develop an algorithm for issuing reps or even rectangular reps

[dankamongmen]

See #1740 for background information. It occurs to me that recognizing linear or even rectangular regions suitable for mass glyph blitting (the latter can be achieved with DECFRA, where supported). The problem is recognizing worthwhile regions of repetition. Note that the styling and coloring must be consistent throughout the region.

There are two basic approaches I can see:

• in each cellrender struct, keep two bools about whether the cells to the right and below are equal to the current cell. the last cell in a row always has false for the former, and all cells of the bottom row have false for the latter. this bool is set during rendering, whenever the last of the two relevant cells is solved. so whenever you solve a cell, you look to the left -- is it solved? if so, are we equal? if so, set that cell's bool high. then look to your right -- is it solved? if so, are we equal? if so, set our own cell's bool high. repeat for the cells above and below. go to step 3.
• when rasterizing, look forward and down to find equality. go to step 3.
• 3: if profitable, issue a multicell update. beating regular glyphs requires at least 5 in a row. beating spaces requires at least 4 in a row.

i think we should forgo the rectangular idea for several reasons:

• it complicates our decision. what if we have 50x of a glyph, and underneath it we have 10x, then 30x spaces, then 10x. ought we issue 2x rectangles and 1x linear? or 3x linear? we could of course solve this.
• the DECFRA escape is much more heavyweight than the rep, requiring 5 parameters
• DECFRA is less widely-supported (no Kitty, for instance), whereas rep is widespread
• rep leaves the cursor right where we want it, DECFRA somewhere else
• DECFRA requires us to carry state across lines about any rectangular regions we need skip
• DECFRA doesn't appear to work on arbitrary glyphs, whereas rep does
• DECFRA doesn't annihilate sixel (on the other hand, DECFRA provides us a ~12B means to update a single glyph underneath a sixel without redrawing the sixel)

now, as to whether we do it in render or raster...i think raster makes more sense. everything's running along natural cache-packed structure, it doesn't require any additional state...

[dankamongmen]

is #1408 a duplicate of this? pretty sure it is.

as noted there, rep seems to work with arbitrary utf-8, which could be profitable:

[schwarzgerat](0) $ tput rep E 8
8888888[schwarzgerat](0) $ tput rep ὂ 8
ὂὂὂὂὂὂὂ[schwarzgerat](0) $ tput rep ▅ 8
▅▅▅▅▅▅▅[schwarzgerat](0) $