pool_load_direct: call egc_rtl() for RTL U200E forcings #850

src/demo/all.c

@@ -8,10 +8,6 @@ problematic_unicode(char32_t wc){
   if(wc >= 0xd800 && wc <= 0xdfff){
     return 1;
   }
-  // right-to-left text fucks with us hard
-  if(uc_bidi_category(wc)){ // FIXME
-    return 1;
-  }
   // these are broken in several terminals ㉈ ㉉ ㉊ ㉋ ㉌ ㉍ ㉎ ㉏
   // https://github.com/dankamongmen/notcurses/issues/881
   if(wc >= 0x3248 && wc <= 0x324f){

src/demo/unicodeblocks.c

@@ -75,11 +75,6 @@ draw_block(struct ncplane* nn, uint32_t blockstart){
         if(wcwidth(w) < 2){
           utf8arr[bwc++] = ' ';
         }
-        if(uc_bidi_category(w)){
-          utf8arr[bwc++] = 0xe2;
-          utf8arr[bwc++] = 0x80;
-          utf8arr[bwc++] = 0x8e;
-        }
         utf8arr[bwc++] = '\0';
       }else{ // don't dump non-printing codepoints
         strcpy(utf8arr, "  ");

src/demo/whiteout.c

@@ -289,8 +289,6 @@ int witherworm_demo(struct notcurses* nc){
     "ብርሃነ ዘርኣይ",
     "ኃይሌ ገብረሥላሴ",
     "ᓱᒻᒪᓂᒃᑯᐊ ᐃᓄᑦᑎᑐᑐᐃᓐᓇᔭᙱᓚᑦ",
-    "∮ E⋅da = Q,  n → ∞, ∑ f(i) = ∏ g(i), ∀x∈ℝ: ⌈x⌉ = −⌊−x⌋, α ∧ ¬β = ¬(¬α ∨ β)"
-    "2H₂ + O₂ ⇌ 2H₂O, R = 4.7 kΩ, ⌀ 200mm",
     "ði ıntəˈnæʃənəl fəˈnɛtık əsoʊsiˈeıʃn",
     "((V⍳V)=⍳⍴V)/V←,V    ⌷←⍳→⍴∆∇⊃‾⍎⍕⌈",
     "Eڿᛯℇ✈ಅΐʐ𝍇Щঅ℻ ⌬⌨ ⌣₰ ⠝ ‱ ‽ ח ֆ ∜ ⨀ IJႪ ⇠ ਐ ῼ இ ╁ ଠ ୭ ⅙ ㈣⧒ ₔ ⅷ ﭗ ゛〃・ ↂ ﻩ ✞ ℼ ⌧",
@@ -501,10 +499,6 @@ int witherworm_demo(struct notcurses* nc){
             idx += eaten;
             continue;
           }
-          if(uc_bidi_category(wcs)){ // FIXME
-            idx += eaten;
-            continue;
-          }
           int ulen = 0;
           int r;
           if(wcwidth(wcs) <= maxx - x){

src/lib/internal.h

@@ -898,6 +898,30 @@ cell_blend_bchannel(cell* cl, unsigned channel, unsigned* blends){
   return cell_set_bchannel(cl, channels_blend(cell_bchannel(cl), channel, blends));
 }
 
+// examine the UTF-8 EGC in the first |*bytes| bytes of |egc|. if the EGC is
+// right-to-left, we make a copy, appending an U+200E to force left-to-right.
+// only the first unicode char of the EGC is currently checked FIXME. if the
+// EGC is not RTL, we return NULL.
+static char*
+egc_rtl(const char* egc, int* bytes){
+  wchar_t w;
+  mbstate_t mbstate = { };
+  size_t r = mbrtowc(&w, egc, *bytes, &mbstate);
+  if(r == (size_t)-1 || r == (size_t)-2){
+    return NULL;
+  }
+  if(!uc_bidi_category(w)){ // FIXME too aggressive, counts punctuation etc
+    return NULL;
+  }
+  // insert U+200E, "LEFT-TO-RIGHT MARK". This ought reset the text direction
+  // after emitting a potentially RTL EGC.
+  const char LTRMARK[] = "\xe2\x80\x8e";
+  char* s = (char*)malloc(*bytes + sizeof(LTRMARK)); // cast for C++ callers
+  memcpy(s, egc, *bytes);
+  memcpy(s + *bytes, LTRMARK, sizeof(LTRMARK));
+  return s;
+}
+
 static inline int
 pool_load_direct(egcpool* pool, cell* c, const char* gcluster, int bytes, int cols){
   if(bytes < 0 || cols < 0){
@@ -923,15 +947,26 @@ pool_load_direct(egcpool* pool, cell* c, const char* gcluster, int bytes, int co
     c->channels |= CELL_NOBACKGROUND_MASK;
     c->channels &= ~CELL_WIDEASIAN_MASK;
   }
+  // checks for RTL and adds U+200E if so FIXME
+  char* rtl = egc_rtl(gcluster, &bytes);
+  if(rtl){
+    gcluster = rtl;
+  }
   if(bytes <= 4){
     if(strcmp(gcluster, (const char*)&c->gcluster)){
       pool_release(pool, c);
       c->gcluster = 0;
       memcpy(&c->gcluster, gcluster, bytes);
     }
+    if(rtl){
+      free(rtl);
+    }
     return bytes;
   }
   int eoffset = egcpool_stash(pool, gcluster, bytes);
+  if(rtl){
+    free(rtl);
+  }
   if(eoffset < 0){
     return -1;
   }

src/lib/render.c

@@ -899,12 +899,6 @@ notcurses_rasterize_inner(notcurses* nc, const struct crender* rvec, FILE* out){
           ++nc->rstate.x;
           ++x;
         }
-        // if the terminal's own motion carried us down to the next line,
-        // we need update our concept of the cursor's true y
-        /*if(nc->rstate.x >= nc->truecols){
-          ++nc->rstate.y; // FIXME not if on last line, right?
-          nc->rstate.x = 0;
-        }*/
       }
 //fprintf(stderr, "damageidx: %ld\n", damageidx);
     }