bandwidth is almost always an issue for local terminals; a 4k (3840×2160) terminal at 120fps and 32bpp is 4 gigabytes per second, and according to https://news.ycombinator.com/item?id=35694801, typically with ddr4 ram you get 30 or 40 gigabytes a second
if you send all that video data through the ddr4 bus twice (once writing, once reading) you're using about 20–25 percent of the machine's memory bandwidth, but the linux tty subsystem actually makes several copies and runs many instructions per byte [citation needed] so you can't get anywhere close to that in practice. especially not if you're spending a lot of time repacking bits into six-bit bytes in order to feign compatibility with the vt340, a misguided engineering dead end from 37 years ago that only supported 16 colors and 800×480 anyway
other protocols like x11 and xpra work considerably better, but x11 video reproduction on a local terminal normally uses xshm or its moral equivalent with xvshmputimage or gl. if you're on a local terminal, you might as well use x11; the only benefit to putting the pixels in the same bytestream as text is if the your terminal is connected to your drawing application over a bytestream such as an ssh connection, a tcp connection, an spi connection, or a uart
i think we could do better by designing a bytestream protocol that minimizes copies and pixel format conversions and is therefore within epsilon of the speed of xshm, but sixel isn't it, and neither is x11 without xshm. a bidirectional bytestream protocol could include flow control to also avoid transmitting pixels that won't be visible and increasing latency due to bufferbloat
as for emoji, they don't improve information density; there are only 2666 emoji in unicode 10.0, so at a maximum they convey only 13 bits of information, and they normally occupy roughly the space of two letters like 'n', so you have slightly fewer bits per pixel. people use emoji because they are cute and colorful, not because they are ithkuil. they're not
some technologies, like the unix shell, smalltalk, and tcp/ip, make hard things easy and easy things possible. others, like retrocomputing, code golf, and malbolge,
make easy things hard and hard things impossible.
sixel makes easy things hard and hard things impossible, like the rest of retrocomputing.
that's a fun way to spend my time when i choose to (e.g., in https://asciinema.org/a/390271 i did real-time 3d graphics in a unicode terminal emulator with braille) but it's not how i want my primary user interface to my computer to work
Sixel is defined for two purposes: the first is to allow display of bitmaps. the second is to allow the definition of characters glyphs (which xterm does NOT have -- due to graphics limitations). If the glyph definition could be done, your bandwidth requirement would be reduced. Back in the 80's we generated custom fonts for PostScript (the Apple laser printer PostScript would cache the glyphs). As needed, the printer would request glyps and these would be supplied by an external driver or box (we called this the "Robin Box" for fairly obscure reasons). This system provided pre-press proofing for many customers (which would include publishers like Ziff-Davis). Please note that the speed of the communication was 9600baud (1000 cps). The image of that printer was 300x300 dpi, with some printers doing 600x600. Sixel would have been good... we used ASCII.
I like sixel -- I use xterm which gives me the option. For everything? No. but if I need a graphic, it is easy enough to use. The alternative for most would be to simply generate PostScript, and run GhostScript to view the results (typically on another system). I have NEVER contemplated watching a video with sixel... I could do it, but the player would be a "labour of love" - I would use a common decoder to produce an uncompressed bitmap (scaling and colour reduction) then convert that into frames of sixel... Just to show StarWars on a 340. But, no urge.
The main issue is that sixel offers the feature. No real reason to NOT have it... and it exists.
Try pbmtoln03, ppmtosixel, imagemagick convert and lsix
sixel is not a protocol: just an encoding. Your idea conflates the two things.
The purpose of the Robin Box? The publishers typically had hundreds or thousands of typefaces. The PostScript Printer? 15 to 50. Since RAM in the printer was limited, this approach allowed the scanning of the target typeface, conversion to outline form and production of dynamic programs executing in the printer. That are discarded but results cached. (and note that
technology is within a period that, in this case, was bounded by RAM space, and typographic conventions -- all of which changed fairly rapidly).
That approach allowed hundreds of typefaces to be used on a single page! With "standard" PostScript; on a printer with only a megabyte or so of RAM available. Sixel is simply a similar tool. You really can't predict how something like would be used if it were generally available.
yeah, i agree that sixel was a reasonable way to implement downloadable fonts in the 01980s, though one that handled line noise and background process output interleaving poorly
(see supdup's graphics protocol for a better design for integrating pixel graphics into a (still retrocomputing) serial terminal. however, supdup foolishly omits any font-downloading facility)
i just think sixel, whatever its merits for the problems of 40 years ago, is a worse way to display graphics today than things like x11, which is itself no shining gem, just not as bad as sixel. also we're sort of stuck with most of x11 for the foreseeable future because there's a lot of software written for it; let's make sure that doesn't happen with sixel
(you point out that sixel is an encoding, not a protocol, which is sort of true, but it has an associated state machine in the terminal, so it's also sort of a protocol, and that's the part that has the poor error-recovery characteristics)
by 'worse' i mean: it requires more effort to achieve an equivalent result; it provides a worse result with the same effort; and there are some results it simply can't achieve that the alternatives can
if you're looking for a puzzle to solve for fun, of course, those are advantages, not disadvantages; but i repeat myself
it's especially an advantage if you find something that was previously thought to be impossible due to laboring under such artificial limitations but turns out to actually be possible despite them
here are some more nontechnical descriptions of sixel that are more correct than any technical description possibly can be
sixel is oulipo programming; encoding your graphics in sixel is like writing a novel without the letter 'e'. btw, a wonderful article about oulipo programming is https://100r.co/site/working_offgrid_efficiently.html
sixel is an art project, not an engineering project
if you send all that video data through the ddr4 bus twice (once writing, once reading) you're using about 20–25 percent of the machine's memory bandwidth, but the linux tty subsystem actually makes several copies and runs many instructions per byte [citation needed] so you can't get anywhere close to that in practice. especially not if you're spending a lot of time repacking bits into six-bit bytes in order to feign compatibility with the vt340, a misguided engineering dead end from 37 years ago that only supported 16 colors and 800×480 anyway
other protocols like x11 and xpra work considerably better, but x11 video reproduction on a local terminal normally uses xshm or its moral equivalent with xvshmputimage or gl. if you're on a local terminal, you might as well use x11; the only benefit to putting the pixels in the same bytestream as text is if the your terminal is connected to your drawing application over a bytestream such as an ssh connection, a tcp connection, an spi connection, or a uart
i think we could do better by designing a bytestream protocol that minimizes copies and pixel format conversions and is therefore within epsilon of the speed of xshm, but sixel isn't it, and neither is x11 without xshm. a bidirectional bytestream protocol could include flow control to also avoid transmitting pixels that won't be visible and increasing latency due to bufferbloat
as for emoji, they don't improve information density; there are only 2666 emoji in unicode 10.0, so at a maximum they convey only 13 bits of information, and they normally occupy roughly the space of two letters like 'n', so you have slightly fewer bits per pixel. people use emoji because they are cute and colorful, not because they are ithkuil. they're not
some technologies, like the unix shell, smalltalk, and tcp/ip, make hard things easy and easy things possible. others, like retrocomputing, code golf, and malbolge, make easy things hard and hard things impossible. sixel makes easy things hard and hard things impossible, like the rest of retrocomputing.
that's a fun way to spend my time when i choose to (e.g., in https://asciinema.org/a/390271 i did real-time 3d graphics in a unicode terminal emulator with braille) but it's not how i want my primary user interface to my computer to work