egc inlining #830

include/notcurses/notcurses.h

@@ -485,7 +485,7 @@ channels_set_bg_default(uint64_t* channels){
 //
 // Each cell occupies 16 static bytes (128 bits). The surface is thus ~1.6MB
 // for a (pretty large) 500x200 terminal. At 80x43, it's less than 64KB.
-// Dynamic requirements (the egcpool) can add up to 32MB to an ncplane, but
+// Dynamic requirements (the egcpool) can add up to 16MB to an ncplane, but
 // such large pools are unlikely in common use.
 //
 // We implement some small alpha compositing. Foreground and background both
@@ -506,9 +506,12 @@ channels_set_bg_default(uint64_t* channels){
 // RGB is used if neither default terminal colors nor palette indexing are in
 // play, and fully supports all transparency options.
 typedef struct cell {
-  // These 32 bits are either a single-byte, single-character grapheme cluster
-  // (values 0--0x7f), or an offset into a per-ncplane attached pool of
-  // varying-length UTF-8 grapheme clusters. This pool may thus be up to 32MB.
+  // These 32 bits are either a complete grapheme cluster in 4 bytes or less,
+  // or 0x01000000 plus an offset into a per-ncplane attached pool of varying-
+  // length UTF-8 EGCs (an egcpool). This pool may thus be up to 16MB.
+  // Obviously, this implies that EGCs beginning with 0x01 are disallowed; such
+  // an EGC, if we wanted to support them, could be spilled to the egcpool. If
+  // the EGC is less than four bytes, it will be padded with zeroes.
   uint32_t gcluster;          // 4B -> 4B
   // 8 bits of zero + 8 reserved bits + NCSTYLE_* attributes (16 bits).
   // (attrword & 0xff000000): reserved, *must be zero*
@@ -648,7 +651,7 @@ cell_wide_left_p(const cell* c){
 // Is the cell simple (a lone ASCII character, encoded as such)?
 static inline bool
 cell_simple_p(const cell* c){
-  return c->gcluster < 0x80;
+  return (c->gcluster >> 24u) != 0x01;
 }
 
 // return a pointer to the NUL-terminated EGC referenced by 'c'. this pointer

src/lib/egcpool.h

@@ -249,7 +249,7 @@ egcpool_dump(egcpool* pool){
 // unsafe results if called on a simple cell.
 static inline uint32_t
 cell_egc_idx(const cell* c){
-  return c->gcluster - 0x80;
+  return c->gcluster & 0x00fffffflu;
 }
 
 __attribute__ ((__returns_nonnull__)) static inline const char*

src/lib/internal.h

@@ -608,7 +608,7 @@ cell_duplicate_far(egcpool* tpool, cell* targ, const ncplane* splane, const cell
   if(eoffset < 0){
     return -1;
   }
-  targ->gcluster = eoffset + 0x80;
+  targ->gcluster = 0x01000000ul + eoffset;
   return ulen;
 }
 

src/lib/notcurses.c

@@ -1364,36 +1364,30 @@ cell_load_direct(ncplane* n, cell* c, const char* gcluster, int bytes, int cols)
   if(bytes < 0 || cols < 0){
     return -1;
   }
-  if(bytes <= 1){
-    assert(cols < 2);
-    cell_release(n, c);
-    c->channels &= ~(CELL_WIDEASIAN_MASK | CELL_NOBACKGROUND_MASK);
-    c->gcluster = *gcluster;
-    return bytes;
-  }
-  if(!cell_simple_p(c)){
-    if(strcmp(gcluster, cell_extended_gcluster(n, c)) == 0){
-      return bytes; // reduce, reuse, recycle
-    }else{
-      cell_release(n, c);
-    }
-  }
-  if(cols > 1){
-    c->channels |= CELL_WIDEASIAN_MASK;
-  }else{
+  if(cols < 2){
     c->channels &= ~CELL_WIDEASIAN_MASK;
+  }else{
+    c->channels |= CELL_WIDEASIAN_MASK;
   }
-  // FIXME also shaded blocks! ░ etc
+  // FIXME also shaded blocks! ░ etc. are there combined EGCs involving these?
   if(strncmp(gcluster, "\xe2\x96\x88", 3)){
     c->channels &= ~CELL_NOBACKGROUND_MASK;
   }else{
     c->channels |= CELL_NOBACKGROUND_MASK;
   }
+  if(bytes <= 4){
+    cell_release(n, c);
+    c->channels &= ~CELL_WIDEASIAN_MASK;
+    c->gcluster = 0;
+    memcpy(&c->gcluster, gcluster, bytes);
+    return bytes;
+  }
   int eoffset = egcpool_stash(&n->pool, gcluster, bytes);
   if(eoffset < 0){
     return -1;
   }
-  c->gcluster = eoffset + 0x80;
+  cell_release(n, c);
+  c->gcluster = 0x01000000ul + eoffset;
   return bytes;
 }
 
@@ -1452,6 +1446,7 @@ int ncplane_putegc_yx(ncplane* n, int y, int x, const char* gclust, int* sbytes)
   if(cell_load_direct(n, targ, gclust, bytes, cols) < 0){
     return -1;
   }
+//fprintf(stderr, "%08x %d %d\n", targ->gcluster, bytes, cols);
   targ->attrword = n->attrword;
   targ->channels = channels;
   if(wide){ // must set our right wide, and check for further damage

src/lib/render.c

@@ -500,16 +500,12 @@ ncfputc(char c, FILE* out){
 static int
 term_putc(FILE* out, const egcpool* e, const cell* c){
   if(cell_simple_p(c)){
-    if(c->gcluster == 0 || iscntrl(c->gcluster)){
-// fprintf(stderr, "[ ]\n");
-      if(ncfputc(' ', out) == EOF){
-        return -1;
-      }
-    }else{
-//fprintf(stderr, "[%c]\n", c->gcluster);
-      if(ncfputc(c->gcluster, out) == EOF){
-        return -1;
-      }
+    char egc[5];
+    memset(egc, 0, sizeof(egc));
+    memcpy(egc, &c->gcluster, sizeof(c->gcluster));
+//fprintf(stderr, "[%ls]\n", egc);
+    if(ncfputs(egc, out) == EOF){
+      return -1;
     }
   }else{
     const char* ext = egcpool_extended_gcluster(e, c);