render.c

Go to the documentation of this file.

#include <ctype.h>
#include <limits.h>
#include <unistd.h>
#include "internal.h"
#include "unixsig.h"
 
sig_atomic_t sigcont_seen_for_render = 0;
 
// update for a new visual area of |rows|x|cols|, neither of which may be zero.
// copies that area of the lastframe (damage map) which is shared between the
// two. new areas are initialized to empty, just like a new plane. lost areas
// have their egcpool entries purged.
static nccell*
restripe_lastframe(notcurses* nc, unsigned rows, unsigned cols){
  assert(rows);
  assert(cols);
  const size_t size = sizeof(*nc->lastframe) * (rows * cols);
  nccell* tmp = malloc(size);
  if(tmp == NULL){
    return NULL;
  }
  size_t copycols = nc->lfdimx > cols ? cols : nc->lfdimx;
  size_t maxlinecopy = sizeof(nccell) * copycols;
  size_t minlineset = sizeof(nccell) * cols - maxlinecopy;
  unsigned zorch = nc->lfdimx > cols ? nc->lfdimx - cols : 0;
  for(unsigned y = 0 ; y < rows ; ++y){
    if(y < nc->lfdimy){
      if(maxlinecopy){
        memcpy(&tmp[cols * y], &nc->lastframe[nc->lfdimx * y], maxlinecopy);
      }
      if(minlineset){
        memset(&tmp[cols * y + copycols], 0, minlineset);
      }
      // excise any egcpool entries from the right of the new plane area
      if(zorch){
        for(unsigned x = copycols ; x < copycols + zorch ; ++x){
          pool_release(&nc->pool, &nc->lastframe[fbcellidx(y, nc->lfdimx, x)]);
        }
      }
    }else{
      memset(&tmp[cols * y], 0, sizeof(nccell) * cols);
    }
  }
  // excise any egcpool entries from below the new plane area
  for(unsigned y = rows ; y < nc->lfdimy ; ++y){
    for(unsigned x = 0 ; x < nc->lfdimx ; ++x){
      pool_release(&nc->pool, &nc->lastframe[fbcellidx(y, nc->lfdimx, x)]);
    }
  }
  free(nc->lastframe);
  nc->lastframe = tmp;
  nc->lfdimy = rows;
  nc->lfdimx = cols;
  return 0;
}
 
// Check whether the terminal geometry has changed, and if so, copy what can
// be copied from the old lastframe. Assumes that the screen is always anchored
// at the same origin. Initiates a resize cascade for the pile containing |pp|.
// The current terminal geometry, changed or not, is written to |rows|/|cols|.
static int
notcurses_resize_internal(ncplane* pp, unsigned* restrict rows, unsigned* restrict cols){
  notcurses* n = ncplane_notcurses(pp);
  unsigned r, c;
  if(rows == NULL){
    rows = &r;
  }
  if(cols == NULL){
    cols = &c;
  }
  ncpile* pile = ncplane_pile(pp);
  unsigned oldrows = pile->dimy;
  unsigned oldcols = pile->dimx;
  *rows = oldrows;
  *cols = oldcols;
  unsigned cgeo_changed;
  unsigned pgeo_changed;
  if(update_term_dimensions(rows, cols, &n->tcache, n->margin_b,
                            &cgeo_changed, &pgeo_changed)){
    return -1;
  }
  n->stats.s.cell_geo_changes += cgeo_changed;
  n->stats.s.pixel_geo_changes += pgeo_changed;
  *rows -= n->margin_t + n->margin_b;
  if(*rows <= 0){
    *rows = 1;
  }
  *cols -= n->margin_l + n->margin_r;
  if(*cols <= 0){
    *cols = 1;
  }
  if(*rows != n->lfdimy || *cols != n->lfdimx){
    if(restripe_lastframe(n, *rows, *cols)){
      return -1;
    }
  }
//fprintf(stderr, "r: %d or: %d c: %d oc: %d\n", *rows, oldrows, *cols, oldcols);
  if(*rows == oldrows && *cols == oldcols){
    return 0; // no change
  }
  pile->dimy = *rows;
  pile->dimx = *cols;
  int ret = 0;
//notcurses_debug(n, stderr);
  // if this pile contains the standard plane, it ought be resized to match
  // the viewing area before invoking any other resize callbacks.
  if(ncplane_pile(notcurses_stdplane(n)) == pile){
    ncplane_resize_maximize(notcurses_stdplane(n));
  }
  // now, begin a resize callback cascade on the root planes of the pile.
  for(ncplane* rootn = pile->roots ; rootn ; rootn = rootn->bnext){
    if(rootn->resizecb){
      ret |= rootn->resizecb(rootn);
    }
  }
  return ret;
}
 
// Check for a window resize on the standard pile.
static int
notcurses_resize(notcurses* n, unsigned* restrict rows, unsigned* restrict cols){
  pthread_mutex_lock(&n->pilelock);
  int ret = notcurses_resize_internal(notcurses_stdplane(n), rows, cols);
  pthread_mutex_unlock(&n->pilelock);
  return ret;
}
 
void nccell_release(ncplane* n, nccell* c){
  pool_release(&n->pool, c);
}
 
// Duplicate one cell onto another when they share a plane. Convenience wrapper.
int nccell_duplicate(ncplane* n, nccell* targ, const nccell* c){
  if(cell_duplicate_far(&n->pool, targ, n, c) < 0){
    logerror("failed duplicating cell");
    return -1;
  }
  return 0;
}
 
// Emit fchannel with RGB changed to contrast effectively against bchannel.
static uint32_t
highcontrast(const tinfo* ti, uint32_t bchannel){
  unsigned r, g, b;
  if(ncchannel_default_p(bchannel)){
    // FIXME what if we couldn't identify the background color?
    r = ncchannel_r(ti->bg_collides_default);
    g = ncchannel_g(ti->bg_collides_default);
    b = ncchannel_b(ti->bg_collides_default);
  }else{
    // FIXME need to handle palette-indexed
    r = ncchannel_r(bchannel);
    g = ncchannel_g(bchannel);
    b = ncchannel_b(bchannel);
  }
  uint32_t conrgb = 0;
  if(r + g + b < 320){
    ncchannel_set(&conrgb, 0xffffff);
  }else{
    ncchannel_set(&conrgb, 0);
  }
  return conrgb;
}
 
// wants coordinates within the sprixel, not absolute
// FIXME if plane is not wholly on-screen, probably need to toss plane,
// at least for this rendering cycle
static void
paint_sprixel(ncplane* p, struct crender* rvec, int starty, int startx,
              int offy, int offx, int dstleny, int dstlenx){
  const notcurses* nc = ncplane_notcurses_const(p);
  sprixel* s = p->sprite;
  int dimy = s->dimy;
  int dimx = s->dimx;
//fprintf(stderr, "START: %d/%d DIM: %d/%d dim(p): %d/%d off: %d/%d DSTLEN %d/%d %p\n", starty, startx, dimy, dimx, ncplane_dim_y(p), ncplane_dim_x(p), offy, offx, dstleny, dstlenx, rvec);
  if(s->invalidated == SPRIXEL_HIDE){ // no need to do work if we're killing it
    return;
  }
  for(int y = starty ; y < dimy ; ++y){
    const int absy = y + offy;
    // once we've passed the physical screen's bottom, we're done
    if(absy >= dstleny || absy < 0){
      break;
    }
    for(int x = startx ; x < dimx ; ++x){ // iteration for each cell
      const int absx = x + offx;
      if(absx >= dstlenx || absx < 0){
        break;
      }
      sprixcell_e state = sprixel_state(s, absy, absx);
      struct crender* crender = &rvec[fbcellidx(absy, dstlenx, absx)];
//fprintf(stderr, "presprixel: %p preid: %d state: %d\n", rvec->sprixel, rvec->sprixel ? rvec->sprixel->id : 0, s->invalidated);
      // if we already have a glyph solved (meaning said glyph is above this
      // sprixel), and we run into a bitmap cell, we need to null that cell out
      // of the bitmap.
      if(crender->p || crender->s.bgblends){
        // if sprite_wipe_cell() fails, we presumably do not have the
        // ability to wipe, and must reprint the character
        if(sprite_wipe(nc, p->sprite, y, x) < 0){
//fprintf(stderr, "damaging due to wipe [%s] %d/%d\n", nccell_extended_gcluster(crender->p, &crender->c), absy, absx);
          crender->s.damaged = 1;
        }
        crender->s.p_beats_sprixel = 1;
      }else if(!crender->p && !crender->s.bgblends){
        // if we are a bitmap, and above a cell that has changed (and
        // will thus be printed), we'll need redraw the sprixel.
        if(crender->sprixel == NULL){
          crender->sprixel = s;
        }
        if(state == SPRIXCELL_ANNIHILATED || state == SPRIXCELL_ANNIHILATED_TRANS){
//fprintf(stderr, "REBUILDING AT %d/%d\n", y, x);
          sprite_rebuild(nc, s, y, x);
//fprintf(stderr, "damaging due to rebuild [%s] %d/%d\n", nccell_extended_gcluster(crender->p, &crender->c), absy, absx);
        }
      }
    }
  }
}
 
// Paints a single ncplane 'p' into the provided scratch framebuffer 'fb' (we
// can't always write directly into lastframe, because we need build state to
// solve certain cells, and need compare their solved result to the last frame).
//
//  dstleny: leny of target rendering area described by rvec
//  dstlenx: lenx of target rendering area described by rvec
//  dstabsy: absy of target rendering area (relative to terminal)
//  dstabsx: absx of target rendering area (relative to terminal)
//
// only those cells where 'p' intersects with the target rendering area are
// rendered.
//
// the sprixelstack orders sprixels of the plane (so we needn't keep them
// ordered between renders). each time we meet a sprixel, extract it from
// the pile's sprixel list, and update the sprixelstack.
__attribute__ ((nonnull (1, 2, 7))) static void
paint(ncplane* p, struct crender* rvec, int dstleny, int dstlenx,
      int dstabsy, int dstabsx, sprixel** sprixelstack,
      unsigned pgeo_changed){
  unsigned y, x, dimy, dimx;
  int offy, offx;
  ncplane_dim_yx(p, &dimy, &dimx);
  offy = p->absy - dstabsy;
  offx = p->absx - dstabsx;
//fprintf(stderr, "PLANE %p %d %d %d %d %d %d %p\n", p, dimy, dimx, offy, offx, dstleny, dstlenx, p->sprite);
  // skip content above or to the left of the physical screen
  unsigned starty, startx;
  if(offy < 0){
    starty = -offy;
  }else{
    starty = 0;
  }
  if(offx < 0){
    startx = -offx;
  }else{
    startx = 0;
  }
  // if we're a sprixel, we must not register ourselves as the active
  // glyph, but we *do* need to null out any cellregions that we've
  // scribbled upon.
  if(p->sprite){
    if(pgeo_changed){
      // do what on failure? FIXME
      sprixel_rescale(p->sprite, ncplane_pile(p)->cellpxy, ncplane_pile(p)->cellpxx);
    }
    paint_sprixel(p, rvec, starty, startx, offy, offx, dstleny, dstlenx);
    // decouple from the pile's sixel list
    if(p->sprite->next){
      p->sprite->next->prev = p->sprite->prev;
    }
    if(p->sprite->prev){
      p->sprite->prev->next = p->sprite->next;
    }else{
      ncplane_pile(p)->sprixelcache = p->sprite->next;
    }
    // stick on the head of the running list: top sprixel is at end
    if(*sprixelstack){
      (*sprixelstack)->prev = p->sprite;
    }
    p->sprite->next = *sprixelstack;
    p->sprite->prev = NULL;
    *sprixelstack = p->sprite;
    return;
  }
  for(y = starty ; y < dimy ; ++y){
    const int absy = y + offy;
    // once we've passed the physical screen's bottom, we're done
    if(absy >= dstleny || absy < 0){
      break;
    }
    for(x = startx ; x < dimx ; ++x){ // iteration for each cell
      const int absx = x + offx;
      if(absx >= dstlenx || absx < 0){
        break;
      }
      struct crender* crender = &rvec[fbcellidx(absy, dstlenx, absx)];
//fprintf(stderr, "p: %p damaged: %u %d/%d\n", p, crender->s.damaged, y, x);
      nccell* targc = &crender->c;
      if(nccell_wide_right_p(targc)){
        continue;
      }
 
      if(nccell_fg_alpha(targc) > NCALPHA_OPAQUE){
        const nccell* vis = &p->fb[nfbcellidx(p, y, x)];
        if(nccell_fg_default_p(vis)){
          vis = &p->basecell;
        }
        if(nccell_fg_alpha(vis) == NCALPHA_HIGHCONTRAST){
          crender->s.highcontrast = true;
          crender->s.hcfgblends = crender->s.fgblends;
          crender->hcfg = cell_fchannel(targc);
        }
        unsigned fgblends = crender->s.fgblends;
        cell_blend_fchannel(ncplane_notcurses(p), targc, cell_fchannel(vis), &fgblends);
        crender->s.fgblends = fgblends;
        // crender->highcontrast can only be true if we just set it, since we're
        // about to set targc opaque based on crender->highcontrast (and this
        // entire stanza is conditional on targc not being NCALPHA_OPAQUE).
        if(crender->s.highcontrast){
          nccell_set_fg_alpha(targc, NCALPHA_OPAQUE);
        }
      }
 
      // Background color takes effect independently of whether we have a
      // glyph. If we've already locked in the background, it has no effect.
      // If it's transparent, it has no effect. Otherwise, update the
      // background channel and balpha.
      // Evaluate the background first, in case we have HIGHCONTRAST fg text.
      if(nccell_bg_alpha(targc) > NCALPHA_OPAQUE){
        const nccell* vis = &p->fb[nfbcellidx(p, y, x)];
        // to be on the blitter stacking path, we need
        //  1) crender->s.blittedquads to be non-zero (we're below semigraphics)
        //  2) cell_blittedquadrants(vis) to be non-zero (we're semigraphics)
        //  3) somewhere crender is 0, blittedquads is 1 (we're visible)
        if(!crender->s.blittedquads || !((~crender->s.blittedquads) & cell_blittedquadrants(vis))){
          if(nccell_bg_default_p(vis)){
            vis = &p->basecell;
          }
          unsigned bgblends = crender->s.bgblends;
          cell_blend_bchannel(ncplane_notcurses(p), targc, cell_bchannel(vis), &bgblends);
          crender->s.bgblends = bgblends;
        }else{ // use the local foreground; we're stacking blittings
          if(nccell_fg_default_p(vis)){
            vis = &p->basecell;
          }
          unsigned bgblends = crender->s.bgblends;
          cell_blend_bchannel(ncplane_notcurses(p), targc, cell_fchannel(vis), &bgblends);
          crender->s.bgblends = bgblends;
          crender->s.blittedquads = 0;
        }
      }
 
      // if we never loaded any content into the cell (or obliterated it by
      // writing in a zero), use the plane's base cell.
      // if we have no character in this cell, we continue to look for a
      // character, but our foreground color will still be used unless it's
      // been set to transparent. if that foreground color is transparent, we
      // still use a character we find here, but its color will come entirely
      // from cells underneath us.
      if(!crender->p){
        const nccell* vis = &p->fb[nfbcellidx(p, y, x)];
        if(vis->gcluster == 0 && !nccell_double_wide_p(vis)){
          vis = &p->basecell;
        }
        // if the following is true, we're a real glyph, and not the right-hand
        // side of a wide glyph (nor the null codepoint).
        if( (targc->gcluster = vis->gcluster) ){ // index copy only
          if(crender->sprixel && crender->sprixel->invalidated == SPRIXEL_HIDE){
//fprintf(stderr, "damaged due to hide %d/%d\n", y, x);
            crender->s.damaged = 1;
          }
          crender->s.blittedquads = cell_blittedquadrants(vis);
          // we can't plop down a wide glyph if the next cell is beyond the
          // screen, nor if we're bisected by a higher plane.
          if(nccell_double_wide_p(vis)){
            // are we on the last column of the real screen? if so, 0x20 us
            if(absx >= dstlenx - 1){
              targc->gcluster = htole(' ');
              targc->width = 1;
            // is the next cell occupied? if so, 0x20 us
            }else if(crender[1].c.gcluster){
//fprintf(stderr, "NULLING out %d/%d (%d/%d) due to %u\n", y, x, absy, absx, crender[1].c.gcluster);
              targc->gcluster = htole(' ');
              targc->width = 1;
            }else{
              targc->stylemask = vis->stylemask;
              targc->width = vis->width;
            }
          }else{
            targc->stylemask = vis->stylemask;
            targc->width = vis->width;
          }
          crender->p = p;
        }else if(nccell_wide_right_p(vis)){
          crender->p = p;
          targc->width = 0;
        }
      }
    }
  }
}
 
// it's not a pure memset(), because NCALPHA_OPAQUE is the zero value, and
// we need NCALPHA_TRANSPARENT
static inline void
init_rvec(struct crender* rvec, int totalcells){
  struct crender c = {0};
  nccell_set_fg_alpha(&c.c, NCALPHA_TRANSPARENT);
  nccell_set_bg_alpha(&c.c, NCALPHA_TRANSPARENT);
  for(int t = 0 ; t < totalcells ; ++t){
    memcpy(&rvec[t], &c, sizeof(c));
  }
}
 
// adjust an otherwise locked-in cell if highcontrast has been requested. this
// should be done at the end of rendering the cell, so that contrast is solved
// against the real background.
static inline void
lock_in_highcontrast(notcurses* nc, const tinfo* ti, nccell* targc, struct crender* crender){
  if(nccell_fg_alpha(targc) == NCALPHA_TRANSPARENT){
    nccell_set_fg_default(targc);
  }
  if(nccell_bg_alpha(targc) == NCALPHA_TRANSPARENT){
    nccell_set_bg_default(targc);
  }
  if(crender->s.highcontrast){
    // highcontrast weighs the original at 1/4 and the contrast at 3/4
    if(!nccell_fg_default_p(targc)){
      unsigned fgblends = 3;
      uint32_t fchan = cell_fchannel(targc);
      uint32_t bchan = cell_bchannel(targc);
      uint32_t hchan = channels_blend(nc, highcontrast(ti, bchan), fchan,
                                      &fgblends, nc->tcache.fg_default);
      cell_set_fchannel(targc, hchan);
      fgblends = crender->s.hcfgblends;
      hchan = channels_blend(nc, hchan, crender->hcfg, &fgblends,
                             nc->tcache.fg_default);
      cell_set_fchannel(targc, hchan);
    }else{
      nccell_set_fg_rgb(targc, highcontrast(ti, cell_bchannel(targc)));
    }
  }
}
 
// Postpaint a single cell (multiple if it is a multicolumn EGC). This means
// checking for and locking in high-contrast, checking for damage, and updating
// 'lastframe' for any cells which are damaged.
static inline void
postpaint_cell(notcurses* nc, const tinfo* ti, nccell* lastframe, unsigned dimx,
               struct crender* crender, egcpool* pool, unsigned y, unsigned* x){
  nccell* targc = &crender->c;
  lock_in_highcontrast(nc, ti, targc, crender);
  nccell* prevcell = &lastframe[fbcellidx(y, dimx, *x)];
  if(cellcmp_and_dupfar(pool, prevcell, crender->p, targc) > 0){
//fprintf(stderr, "damaging due to cmp [%s] %d %d\n", nccell_extended_gcluster(crender->p, &crender->c), y, *x);
    if(crender->sprixel){
      sprixcell_e state = sprixel_state(crender->sprixel, y, *x);
//fprintf(stderr, "state under candidate sprixel: %d %d/%d\n", state, y, *x);
      // we don't need to change it when under an opaque cell, because
      // that's always printed on top.
      if(!crender->s.p_beats_sprixel){
        if(state != SPRIXCELL_OPAQUE_SIXEL){
//fprintf(stderr, "damaged due to opaque %d/%d\n", y, *x);
          crender->s.damaged = 1;
        }
      }
    }else{
//fprintf(stderr, "damaged due to lastframe disagree %d/%d\n", y, *x);
      crender->s.damaged = 1;
    }
    assert(!nccell_wide_right_p(targc));
    const int width = targc->width;
    for(int i = 1 ; i < width ; ++i){
      const ncplane* tmpp = crender->p;
      ++crender;
      crender->p = tmpp;
      ++*x;
      ++prevcell;
      targc = &crender->c;
      targc->gcluster = 0;
      targc->channels = crender[-i].c.channels;
      targc->stylemask = crender[-i].c.stylemask;
      if(cellcmp_and_dupfar(pool, prevcell, crender->p, targc) > 0){
//fprintf(stderr, "damaging due to cmp2 %d/%d\n", y, *x);
        crender->s.damaged = 1;
      }
    }
  }
}
 
 
// iterate over the rendered frame, adjusting the foreground colors for any
// cells marked NCALPHA_HIGHCONTRAST, and clearing any cell covered by a
// wide glyph to its left.
//
// FIXME this cannot be performed at render time (we don't yet know the
//       lastframe, and thus can't compute damage), but we *could* unite it
//       with rasterization--factor out the single cell iteration...
// FIXME can we not do the blend a single time here, if we track sums in
//       paint()? tried this before and didn't get a win...
static void
postpaint(notcurses* nc, const tinfo* ti, nccell* lastframe,
          unsigned dimy, unsigned dimx, struct crender* rvec, egcpool* pool){
//fprintf(stderr, "POSTPAINT BEGINS! %zu %p %d/%d\n", sizeof(*rvec), rvec, dimy, dimx);
  for(unsigned y = 0 ; y < dimy ; ++y){
    for(unsigned x = 0 ; x < dimx ; ++x){
      struct crender* crender = &rvec[fbcellidx(y, dimx, x)];
      postpaint_cell(nc, ti, lastframe, dimx, crender, pool, y, &x);
    }
  }
}
 
// merging one plane down onto another is basically just performing a render
// using only these two planes, with the result written to the lower plane.
int ncplane_mergedown(ncplane* restrict src, ncplane* restrict dst,
                      int begsrcy, int begsrcx, unsigned leny, unsigned lenx,
                      int dsty, int dstx){
//fprintf(stderr, "Merging down %d/%d @ %d/%d to %d/%d\n", leny, lenx, begsrcy, begsrcx, dsty, dstx);
  if(dsty < 0){
    if(dsty != -1){
      logerror("invalid dsty %d", dsty);
      return -1;
    }
    dsty = dst->y;
  }
  if(dstx < 0){
    if(dstx != -1){
      logerror("invalid dstx %d", dstx);
      return -1;
    }
    dstx = dst->x;
  }
  if((unsigned)dsty >= dst->leny || (unsigned)dstx >= dst->lenx){
    logerror("dest origin %u/%u ≥ dest dimensions %d/%d",
             dsty, dstx, dst->leny, dst->lenx);
    return -1;
  }
  if(begsrcy < 0){
    if(begsrcy != -1){
      logerror("invalid begsrcy %d", begsrcy);
      return -1;
    }
    begsrcy = src->y;
  }
  if(begsrcx < 0){
    if(begsrcx != -1){
      logerror("invalid begsrcx %d", begsrcx);
      return -1;
    }
    begsrcx = src->x;
  }
  if((unsigned)begsrcy >= src->leny || (unsigned)begsrcx >= src->lenx){
    logerror("source origin %u/%u ≥ source dimensions %d/%d",
             begsrcy, begsrcx, src->leny, src->lenx);
    return -1;
  }
  if(leny == 0){
    if((leny = src->leny - begsrcy) == 0){
      logerror("source area was zero height");
      return -1;
    }
  }
  if(lenx == 0){
    if((lenx = src->lenx - begsrcx) == 0){
      logerror("source area was zero width");
      return -1;
    }
  }
  if(dst->leny - leny < (unsigned)dsty || dst->lenx - lenx < (unsigned)dstx){
    logerror("dest len %u/%u ≥ dest dimensions %d/%d",
             leny, lenx, dst->leny, dst->lenx);
    return -1;
  }
  if(src->leny - leny < (unsigned)begsrcy || src->lenx - lenx < (unsigned)begsrcx){
    logerror("source len %u/%u ≥ source dimensions %d/%d",
             leny, lenx, src->leny, src->lenx);
    return -1;
  }
  if(src->sprite || dst->sprite){
    logerror("can't merge sprixel planes");
    return -1;
  }
  const int totalcells = dst->leny * dst->lenx;
  nccell* rendfb = calloc(totalcells, sizeof(*rendfb));
  const size_t crenderlen = sizeof(struct crender) * totalcells;
  struct crender* rvec = malloc(crenderlen);
  if(!rendfb || !rvec){
    logerror("error allocating render state for %ux%u", leny, lenx);
    free(rendfb);
    free(rvec);
    return -1;
  }
  init_rvec(rvec, totalcells);
  sprixel* s = NULL;
  paint(src, rvec, dst->leny, dst->lenx, dst->absy, dst->absx, &s, 0);
  assert(NULL == s);
  paint(dst, rvec, dst->leny, dst->lenx, dst->absy, dst->absx, &s, 0);
  assert(NULL == s);
//fprintf(stderr, "Postpaint start (%dx%d)\n", dst->leny, dst->lenx);
  const struct tinfo* ti = &ncplane_notcurses_const(dst)->tcache;
  postpaint(ncplane_notcurses(dst), ti, rendfb, dst->leny, dst->lenx, rvec, &dst->pool);
//fprintf(stderr, "Postpaint done (%dx%d)\n", dst->leny, dst->lenx);
  free(dst->fb);
  dst->fb = rendfb;
  free(rvec);
  return 0;
}
 
int ncplane_mergedown_simple(ncplane* restrict src, ncplane* restrict dst){
  return ncplane_mergedown(src, dst, 0, 0, 0, 0, 0, 0);
}
 
// write the nccell's UTF-8 extended grapheme cluster to the provided fbuf.
static inline int
term_putc(fbuf* f, const egcpool* e, const nccell* c){
  if(cell_simple_p(c)){
//fprintf(stderr, "[%.4s] %08x\n", (const char*)&c->gcluster, c->gcluster); }
    // we must not have any 'cntrl' characters at this point, except for
    // nil or newline
    if(c->gcluster == 0 || c->gcluster == '\n'){
      if(fbuf_putc(f, ' ') < 0){
        return -1;
      }
    }else if(fbuf_puts(f, (const char*)&c->gcluster) == EOF){
      return -1;
    }
  }else{
    if(fbuf_puts(f, egcpool_extended_gcluster(e, c)) == EOF){
      return -1;
    }
  }
  return 0;
}
 
// write any escape sequences necessary to set the desired style
static inline int
term_setstyles(fbuf* f, notcurses* nc, const nccell* c){
  unsigned normalized = false;
  int ret = coerce_styles(f, &nc->tcache, &nc->rstate.curattr,
                          nccell_styles(c), &normalized);
  if(normalized){
    nc->rstate.fgdefelidable = true;
    nc->rstate.bgdefelidable = true;
    nc->rstate.bgelidable = false;
    nc->rstate.fgelidable = false;
    nc->rstate.bgpalelidable = false;
    nc->rstate.fgpalelidable = false;
  }
  return ret;
}
 
// u8->str lookup table used in term_esc_rgb below
static const char* const NUMBERS[] = {
"0;", "1;", "2;", "3;", "4;", "5;", "6;", "7;", "8;", "9;", "10;", "11;", "12;", "13;", "14;", "15;", "16;",
"17;", "18;", "19;", "20;", "21;", "22;", "23;", "24;", "25;", "26;", "27;", "28;", "29;", "30;", "31;", "32;",
"33;", "34;", "35;", "36;", "37;", "38;", "39;", "40;", "41;", "42;", "43;", "44;", "45;", "46;", "47;", "48;",
"49;", "50;", "51;", "52;", "53;", "54;", "55;", "56;", "57;", "58;", "59;", "60;", "61;", "62;", "63;", "64;",
"65;", "66;", "67;", "68;", "69;", "70;", "71;", "72;", "73;", "74;", "75;", "76;", "77;", "78;", "79;", "80;",
"81;", "82;", "83;", "84;", "85;", "86;", "87;", "88;", "89;", "90;", "91;", "92;", "93;", "94;", "95;", "96;",
"97;", "98;", "99;", "100;", "101;", "102;", "103;", "104;", "105;", "106;", "107;", "108;", "109;", "110;", "111;", "112;",
"113;", "114;", "115;", "116;", "117;", "118;", "119;", "120;", "121;", "122;", "123;", "124;", "125;", "126;", "127;", "128;",
"129;", "130;", "131;", "132;", "133;", "134;", "135;", "136;", "137;", "138;", "139;", "140;", "141;", "142;", "143;", "144;",
"145;", "146;", "147;", "148;", "149;", "150;", "151;", "152;", "153;", "154;", "155;", "156;", "157;", "158;", "159;", "160;",
"161;", "162;", "163;", "164;", "165;", "166;", "167;", "168;", "169;", "170;", "171;", "172;", "173;", "174;", "175;", "176;",
"177;", "178;", "179;", "180;", "181;", "182;", "183;", "184;", "185;", "186;", "187;", "188;", "189;", "190;", "191;", "192;",
"193;", "194;", "195;", "196;", "197;", "198;", "199;", "200;", "201;", "202;", "203;", "204;", "205;", "206;", "207;", "208;",
"209;", "210;", "211;", "212;", "213;", "214;", "215;", "216;", "217;", "218;", "219;", "220;", "221;", "222;", "223;", "224;",
"225;", "226;", "227;", "228;", "229;", "230;", "231;", "232;", "233;", "234;", "235;", "236;", "237;", "238;", "239;", "240;",
"241;", "242;", "243;", "244;", "245;", "246;", "247;", "248;", "249;", "250;", "251;", "252;", "253;", "254;", "255;", };
 
static inline int
term_esc_rgb(fbuf* f, bool foreground, unsigned r, unsigned g, unsigned b){
  // The correct way to do this is using tiparm+tputs, but doing so (at least
  // as of terminfo 6.1.20191019) both emits ~3% more bytes for a run of 'rgb'
  // and gives rise to some inaccurate colors (possibly due to special handling
  // of values < 256; I'm not at this time sure). So we just cons up our own.
  /*if(esc == 4){
    return fbuf_emit(f, "setab", tiparm(nc->setab, (int)((r << 16u) | (g << 8u) | b)));
  }else if(esc == 3){
    return fbuf_emit(f, "setaf", tiparm(nc->setaf, (int)((r << 16u) | (g << 8u) | b)));
  }else{
    return -1;
  }*/
  #define RGBESC1 "\x1b" "["
  // we'd like to use the proper ITU T.416 colon syntax i.e. "8:2::", but it is
  // not supported by several terminal emulators :/.
  #define RGBESC2 "8;2;"
  // fprintf() was sitting atop our profiles, so we put the effort into a fast solution
  // here. assemble a buffer using constants and a lookup table.
  fbuf_putn(f, RGBESC1, strlen(RGBESC1));
  fbuf_putc(f, foreground ? '3' : '4');
  fbuf_putn(f, RGBESC2, strlen(RGBESC2));
  const char* s = NUMBERS[r];
  while(*s){
    fbuf_putc(f, *s++);
  }
  s = NUMBERS[g];
  while(*s){
    fbuf_putc(f, *s++);
  }
  s = NUMBERS[b];
  while(*s != ';'){
    fbuf_putc(f, *s++);
  }
  fbuf_putc(f, 'm');
  return 0;
}
 
static inline int
term_bg_rgb8(const tinfo* ti, fbuf* f, unsigned r, unsigned g, unsigned b){
  // We typically want to use tputs() and tiperm() to acquire and write the
  // escapes, as these take into account terminal-specific delays, padding,
  // etc. For the case of DirectColor, there is no suitable terminfo entry, but
  // we're also in that case working with hopefully more robust terminals.
  // If it doesn't work, eh, it doesn't work. Fuck the world; save yourself.
  if(ti->caps.rgb){
    if((ti->bg_collides_default & 0xff000000) == 0x01000000){
      if((r == ncchannel_r(ti->bg_collides_default)) &&
         (g == ncchannel_g(ti->bg_collides_default)) &&
         (b == ncchannel_b(ti->bg_collides_default))){
        // the human eye has fewer blue cones than red or green. toggle
        // the last bit in the blue component to avoid a collision.
        b ^= 0x00000001;
      }
    }
    return term_esc_rgb(f, false, r, g, b);
  }else{
    const char* setab = get_escape(ti, ESCAPE_SETAB);
    if(setab){
      // For 256-color indexed mode, start constructing a palette based off
      // the inputs *if we can change the palette*. If more than 256 are used on
      // a single screen, start... combining close ones? For 8-color mode, simple
      // interpolation. I have no idea what to do for 88 colors. FIXME
      if(ti->caps.colors >= 256){
        return fbuf_emit(f, tiparm(setab, rgb_quantize_256(r, g, b)));
      }else if(ti->caps.colors >= 8){
        return fbuf_emit(f, tiparm(setab, rgb_quantize_8(r, g, b)));
      }
    }
  }
  return 0;
}
 
int term_fg_rgb8(const tinfo* ti, fbuf* f, unsigned r, unsigned g, unsigned b){
  // We typically want to use tputs() and tiperm() to acquire and write the
  // escapes, as these take into account terminal-specific delays, padding,
  // etc. For the case of DirectColor, there is no suitable terminfo entry, but
  // we're also in that case working with hopefully more robust terminals.
  // If it doesn't work, eh, it doesn't work. Fuck the world; save yourself.
  if(ti->caps.rgb){
    return term_esc_rgb(f, true, r, g, b);
  }else{
    const char* setaf = get_escape(ti, ESCAPE_SETAF);
    if(setaf){
      // For 256-color indexed mode, start constructing a palette based off
      // the inputs *if we can change the palette*. If more than 256 are used on
      // a single screen, start... combining close ones? For 8-color mode, simple
      // interpolation. I have no idea what to do for 88 colors. FIXME
      if(ti->caps.colors >= 256){
        return fbuf_emit(f, tiparm(setaf, rgb_quantize_256(r, g, b)));
      }else if(ti->caps.colors >= 8){
        return fbuf_emit(f, tiparm(setaf, rgb_quantize_8(r, g, b)));
      }
    }
  }
  return 0;
}
 
static inline int
update_palette(notcurses* nc, fbuf* f){
  if(nc->tcache.caps.can_change_colors){
    const char* initc = get_escape(&nc->tcache, ESCAPE_INITC);
    for(size_t damageidx = 0 ; damageidx < sizeof(nc->palette.chans) / sizeof(*nc->palette.chans) ; ++damageidx){
      unsigned r, g, b;
      if(nc->palette_damage[damageidx]){
        nc->touched_palette = true;
        ncchannel_rgb8(nc->palette.chans[damageidx], &r, &g, &b);
        // Need convert RGB values [0..256) to [0..1000], ugh
        // FIXME need handle HSL case also
        r = r * 1000 / 255;
        g = g * 1000 / 255;
        b = b * 1000 / 255;
        if(fbuf_emit(f, tiparm(initc, damageidx, r, g, b)) < 0){
          return -1;
        }
        nc->palette_damage[damageidx] = false;
      }
    }
  }
  return 0;
}
 
// at least one of the foreground and background are the default. emit the
// necessary return to default (if one is necessary), and update rstate.
static inline int
raster_defaults(notcurses* nc, bool fgdef, bool bgdef, fbuf* f){
  const char* op = get_escape(&nc->tcache, ESCAPE_OP);
  if(op == NULL){ // if we don't have op, we don't have fgop/bgop
    return 0;
  }
  const char* fgop = get_escape(&nc->tcache, ESCAPE_FGOP);
  const char* bgop = get_escape(&nc->tcache, ESCAPE_BGOP);
  bool mustsetfg = fgdef && !nc->rstate.fgdefelidable;
  bool mustsetbg = bgdef && !nc->rstate.bgdefelidable;
  if(!mustsetfg && !mustsetbg){ // needn't emit anything
    ++nc->stats.s.defaultelisions;
    return 0;
  }else if((mustsetfg && mustsetbg) || !fgop || !bgop){
    if(fbuf_emit(f, op)){
      return -1;
    }
    nc->rstate.fgdefelidable = true;
    nc->rstate.bgdefelidable = true;
    nc->rstate.fgelidable = false;
    nc->rstate.bgelidable = false;
    nc->rstate.fgpalelidable = false;
    nc->rstate.bgpalelidable = false;
  }else if(mustsetfg){ // if we reach here, we must have fgop
    if(fbuf_emit(f, fgop)){
      return -1;
    }
    nc->rstate.fgdefelidable = true;
    nc->rstate.fgelidable = false;
    nc->rstate.fgpalelidable = false;
  }else{ // mustsetbg and !mustsetfg and bgop != NULL
    if(fbuf_emit(f, bgop)){
      return -1;
    }
    nc->rstate.bgdefelidable = true;
    nc->rstate.bgelidable = false;
    nc->rstate.bgpalelidable = false;
  }
  ++nc->stats.s.defaultemissions;
  return 0;
}
 
// these are unlikely, so we leave it uninlined
static int
emit_fg_palindex(notcurses* nc, fbuf* f, const nccell* srccell){
  unsigned palfg = nccell_fg_palindex(srccell);
  // we overload lastr for the palette index; both are 8 bits
  if(nc->rstate.fgpalelidable && nc->rstate.lastr == palfg){
    ++nc->stats.s.fgelisions;
  }else{
    if(term_fg_palindex(nc, f, palfg)){
      return -1;
    }
    ++nc->stats.s.fgemissions;
    nc->rstate.fgpalelidable = true;
  }
  nc->rstate.lastr = palfg;
  nc->rstate.fgdefelidable = false;
  nc->rstate.fgelidable = false;
  return 0;
}
 
static int
emit_bg_palindex(notcurses* nc, fbuf* f, const nccell* srccell){
  unsigned palbg = nccell_bg_palindex(srccell);
  if(nc->rstate.bgpalelidable && nc->rstate.lastbr == palbg){
    ++nc->stats.s.bgelisions;
  }else{
    if(term_bg_palindex(nc, f, palbg)){
      return -1;
    }
    ++nc->stats.s.bgemissions;
    nc->rstate.bgpalelidable = true;
  }
  nc->rstate.lastr = palbg;
  nc->rstate.bgdefelidable = false;
  nc->rstate.bgelidable = false;
  return 0;
}
 
// this first phase of sprixel rasterization is responsible for:
//  1) invalidating all QUIESCENT sprixels if the pile has changed (because
//      it would have been destroyed when switching away from our pile).
//      for the same reason, invalidate all MOVE sprixels in this case.
//  2) damaging all cells under a HIDE sixel, so text phase 1 consumes it
//      (not necessary for kitty graphics)
//  3) damaging uncovered cells under a MOVE (not necessary for kitty)
//  4) drawing invalidated sixels and loading invalidated kitty graphics
//      (new kitty graphics are *not* yet made visible)
// by the end of this pass, all sixels are *complete*. all kitty graphics
// are loaded, but old kitty graphics remain visible, and new/updated kitty
// graphics are not yet visible, and they have not moved.
static int64_t
clean_sprixels(notcurses* nc, ncpile* p, fbuf* f, int scrolls){
  sprixel* s;
  sprixel** parent = &p->sprixelcache;
  int64_t bytesemitted = 0;
  while( (s = *parent) ){
    loginfo("phase 1 sprixel %u state %d loc %d/%d", s->id,
            s->invalidated, s->n ? s->n->absy : -1, s->n ? s->n->absx : -1);
    if(s->invalidated == SPRIXEL_QUIESCENT){
      if(p != nc->last_pile){
        s->invalidated = SPRIXEL_UNSEEN;
      }
    }else if(s->invalidated == SPRIXEL_HIDE){
//fprintf(stderr, "OUGHT HIDE %d [%dx%d] %p\n", s->id, s->dimy, s->dimx, s);
      int r = sprite_scrub(nc, p, s);
      if(r < 0){
        return -1;
      }else if(r > 0){
        if( (*parent = s->next) ){
          s->next->prev = s->prev;
        }
        sprixel_free(s);
        // need to avoid the rest of the iteration, as s is dead
        continue; // don't account as an elision
      }
    }
    if(s->invalidated == SPRIXEL_INVALIDATED && nc->tcache.pixel_refresh){
      nc->tcache.pixel_refresh(p, s);
    }else if(s->invalidated == SPRIXEL_MOVED ||
             s->invalidated == SPRIXEL_UNSEEN ||
             s->invalidated == SPRIXEL_INVALIDATED){
//fprintf(stderr, "1 MOVING BITMAP %d STATE %d AT %d/%d for %p\n", s->id, s->invalidated, y + nc->margin_t, x + nc->margin_l, s->n);
      if(s->invalidated == SPRIXEL_MOVED){
        if(p != nc->last_pile){
          s->invalidated = SPRIXEL_UNSEEN;
        }else{
          if(s->n->absx == s->movedfromx){
            if(s->movedfromy - s->n->absy == scrolls){
              s->invalidated = SPRIXEL_INVALIDATED;
              continue;
            }
          }
        }
      }
      // otherwise it's a new pile, so we couldn't have been on-screen
      int r = sprite_redraw(nc, p, s, f, nc->margin_t, nc->margin_l);
      if(r < 0){
        return -1;
      }
      bytesemitted += r;
      ++nc->stats.s.sprixelemissions;
    }else{
      ++nc->stats.s.sprixelelisions;
    }
    parent = &s->next;
//fprintf(stderr, "SPRIXEL STATE: %d\n", s->invalidated);
  }
  return bytesemitted;
}
 
// scroll the lastframe data |rows| up, to reflect scrolling reality.
// FIXME we could virtualize this as we do for scrolling planes; this
// method involves a lot of unnecessary copying.
static void
scroll_lastframe(notcurses* nc, unsigned rows){
  // the top |rows| rows need be released (though not more than the actual
  // number of rows!)
  if(rows > nc->lfdimy){
    rows = nc->lfdimy;
  }
  for(unsigned targy = 0 ; targy < rows ; ++targy){
    for(unsigned targx = 0 ; targx < nc->lfdimx ; ++targx){
      const size_t damageidx = targy * nc->lfdimx + targx;
      nccell* c = &nc->lastframe[damageidx];
      pool_release(&nc->pool, c);
    }
  }
  // now for all rows subsequent, up through lfdimy - rows, move them back.
  // if we scrolled all rows, we will not move anything (and we just
  // released everything).
  for(unsigned targy = 0 ; targy < nc->lfdimy - rows ; ++targy){
    const size_t dstidx = targy * nc->lfdimx;
    nccell* dst = &nc->lastframe[dstidx];
    const size_t srcidx = dstidx + rows * nc->lfdimx;
    const nccell* src = &nc->lastframe[srcidx];
    memcpy(dst, src, sizeof(*dst) * nc->lfdimx);
  }
  // now for the last |rows| rows, initialize them to 0.
  unsigned targy = nc->lfdimy - rows;
  while(targy < nc->lfdimy){
    const size_t dstidx = targy * nc->lfdimx;
    nccell* dst = &nc->lastframe[dstidx];
    memset(dst, 0, sizeof(*dst) * nc->lfdimx);
    ++targy;
  }
}
 
// "%d tardies to work off, by far the most in the class!\n", p->scrolls
static int
rasterize_scrolls(const ncpile* p, fbuf* f){
  int scrolls = p->scrolls;
  if(scrolls == 0){
    return 0;
  }
  logdebug("order-%d scroll", scrolls);
  /*if(p->nc->rstate.logendy >= 0){
    p->nc->rstate.logendy -= scrolls;
    if(p->nc->rstate.logendy < 0){
      p->nc->rstate.logendy = 0;
      p->nc->rstate.logendx = 0;
    }
  }*/
  // FIXME probably need this to take place at the end of cycle...
  if(p->nc->tcache.pixel_scroll){
    p->nc->tcache.pixel_scroll(p, &p->nc->tcache, scrolls);
  }
  if(goto_location(p->nc, f, p->dimy, 0, NULL)){
    return -1;
  }
  // terminals advertising 'bce' will scroll in the current background color;
  // switch back to the default explicitly.
  if(p->nc->tcache.bce){
    if(raster_defaults(p->nc, false, true, f)){
      return -1;
    }
  }
  return emit_scrolls_track(p->nc, scrolls, f);
}
 
// second sprixel pass in rasterization. by this time, all sixels are handled
// (and in the QUIESCENT state); only persistent kitty graphics still require
// operation. responsibilities of this second pass include:
//
// 1) if we're a different pile, issue the kitty universal clear
// 2) first, hide all sprixels in the HIDE state
// 3) then, make allo LOADED sprixels visible
//
// don't account for sprixelemissions here, as they were already counted.
static int64_t
rasterize_sprixels(notcurses* nc, ncpile* p, fbuf* f){
  int64_t bytesemitted = 0;
  sprixel* s;
  sprixel** parent = &p->sprixelcache;
  while( (s = *parent) ){
//fprintf(stderr, "raster YARR HARR HARR SPIRXLE %u STATE %d\n", s->id, s->invalidated);
    if(s->invalidated == SPRIXEL_INVALIDATED){
//fprintf(stderr, "3 DRAWING BITMAP %d STATE %d AT %d/%d for %p\n", s->id, s->invalidated, nc->margin_t, nc->margin_l, s->n);
      int r = sprite_draw(&nc->tcache, p, s, f, nc->margin_t, nc->margin_l);
      if(r < 0){
        return -1;
      }else if(r > 0){
        bytesemitted += r;
        nc->rstate.y = -1;
        nc->rstate.x = -1;
        ++nc->stats.s.sprixelemissions;
      }
    }else if(s->invalidated == SPRIXEL_LOADED){
      if(nc->tcache.pixel_commit){
        int y, x;
        ncplane_abs_yx(s->n, &y, &x);
        if(goto_location(nc, f, y + nc->margin_t, x + nc->margin_l, NULL)){
          return -1;
        }
        if(sprite_commit(&nc->tcache, f, s, false)){
          return -1;
        }
        nc->rstate.y = -1;
        nc->rstate.x = -1;
      }
    }else if(s->invalidated == SPRIXEL_HIDE){
      if(nc->tcache.pixel_remove){
        if(nc->tcache.pixel_remove(s->id, f) < 0){
          return -1;
        }
        if( (*parent = s->next) ){
          s->next->prev = s->prev;
        }
        sprixel_free(s);
        continue;
      }
    }
    parent = &s->next;
  }
  return bytesemitted;
}
 
// bitmap backends which don't use the bytestream (currently only fbcon)
// need go at the very end, following writeout. pass again, invoking
// pixel_draw_late if defined.
static int64_t
rasterize_sprixels_post(notcurses* nc, ncpile* p){
  if(!nc->tcache.pixel_draw_late){
    return 0;
  }
  int64_t bytesemitted = 0;
  sprixel* s;
  sprixel** parent = &p->sprixelcache;
  while( (s = *parent) ){
//fprintf(stderr, "YARR HARR HARR SPIRXLE %u STATE %d\n", s->id, s->invalidated);
    if(s->invalidated == SPRIXEL_INVALIDATED || s->invalidated == SPRIXEL_UNSEEN){
      int offy, offx;
      ncplane_abs_yx(s->n, &offy, &offx);
//fprintf(stderr, "5 DRAWING BITMAP %d STATE %d AT %d/%d for %p\n", s->id, s->invalidated, nc->margin_t + offy, nc->margin_l + offx, s->n);
      int r = nc->tcache.pixel_draw_late(&nc->tcache, s, nc->margin_t + offy, nc->margin_l + offx);
      if(r < 0){
        return -1;
      }
      bytesemitted += r;
    }
    parent = &s->next;
  }
  return bytesemitted;
}
 
// Producing the frame requires three steps:
//  * render -- build up a flat framebuffer from a set of ncplanes
//  * rasterize -- build up a UTF-8/ASCII stream of escapes and EGCs
//  * refresh -- write the stream to the emulator
 
// Takes a rendered frame (a flat framebuffer, where each cell has the desired
// EGC, attribute, and channels), which has been written to nc->lastframe, and
// spits out an optimal sequence of terminal-appropriate escapes and EGCs. There
// should be an rvec entry for each cell, but only the 'damaged' field is used.
// lastframe has *not yet been written to the screen*, i.e. it's only about to
// *become* the last frame rasterized.
static int
rasterize_core(notcurses* nc, const ncpile* p, fbuf* f, unsigned phase){
  struct crender* rvec = p->crender;
  // we only need to emit a coordinate if it was damaged. the damagemap is a
  // bit per coordinate, one per struct crender.
  for(unsigned y = nc->margin_t; y < p->dimy + nc->margin_t ; ++y){
    const int innery = y - nc->margin_t;
    bool saw_linefeed = 0;
    for(unsigned x = nc->margin_l ; x < p->dimx + nc->margin_l ; ++x){
      const int innerx = x - nc->margin_l;
      const size_t damageidx = innery * nc->lfdimx + innerx;
      unsigned r, g, b, br, bg, bb;
      nccell* srccell = &nc->lastframe[damageidx];
      if(!rvec[damageidx].s.damaged){
        // no need to emit a cell; what we rendered appears to already be
        // here. no updates are performed to elision state nor lastframe.
        ++nc->stats.s.cellelisions;
        if(nccell_wide_left_p(srccell)){
          ++x;
        }
      }else if(phase != 0 || !rvec[damageidx].s.p_beats_sprixel){
//fprintf(stderr, "phase %u damaged at %d/%d %d\n", phase, innery, innerx, x);
        // in the first text phase, we draw only those glyphs where the glyph
        // was not above a sprixel (and the cell is damaged). in the second
        // phase, we draw everything that remains damaged.
        ++nc->stats.s.cellemissions;
        if(goto_location(nc, f, y, x, rvec[damageidx].p)){
          return -1;
        }
        // set the style. this can change the color back to the default; if it
        // does, we need update our elision possibilities.
        if(term_setstyles(f, nc, srccell)){
          return -1;
        }
        // if our cell has a default foreground *or* background, we can elide
        // the default set iff one of:
        //  * we are a partial glyph, and the previous was default on both, or
        //  * we are a no-foreground glyph, and the previous was default background, or
        //  * we are a no-background glyph, and the previous was default foreground
        bool nobackground = nccell_nobackground_p(srccell);
        bool rgbequal = nccell_rgbequal_p(srccell);
        if((nccell_fg_default_p(srccell)) || (!nobackground && nccell_bg_default_p(srccell))){
          if(raster_defaults(nc, nccell_fg_default_p(srccell),
                             !nobackground && nccell_bg_default_p(srccell), f)){
            return -1;
          }
        }
        if(nccell_fg_palindex_p(srccell)){ // palette-indexed foreground
          if(emit_fg_palindex(nc, f, srccell)){
            return -1;
          }
        }else if(!nccell_fg_default_p(srccell)){ // rgb foreground
          // if our cell has a non-default foreground, we can elide the
          // non-default foreground set iff either:
          //  * the previous was non-default, and matches what we have now, or
          //  * we are a no-foreground glyph (iswspace() is true) FIXME
          nccell_fg_rgb8(srccell, &r, &g, &b);
          if(nc->rstate.fgelidable && nc->rstate.lastr == r && nc->rstate.lastg == g && nc->rstate.lastb == b){
            ++nc->stats.s.fgelisions;
          }else{
            if(!rgbequal){ // if rgbequal, no need to set fg
              if(term_fg_rgb8(&nc->tcache, f, r, g, b)){
                return -1;
              }
              ++nc->stats.s.fgemissions;
              nc->rstate.fgelidable = true;
            }else{
              r = nc->rstate.lastr; g = nc->rstate.lastg; b = nc->rstate.lastb;
            }
          }
          nc->rstate.lastr = r; nc->rstate.lastg = g; nc->rstate.lastb = b;
          nc->rstate.fgdefelidable = false;
          nc->rstate.fgpalelidable = false;
        }
        if(nobackground){
          ++nc->stats.s.bgelisions;
        }else if(nccell_bg_palindex_p(srccell)){ // palette-indexed background
          if(emit_bg_palindex(nc, f, srccell)){
            return -1;
          }
        }else if(!nccell_bg_default_p(srccell)){ // rgb background
          // if our cell has a non-default background, we can elide the
          // non-default background set iff either:
          //  * we do not use the background, because the cell is all-foreground,
          //  * the previous was non-default, and matches what we have now, or
          nccell_bg_rgb8(srccell, &br, &bg, &bb);
          if(nc->rstate.bgelidable && nc->rstate.lastbr == br && nc->rstate.lastbg == bg && nc->rstate.lastbb == bb){
            ++nc->stats.s.bgelisions;
          }else{
            if(term_bg_rgb8(&nc->tcache, f, br, bg, bb)){
              return -1;
            }
            ++nc->stats.s.bgemissions;
            nc->rstate.bgelidable = true;
          }
          nc->rstate.lastbr = br; nc->rstate.lastbg = bg; nc->rstate.lastbb = bb;
          nc->rstate.bgdefelidable = false;
          nc->rstate.bgpalelidable = false;
          // FIXME see above -- if we don't have to change the fg, but do need
          // to change the bg, emit full block and no rgb at all
          if(rgbequal){
            // FIXME need one per column of original glyph
            pool_load_direct(&nc->pool, srccell, " ", 1, 1);
          }
        }
//fprintf(stderr, "RAST %08x [%s] to %d/%d cols: %u %016" PRIx64 "\n", srccell->gcluster, pool_extended_gcluster(&nc->pool, srccell), y, x, srccell->width, srccell->channels);
        // this is used to invalidate the sprixel in the first text round,
        // which is only necessary for sixel, not kitty.
        if(rvec[damageidx].sprixel){
          sprixcell_e scstate = sprixel_state(rvec[damageidx].sprixel, y - nc->margin_t, x - nc->margin_l);
          if((scstate == SPRIXCELL_MIXED_SIXEL || scstate == SPRIXCELL_OPAQUE_SIXEL)
             && !rvec[damageidx].s.p_beats_sprixel){
//fprintf(stderr, "INVALIDATING at %d/%d (%u)\n", y, x, rvec[damageidx].s.p_beats_sprixel);
            sprixel_invalidate(rvec[damageidx].sprixel, y - nc->margin_t, x - nc->margin_l);
          }
        }
        if(term_putc(f, &nc->pool, srccell)){
          return -1;
        }
        if(srccell->gcluster == '\n'){
          saw_linefeed = true;
        }
        rvec[damageidx].s.damaged = 0;
        rvec[damageidx].s.p_beats_sprixel = 0;
        nc->rstate.x += srccell->width;
        if(srccell->width){ // check only necessary when undamaged; be safe
          x += srccell->width - 1;
        }else{
          ++nc->rstate.x;
        }
        if((int)y > nc->rstate.logendy || ((int)y == nc->rstate.logendy && (int)x > nc->rstate.logendx)){
          if((int)y > nc->rstate.logendy){
//fprintf(stderr, "**************8NATURAL PLACEMENT AT %u/ %u\n", y, x);
            nc->rstate.logendy = y;
            nc->rstate.logendx = 0;
          }
          if(x >= p->dimx + nc->margin_l - 1){
            if(nc->rstate.logendy < (int)(p->dimy + nc->margin_t - 1)){
              ++nc->rstate.logendy;
            }
            nc->rstate.logendx = 0;
            saw_linefeed = false;
//fprintf(stderr, "**********8SCROLLING PLACEMENT AT %u %u\n", nc->rstate.logendy, nc->rstate.logendx);
          }else if((int)x >= nc->rstate.logendx){
            nc->rstate.logendx = x;
//fprintf(stderr, "**********HORIZ PLACEMENT AT %u %u\n", nc->rstate.logendy, nc->rstate.logendx);
          }
        }
      }
//fprintf(stderr, "damageidx: %ld\n", damageidx);
    }
    // shouldn't this happen only if the linefeed was on the last line? FIXME
    if(saw_linefeed){
      nc->rstate.logendx = 0;
    }
  }
  return 0;
}
 
// 'asu' on input is non-0 if application-synchronized updates are permitted
// (they are not, for instance, when rendering to a non-tty). on output,
// assuming success, it is non-0 if application-synchronized updates are
// desired; in this case, a SUM footer is present at the end of the buffer.
static int
notcurses_rasterize_inner(notcurses* nc, ncpile* p, fbuf* f, unsigned* asu){
  logdebug("pile %p ymax: %d xmax: %d", p, p->dimy + nc->margin_t, p->dimx + nc->margin_l);
  // don't write a clearscreen. we only update things that have been changed.
  // we explicitly move the cursor at the beginning of each output line, so no
  // need to home it expliticly.
  update_palette(nc, f);
  int scrolls = p->scrolls;
  logdebug("sprixel phase 1");
  int64_t sprixelbytes = clean_sprixels(nc, p, f, scrolls);
  if(sprixelbytes < 0){
    return -1;
  }
  logdebug("glyph phase 1");
  if(rasterize_core(nc, p, f, 0)){
    return -1;
  }
  logdebug("sprixel phase 2");
  int64_t rasprixelbytes = rasterize_sprixels(nc, p, f);
  if(rasprixelbytes < 0){
    return -1;
  }
  sprixelbytes += rasprixelbytes;
  pthread_mutex_lock(&nc->stats.lock);
    nc->stats.s.sprixelbytes += sprixelbytes;
  pthread_mutex_unlock(&nc->stats.lock);
  logdebug("glyph phase 2");
  if(rasterize_scrolls(p, f)){
    return -1;
  }
  p->scrolls = 0;
  if(rasterize_core(nc, p, f, 1)){
    return -1;
  }
#define MIN_SUMODE_SIZE BUFSIZ
  if(*asu){
    if(nc->rstate.f.used >= MIN_SUMODE_SIZE){
      const char* endasu = get_escape(&nc->tcache, ESCAPE_ESUM);
      if(endasu){
        if(fbuf_puts(f, endasu) < 0){
          *asu = 0;
        }
      }else{
        *asu = 0;
      }
    }else{
      *asu = 0;
    }
  }
#undef MIN_SUMODE_SIZE
  return nc->rstate.f.used;
}
 
// rasterize the rendered frame, and blockingly write it out to the terminal.
static int
raster_and_write(notcurses* nc, ncpile* p, fbuf* f){
  fbuf_reset(f);
  // will we be using application-synchronized updates? if this comes back as
  // non-zero, we are, and must emit the header. no SUM without a tty, and we
  // can't have the escape without being connected to one...
  const char* basu = get_escape(&nc->tcache, ESCAPE_BSUM);
  unsigned useasu = basu ? 1 : 0;
  // if we have SUM support, emit a BSU speculatively. if we do so, but don't
  // actually use an ESU, this BSUM must be skipped on write.
  if(useasu){
    if(fbuf_puts(f, basu) < 0){
      return -1;
    }
  }
  if(notcurses_rasterize_inner(nc, p, f, &useasu) < 0){
    return -1;
  }
  // if we loaded a BSU into the front, but don't actually want to use it,
  // we start printing after the BSU.
  size_t moffset = 0;
  if(basu){
    if(useasu){
      ++nc->stats.s.appsync_updates;
    }else{
      moffset = strlen(basu);
    }
  }
  int ret = 0;
  sigset_t oldmask;
  block_signals(&oldmask);
  if(blocking_write(fileno(nc->ttyfp), nc->rstate.f.buf + moffset,
                    nc->rstate.f.used - moffset)){
    ret = -1;
  }
  unblock_signals(&oldmask);
  rasterize_sprixels_post(nc, p);
//fprintf(stderr, "%lu/%lu %lu/%lu %lu/%lu %d\n", nc->stats.defaultelisions, nc->stats.defaultemissions, nc->stats.fgelisions, nc->stats.fgemissions, nc->stats.bgelisions, nc->stats.bgemissions, ret);
  if(ret < 0){
    return ret;
  }
  return nc->rstate.f.used;
}
 
// if the cursor is enabled, store its location and disable it. then, once done
// rasterizing, enable it afresh, moving it to the stored location. if left on
// during rasterization, we'll get grotesque flicker. 'out' is a memstream
// used to collect a buffer.
static inline int
notcurses_rasterize(notcurses* nc, ncpile* p, fbuf* f){
  const int cursory = nc->cursory;
  const int cursorx = nc->cursorx;
  if(cursory >= 0){ // either both are good, or neither is
    notcurses_cursor_disable(nc);
  }
  int ret = raster_and_write(nc, p, f);
  fbuf_reset(f);
  if(cursory >= 0){
    notcurses_cursor_enable(nc, cursory, cursorx);
  }else if(nc->rstate.logendy >= 0){
    goto_location(nc, f, nc->rstate.logendy, nc->rstate.logendx, nc->rstate.lastsrcp);
    if(fbuf_flush(f, nc->ttyfp)){
      ret = -1;
    }
  }
  nc->last_pile = p;
  return ret;
}
 
// get the cursor to the upper-left corner by one means or another, clearing
// the screen while doing so.
int clear_and_home(notcurses* nc, tinfo* ti, fbuf* f){
  // clear clears the screen and homes the cursor by itself
  const char* clearscr = get_escape(ti, ESCAPE_CLEAR);
  if(clearscr){
    if(fbuf_emit(f, clearscr) == 0){
      nc->rstate.x = nc->rstate.y = 0;
      return 0;
    }
  }
  if(emit_scrolls_track(nc, ncplane_dim_y(notcurses_stdplane_const(nc)), f)){
    return -1;
  }
  if(goto_location(nc, f, 0, 0, NULL)){
    return -1;
  }
  return 0;
}
 
int notcurses_refresh(notcurses* nc, unsigned* restrict dimy, unsigned* restrict dimx){
  if(notcurses_resize(nc, dimy, dimx)){
    return -1;
  }
  fbuf_reset(&nc->rstate.f);
  if(clear_and_home(nc, &nc->tcache, &nc->rstate.f)){
    return -1;
  }
  if(fbuf_flush(&nc->rstate.f, nc->ttyfp)){
    return -1;
  }
  if(nc->lfdimx == 0 || nc->lfdimy == 0){
    return 0;
  }
  ncpile p = {0};
  p.dimy = nc->lfdimy;
  p.dimx = nc->lfdimx;
  const int count = p.dimy * p.dimx;
  p.crender = malloc(count * sizeof(*p.crender));
  if(p.crender == NULL){
    return -1;
  }
  init_rvec(p.crender, count);
  for(int i = 0 ; i < count ; ++i){
    p.crender[i].s.damaged = 1;
  }
  int ret = notcurses_rasterize(nc, &p, &nc->rstate.f);
  free(p.crender);
  if(ret < 0){
    return -1;
  }
  ++nc->stats.s.refreshes;
  return 0;
}
 
int ncpile_render_to_file(ncplane* n, FILE* fp){
  notcurses* nc = ncplane_notcurses(n);
  ncpile* p = ncplane_pile(n);
  if(nc->lfdimx == 0 || nc->lfdimy == 0){
    return 0;
  }
  fbuf f = {0};
  if(fbuf_init(&f)){
    return -1;
  }
  const unsigned count = (nc->lfdimx > p->dimx ? nc->lfdimx : p->dimx) *
                         (nc->lfdimy > p->dimy ? nc->lfdimy : p->dimy);
  p->crender = malloc(count * sizeof(*p->crender));
  if(p->crender == NULL){
    fbuf_free(&f);
    return -1;
  }
  init_rvec(p->crender, count);
  for(unsigned i = 0 ; i < count ; ++i){
    p->crender[i].s.damaged = 1;
  }
  int ret = raster_and_write(nc, p, &f);
  free(p->crender);
  if(ret > 0){
    if(fwrite(f.buf, f.used, 1, fp) == 1){
      ret = 0;
    }else{
      ret = -1;
    }
  }
  fbuf_free(&f);
  return ret;
}
 
// We execute the painter's algorithm, starting from our topmost plane. The
// damagevector should be all zeros on input. On success, it will reflect
// which cells were changed. We solve for each coordinate's cell by walking
// down the z-buffer, looking at intersections with ncplanes. This implies
// locking down the EGC, the attributes, and the channels for each cell.
// if |pgeo_changed|, the cell-pixel geometry for the pile has changed
// since the last render, and thus all sprixels need be rescaled.
static void
ncpile_render_internal(ncpile* p, unsigned pgeo_changed){
  struct crender* rvec = p->crender;
//fprintf(stderr, "rendering %dx%d\n", p->dimy, p->dimx);
  ncplane* pl = p->top;
  sprixel* sprixel_list = NULL;
  while(pl){
    paint(pl, rvec, p->dimy, p->dimx, 0, 0, &sprixel_list, pgeo_changed);
    pl = pl->below;
  }
  if(sprixel_list){
    if(p->sprixelcache){
      sprixel* s = sprixel_list;
      while(s->next){
        s = s->next;
      }
      if( (s->next = p->sprixelcache) ){
        p->sprixelcache->prev = s;
      }
    }
    p->sprixelcache = sprixel_list;
  }
}
 
int ncpile_rasterize(ncplane* n){
  struct timespec start, rasterdone, writedone;
  clock_gettime(CLOCK_MONOTONIC, &start);
  ncpile* pile = ncplane_pile(n);
  struct notcurses* nc = ncpile_notcurses(pile);
  const struct tinfo* ti = &ncplane_notcurses_const(n)->tcache;
  postpaint(nc, ti, nc->lastframe, pile->dimy, pile->dimx, pile->crender, &nc->pool);
  clock_gettime(CLOCK_MONOTONIC, &rasterdone);
  int bytes = notcurses_rasterize(nc, pile, &nc->rstate.f);
  clock_gettime(CLOCK_MONOTONIC, &writedone);
  pthread_mutex_lock(&nc->stats.lock);
    // accepts negative |bytes| as an indication of failure
    update_raster_bytes(&nc->stats.s, bytes);
    update_raster_stats(&rasterdone, &start, &nc->stats.s);
    update_write_stats(&writedone, &rasterdone, &nc->stats.s, bytes);
  pthread_mutex_unlock(&nc->stats.lock);
  // we want to refresh if the screen geometry changed (or if we were just
  // woken up from SIGSTOP), but we mustn't do so until after rasterizing
  // the solved rvec, since this might result in a geometry update.
  if(sigcont_seen_for_render){
    sigcont_seen_for_render = 0;
    notcurses_refresh(ncplane_notcurses(n), NULL, NULL);
  }
  if(bytes < 0){
    return -1;
  }
  return 0;
}
 
// ensure the crender vector of 'n' is properly sized for 'n'->dimy x 'n'->dimx,
// and initialize the rvec afresh for a new render.
static int
engorge_crender_vector(ncpile* p){
  if(p->dimy <= 0 || p->dimx <= 0){
    return -1;
  }
  const size_t crenderlen = p->dimy * p->dimx; // desired size
//fprintf(stderr, "crlen: %d y: %d x:%d\n", crenderlen, dimy, dimx);
  if(crenderlen != p->crenderlen){
    loginfo("resizing rvec (%" PRIuPTR ") for %p to %" PRIuPTR,
            p->crenderlen, p, crenderlen);
    struct crender* tmp = realloc(p->crender, sizeof(*tmp) * crenderlen);
    if(tmp == NULL){
      return -1;
    }
    p->crender = tmp;
    p->crenderlen = crenderlen;
  }
  init_rvec(p->crender, crenderlen);
  return 0;
}
 
int ncpile_render(ncplane* n){
  scroll_lastframe(ncplane_notcurses(n), ncplane_pile(n)->scrolls);
  struct timespec start, renderdone;
  clock_gettime(CLOCK_MONOTONIC, &start);
  notcurses* nc = ncplane_notcurses(n);
  ncpile* pile = ncplane_pile(n);
  // update our notion of screen geometry, and render against that
  unsigned pgeo_changed = 0;
  notcurses_resize_internal(n, NULL, NULL);
  if(pile->cellpxy != nc->tcache.cellpxy || pile->cellpxx != nc->tcache.cellpxx){
    pile->cellpxy = nc->tcache.cellpxy;
    pile->cellpxx = nc->tcache.cellpxx;
    pgeo_changed = 1;
  }
  if(engorge_crender_vector(pile)){
    return -1;
  }
  ncpile_render_internal(pile, pgeo_changed);
  clock_gettime(CLOCK_MONOTONIC, &renderdone);
  pthread_mutex_lock(&nc->stats.lock);
    update_render_stats(&renderdone, &start, &nc->stats.s);
  pthread_mutex_unlock(&nc->stats.lock);
  return 0;
}
 
// run the top half of notcurses_render(), and steal the buffer from rstate.
int ncpile_render_to_buffer(ncplane* p, char** buf, size_t* buflen){
  if(ncpile_render(p)){
    return -1;
  }
  notcurses* nc = ncplane_notcurses(p);
  unsigned useasu = false; // no SUM with file
  fbuf_reset(&nc->rstate.f);
  int bytes = notcurses_rasterize_inner(nc, ncplane_pile(p), &nc->rstate.f, &useasu);
  pthread_mutex_lock(&nc->stats.lock);
    update_raster_bytes(&nc->stats.s, bytes);
  pthread_mutex_unlock(&nc->stats.lock);
  if(bytes < 0){
    return -1;
  }
  *buf = nc->rstate.f.buf;
  *buflen = nc->rstate.f.used;
  fbuf_reset(&nc->rstate.f);
  return 0;
}
 
// copy the UTF8-encoded EGC out of the cell, whether simple or complex. the
// result is not tied to the ncplane, and persists across erases / destruction.
static inline char*
pool_egc_copy(const egcpool* e, const nccell* c){
  if(cell_simple_p(c)){
    return strdup((const char*)&c->gcluster);
  }
  return strdup(egcpool_extended_gcluster(e, c));
}
 
char* notcurses_at_yx(notcurses* nc, unsigned yoff, unsigned xoff, uint16_t* stylemask, uint64_t* channels){
  if(nc->lastframe == NULL){
    logerror("haven't yet rendered");
    return NULL;
  }
  if(yoff >= nc->lfdimy){
    logerror("invalid coordinates: %u/%u", yoff, xoff);
    return NULL;
  }
  if(xoff >= nc->lfdimx){
    logerror("invalid coordinates: %u/%u", yoff, xoff);
    return NULL;
  }
  const nccell* srccell = &nc->lastframe[yoff * nc->lfdimx + xoff];
  if(nccell_wide_right_p(srccell)){
    return notcurses_at_yx(nc, yoff, xoff - 1, stylemask, channels);
  }
  if(stylemask){
    *stylemask = srccell->stylemask;
  }
  if(channels){
    *channels = srccell->channels;
  }
//fprintf(stderr, "COPYING: %d from %p\n", srccell->gcluster, &nc->pool);
  return pool_egc_copy(&nc->pool, srccell);
}
 
int ncdirect_set_bg_rgb_f(ncdirect* nc, unsigned rgb, fbuf* f){
  if(rgb > 0xffffffu){
    return -1;
  }
  if(!ncdirect_bg_default_p(nc) && !ncdirect_bg_palindex_p(nc)
     && ncchannels_bg_rgb(nc->channels) == rgb){
    return 0;
  }
  if(term_bg_rgb8(&nc->tcache, f, (rgb & 0xff0000u) >> 16u, (rgb & 0xff00u) >> 8u, rgb & 0xffu)){
    return -1;
  }
  ncchannels_set_bg_rgb(&nc->channels, rgb);
  return 0;
}
 
int ncdirect_set_bg_rgb(ncdirect* nc, unsigned rgb){
  fbuf f = {0};
  if(fbuf_init_small(&f)){
    return -1;
  }
  if(ncdirect_set_bg_rgb_f(nc, rgb, &f)){
    fbuf_free(&f);
    return -1;
  }
  if(fbuf_finalize(&f, nc->ttyfp) < 0){
    return -1;
  }
  return 0;
}
 
int ncdirect_set_fg_rgb_f(ncdirect* nc, unsigned rgb, fbuf* f){
  if(rgb > 0xffffffu){
    return -1;
  }
  if(!ncdirect_fg_default_p(nc) && !ncdirect_fg_palindex_p(nc)
     && ncchannels_fg_rgb(nc->channels) == rgb){
    return 0;
  }
  if(term_fg_rgb8(&nc->tcache, f, (rgb & 0xff0000u) >> 16u, (rgb & 0xff00u) >> 8u, rgb & 0xffu)){
    return -1;
  }
  ncchannels_set_fg_rgb(&nc->channels, rgb);
  return 0;
}
 
int ncdirect_set_fg_rgb(ncdirect* nc, unsigned rgb){
  fbuf f = {0};
  if(fbuf_init_small(&f)){
    return -1;
  }
  if(ncdirect_set_fg_rgb_f(nc, rgb, &f)){
    fbuf_free(&f);
    return -1;
  }
  if(fbuf_finalize(&f, nc->ttyfp) < 0){
    return -1;
  }
  return 0;
}
 
int notcurses_default_foreground(const struct notcurses* nc, uint32_t* fg){
  const tinfo* ti = &nc->tcache;
  if(ti->fg_default & 0x80000000){
    logerror("default foreground could not be determined");
    return -1;
  }
  *fg = ti->fg_default & NC_BG_RGB_MASK;
  return 0;
}
 
int notcurses_default_background(const struct notcurses* nc, uint32_t* bg){
  const tinfo* ti = &nc->tcache;
  if(ti->bg_collides_default & 0x80000000){
    logerror("default background could not be determined");
    return -1;
  }
  *bg = ti->bg_collides_default & NC_BG_RGB_MASK;
  return 0;
}
 
int notcurses_cursor_yx(const notcurses* nc, int* y, int* x){
  *y = nc->rstate.y;
  *x = nc->rstate.x;
  return 0;
}
 
int notcurses_cursor_enable(notcurses* nc, int y, int x){
  if(y < 0 || x < 0){
    logerror("illegal cursor placement: %d, %d", y, x);
    return -1;
  }
  // if we're already at the demanded location, we must already be visible, and
  // we needn't move the cursor -- return success immediately.
  if(nc->cursory == y && nc->cursorx == x){
    return 0;
  }
  fbuf f = {0};
  if(fbuf_init_small(&f)){
    return -1;
  }
  // updates nc->rstate.cursor{y,x}
  if(goto_location(nc, &f, y + nc->margin_t, x + nc->margin_l, nc->rstate.lastsrcp)){
    fbuf_free(&f);
    return -1;
  }
  if(nc->cursory < 0){ // we weren't visible before, need cnorm
    const char* cnorm = get_escape(&nc->tcache, ESCAPE_CNORM);
    if(!cnorm || fbuf_emit(&f, cnorm)){
      fbuf_free(&f);
      return -1;
    }
  }
  if(fbuf_finalize(&f, nc->ttyfp)){
    return -1;
  }
  nc->cursory = y;
  nc->cursorx = x;
  return 0;
}
 
int notcurses_cursor_disable(notcurses* nc){
  if(nc->cursorx < 0 || nc->cursory < 0){
    logerror("cursor is not enabled");
    return -1;
  }
  const char* cinvis = get_escape(&nc->tcache, ESCAPE_CIVIS);
  if(cinvis){
    if(!tty_emit(cinvis, nc->tcache.ttyfd) && !ncflush(nc->ttyfp)){
      nc->cursory = -1;
      nc->cursorx = -1;
      return 0;
    }
  }
  return -1;
}