4-bit code.
The point I'm making is that those colours are defined by the terminal and configurable by the user. Which means individuals can set a colour profile that is readable for them. This is why CLI and TUI developers should default to, and respect, those preferences.
Take a read through my comment history. I'm an author of a multitude of advanced CLI and TUI applications as well as terminal emulator and thus I've discussed these topics at great lengths with other people.
Agreed to an extent. I wouldn't call ignoring the systems colour preferences as "good defaults". But I do agree with your more general point.
> 8bit and 24bit colors are already opt-in and you can configure it. Your TERM and COLORTERM environmental variables determine if a (well behaved) terminal program will use those colors.
Unfortunately it's not that simple. Both the "256color" suffix to $TERM and $COLORTERM env vars report terminal compatibility -- not user preference.
Neither are standards either. It's a push to even call them a de facto standard because they're often not used by either terminal emulators nor application developer alike.
There is another env var that does report user preferences: $COLORFGBG, but that receives even less support than the former two vars.
There's also $NO_COLOR (or is $NOCOLOR?), which seems to be honored a little better. However the problem with this is it turns colour off completely. ie you cannot specify "use 4bit but not 8bit nor 24bit colour".
Like all things terminal detection wise, it's all a big fat steaming mess of inconsistencies. Which is why I think the best default is 3/4bit or no colour at all if $NO_COLOR / not a TTY. And have all the cool 24bit themes as an optional extra.