Merge pull request #407 from phi-gamma/masterHEADmaster

fixes, mostly

5 files changed, 139 insertions, 54 deletions

diff --git a/doc/luaotfload-main.tex b/doc/luaotfload-main.tex
index c26235e..21e001b 100644
--- a/doc/luaotfload-main.tex
+++ b/doc/luaotfload-main.tex
@@ -183,7 +183,7 @@ for a more formal description see figure \ref{font-syntax}.
       \alt                  <unquoted font request> ;
 
       <unquoted font request> ::= <specification>, [`:', <feature list> ]
-      \alt                        `[', <path lookup> `]', [ [`:'], <feature list> ] ;
+      \alt                        <path lookup>, [ [`:'], <feature list> ] ;
 
       <specification>    ::= <prefixed spec>, [ <subfont no> ], \{ <modifier> \}
       \alt                   <anon lookup>, \{ <modifier> \} ;
@@ -218,7 +218,13 @@ for a more formal description see figure \ref{font-syntax}.
 
       <anon lookup>      ::= {\sc tfmname} | <name lookup> ;
 
-      <path lookup>      ::= \{ {\sc all_characters} - `]' \} ;
+      <path lookup>      ::= `[', \{ <path content> \}, `]', [ <subfont no> ] ;
+
+      <path content>     ::= <path balanced>
+      \alt                   `\\', {\sc all_characters}
+      \alt                   {\sc all_characters} - `]'
+
+      <path balanced>    ::= `[', [ <path content> ], `]'
 
       <modifier>         ::= `/', (`I' | `B' | `BI' | `IB' | `S=', \{ {\sc digit} \} ) ;
 
@@ -234,6 +240,7 @@ for a more formal description see figure \ref{font-syntax}.
       <name character>   ::= {\sc all_characters} - ( `(' | `/' | `:' ) ;
 \endsyntaxfloat
 
+%% Below guarded space gets borked in index; why‽
 \beginsubsection{Prefix -- the \identifier{luaotfload}{ }Way}
 
 In \identifier{luaotfload}, the canonical syntax for font requests
@@ -329,13 +336,12 @@ The file names corresponding to the example font names above are
 
 \endsubsection
 
-\beginsubsection {Compatibility Layer}
-
-In addition to the regular prefixed requests, \identifier{luaotfload}
-accepts loading fonts the \XETEX way.
+\beginsubsection {Bracketed Lookups}
+\label{sec:conf}
+Bracketed lookups allow for arbitrary character content to be used in a
+definition.
 %
-There are again two modes: bracketed and unbracketed.
-A bracketed request looks as follows.
+A simple bracketed request looks follows the scheme
 
 \beginnarrower
   \nonproportional{\string\font\string\fontname\space = [}%
@@ -344,21 +350,41 @@ A bracketed request looks as follows.
 \endnarrower
 
 \noindent
-Inside the square brackets, every character except for a closing
-bracket is permitted, allowing for specifying paths to a font file.
+Inside the square brackets, every character except for a closing bracket is
+permitted, allowing for  arbitrary paths to a font file -- including Windows
+style paths with UNC or drive letter prepended -- to be specified.
 %
+The \identifier{Luaotfload} syntax differs from \XETEX in that the subfont
+selector goes \emphasis{after} the closing bracket:
+
+\beginnarrower
+  \nonproportional{\string\font\string\fontname\space = [}%
+  \meta{/path/to/file}%
+  \nonproportional{]}
+  \nonproportional{(}n\nonproportional{)}
+\endnarrower
+
 Naturally, path-less file names are equally valid and processed the
 same way as an ordinary \inlinecode {file:} lookup.
 
+\beginsubsection {Compatibility}
+
+In addition to the regular prefixed requests, \identifier{luaotfload}
+accepts loading fonts the \XETEX way.
+%
+There are again two modes: bracketed and unbracketed.
+For the bracketed variety, see above, \ref{sec:conf}.
+
+Unbracketed (or, for lack of a better word: \emphasis{anonymous})
+font requests resemble the conventional \TEX syntax.
+
 \beginnarrower
   \nonproportional{\string\font\string\fontname\space= }%
   \meta{font name}
   \dots
 \endnarrower
+\endsubsection
 
-Unbracketed (or, for lack of a better word: \emphasis{anonymous})
-font requests resemble the conventional \TEX syntax.
-%
 However, they have a broader spectrum of possible interpretations:
 before anything else, \identifier{luaotfload} attempts to load a
 traditional \TEX Font Metric (\abbrev{tfm} or \abbrev{ofm}).
@@ -437,6 +463,13 @@ name.\footnote{%
   \font \cambriamath = "file:cambria.ttc(1)" at 10pt
 \endlisting
 
+and likewise, requesting subfont inside a TTC container by path:
+
+\beginlisting
+  \font \asanamain = "[/home/typesetter/.fonts/math/asana.ttc](0):mode=node;+tlig" at 10pt
+  \font \asanamath = "[/home/typesetter/.fonts/math/asana.ttc](1):mode=base" at 10pt
+\endlisting
+
 \endsubsubsection
 
 \beginsubsubsection{Loading by Font Name}
@@ -608,7 +641,7 @@ obviously, \inlinecode{random}.
            For scripts derived from the Latin alphabet the value
            \inlinecode{latn} is good choice.
          }
-         the default value is \inlinecode{dlft}.
+         the default value is \inlinecode{dflt}.
          %
          Some fonts, including very popular ones by foundries like Adobe,
          do not assign features to the \inlinecode{dflt} script, in
diff --git a/src/fontloader/misc/fontloader-font-gbn.lua b/src/fontloader/misc/fontloader-font-gbn.lua
index 1f8df64..21c6061 100644
--- a/src/fontloader/misc/fontloader-font-gbn.lua
+++ b/src/fontloader/misc/fontloader-font-gbn.lua
@@ -37,6 +37,7 @@ local getdisc     = nuts.getdisc
 local setchar     = nuts.setchar
 local setlink     = nuts.setlink
 local setprev     = nuts.setprev
+local nodetail    = nuts.tail
 
 -- from now on we apply ligaturing and kerning here because it might interfere with complex
 -- opentype discretionary handling where the base ligature pass expect some weird extra
@@ -200,21 +201,15 @@ function nodes.handlers.nodepass(head)
                 local stop  = range[2]
                 if start then
                     local front = nuthead == start
-                    local prev, next
+                    if not stop then
+                        stop = nodetail(start)
+                    end
                     if stop then
-                        next = getnext(stop)
                         start, stop = ligaturing(start,stop)
                         start, stop = kerning(start,stop)
-                    else
-                        prev  = getprev(start)
-                        start = ligaturing(start)
-                        start = kerning(start)
-                    end
-                    if prev then
-                        setlink(prev,start)
-                    end
-                    if next then
-                        setlink(stop,next)
+                    elseif start then -- safeguard
+                        start, stop = ligaturing(start,stop)
+                        start, stop = kerning(start,stop)
                     end
                     if front and nuthead ~= start then
                         head = tonode(start)
diff --git a/src/luaotfload-features.lua b/src/luaotfload-features.lua
index 5c35031..f11f05b 100644
--- a/src/luaotfload-features.lua
+++ b/src/luaotfload-features.lua
@@ -1274,9 +1274,17 @@ local tlig_specification = {
         type      = "substitution",
         features  = everywhere,
         data      = {
-            [0x0022] = 0x201D,                   -- quotedblright
-            [0x0027] = 0x2019,                   -- quoteleft
-            [0x0060] = 0x2018,                   -- quoteright
+            --- quotedblright:
+            --- " (QUOTATION MARK)   → ” (RIGHT DOUBLE QUOTATION MARK)
+            [0x0022] = 0x201D,
+
+            --- quoteleft:
+            --- ' (APOSTROPHE)       → ’ (RIGHT SINGLE QUOTATION MARK)
+            [0x0027] = 0x2019,
+
+            --- quoteright:
+            --- ` (GRAVE ACCENT)     → ‘ (LEFT SINGLE QUOTATION MARK)
+            [0x0060] = 0x2018,
         },
         flags     = noflags,
         order     = { "tlig" },
@@ -1286,13 +1294,43 @@ local tlig_specification = {
         type     = "ligature",
         features = everywhere,
         data     = {
-            [0x2013] = {0x002D, 0x002D},         -- endash
-            [0x2014] = {0x002D, 0x002D, 0x002D}, -- emdash
-            --- next three originate in T1 encoding; Xetex applies
-            --- them too
-            [0x201E] = {0x002C, 0x002C},         -- quotedblbase
-            [0x00AB] = {0x003C, 0x003C},         -- LEFT-POINTING DOUBLE ANGLE QUOTATION MARK
-            [0x00BB] = {0x003E, 0x003E},         -- RIGHT-POINTING DOUBLE ANGLE QUOTATION MARK
+
+            --- endash:
+            --- [--] (HYPHEN-MINUS, HYPHEN-MINUS)                   → – (EN DASH)
+            [0x2013] = {0x002D, 0x002D},
+
+            --- emdash:
+            --- [---] (HYPHEN-MINUS, HYPHEN-MINUS, HYPHEN-MINUS)    → — (EM DASH)
+            [0x2014] = {0x002D, 0x002D, 0x002D},
+
+            --- quotedblleft:
+            --- [''] (GRAVE ACCENT, GRAVE ACCENT)                   → “ (LEFT DOUBLE QUOTATION MARK)
+            [0x201C] = {0x0060, 0x0060},
+
+            --- quotedblright:
+            --- [``] (APOSTROPHE, APOSTROPHE)                       → ” (RIGHT DOUBLE QUOTATION MARK)
+            [0x201D] = {0x0027, 0x0027},
+
+            --- exclamdown:
+            --- [!'] (EXCLAMATION MARK, GRAVE ACCENT)               → ¡ (INVERTED EXCLAMATION MARK)
+            [0x00A1] = {0x0021, 0x0060},
+
+            --- questiondown:
+            --- [?'] (QUESTION MARK, GRAVE ACCENT)                  → ¡ (INVERTED EXCLAMATION MARK)
+            [0x00BF] = {0x003F, 0x0060},
+
+            --- next three originate in T1 encoding (Xetex applies them too)
+            --- quotedblbase:
+            --- [,,] (COMMA, COMMA)                                 → ¡ (DOUBLE LOW-9 QUOTATION MARK)
+            [0x201E] = {0x002C, 0x002C},
+
+            --- LEFT-POINTING DOUBLE ANGLE QUOTATION MARK:
+            --- [,,] (LESS-THAN SIGN, LESS-THAN SIGN)               → ¡ (LEFT-POINTING ANGLE QUOTATION MARK)
+            [0x00AB] = {0x003C, 0x003C},
+
+            --- RIGHT-POINTING DOUBLE ANGLE QUOTATION MARK:
+            --- [,,] (GREATER-THAN SIGN, GREATER-THAN SIGN)         → ¡ (RIGHT-POINTING DOUBLE ANGLE QUOTATION MARK)
+            [0x00BB] = {0x003E, 0x003E},
         },
         flags    = noflags,
         order    = { "tlig" },
diff --git a/src/luaotfload-loaders.lua b/src/luaotfload-loaders.lua
index 87275ef..802776c 100644
--- a/src/luaotfload-loaders.lua
+++ b/src/luaotfload-loaders.lua
@@ -151,6 +151,11 @@ do
       logreport ("both", 0, "loaders", "   > fontname %q", result.fontname or "<nil>")
       logreport ("both", 0, "loaders", "   > fullname %q", result.fullname or "<nil>")
       logreport ("both", 0, "loaders", "   > type %s",     result.type     or "<nil>")
+      local spec = result.specification
+      if spec then
+        logreport ("both", 0, "loaders", "   > file %q",     spec.filename or "<nil>")
+        logreport ("both", 0, "loaders", "   > subfont %s",  spec.sub      or "<nil>")
+      end
       return result
     end
   end
diff --git a/src/luaotfload-parsers.lua b/src/luaotfload-parsers.lua
index c2b4f2e..b188bb0 100644
--- a/src/luaotfload-parsers.lua
+++ b/src/luaotfload-parsers.lua
@@ -12,7 +12,8 @@ local traversal_maxdepth  = 42 --- prevent stack overflows
 local rawset            = rawset
 
 local lpeg              = require "lpeg"
-local P, R, S           = lpeg.P, lpeg.R, lpeg.S
+local patterns          = lpeg.patterns
+local P, R, S, V        = lpeg.P, lpeg.R, lpeg.S, lpeg.V
 local lpegmatch         = lpeg.match
 local C, Cc, Cf         = lpeg.C, lpeg.Cc, lpeg.Cf
 local Cg, Cmt, Cs, Ct   = lpeg.Cg, lpeg.Cmt, lpeg.Cs, lpeg.Ct
@@ -57,6 +58,7 @@ local equals            = P"="
 local dash              = P"-"
 local gartenzaun        = P"#"
 local lbrk, rbrk        = P"[", P"]"
+local brackets          = lbrk + rbrk
 local squote            = P"'"
 local dquote            = P"\""
 
@@ -602,6 +604,22 @@ local combodef1         = Ct(comboidx * combomapsep * comboid) --> no chars
 local combodef          = Ct(comboidx * combomapsep * comboidchars)
 local combolist         = Ct(combodef1 * (comborowsep * combodef)^1)
 
+--- subfonts ----------------------------------------------------------
+--- This rule is present in the original parser. It sets the “sub”
+--- field of the specification which allows addressing a specific
+--- font inside a TTC container. Neither in Luatex-Fonts nor in
+--- Luaotfload is this documented, so we might as well silently drop
+--- it. However, as backward compatibility is one of our prime goals we
+--- just insert it here and leave it undocumented until someone cares
+--- to ask. (Note: afair subfonts are numbered, but this rule matches a
+--- string; I won’t mess with it though until someone reports a
+--- problem.)
+--- local subvalue   = P("(") * (C(P(1-S("()"))^1)/issub) * P(")") -- for Kim
+--- Note to self: subfonts apparently start at index 0. Tested with
+--- Cambria.ttc that includes “Cambria Math” at 0 and “Cambria” at 1.
+--- Other values cause luatex to segfault.
+local subfont           = P"(" * Cg((1 - S"()")^1, "sub") * P")"
+
 --- lookups -----------------------------------------------------------
 local fontname          = C((1-S":(/")^1)  --- like luatex-fonts
 local unsupported       = Cmt((1-S":(")^1, check_garbage)
@@ -618,7 +636,19 @@ local prefixed          = P"name:" * ws * Cg(fontname, "name")
                         + P"kpse:" * ws * Cg(fontname, "kpse")
                         + P"my:" * ws * Cg(fontname, "my")
 local unprefixed        = Cg(fontname, "anon")
-local path_lookup       = lbrk * Cg(C((1-rbrk)^1), "path") * rbrk
+--- Bracketed “path” lookups: These may contain any character except
+--- for unbalanced brackets. A backslash escapes any following
+--- character. Everything inside the outermost brackets is treated as
+--- part of the filename or path to look up. Subfonts may be specified
+--- in parens *after* the closing bracket. Note that this differs from
+--- Xetex whose syntax expects the subfont passed inside the brackets,
+--- separated by a colon.
+local path_escape       = backslash / "" * patterns.utf8char
+local path_content      = path_escape + (1 - brackets)
+local path_balanced     = { (path_content + V(2))^1
+                          , lbrk * V(1)^-1 * rbrk }
+local path_lookup       = lbrk * Cg(Cs(path_balanced), "path") * rbrk
+                        * subfont^-1
 
 --- features ----------------------------------------------------------
 local field_char        = anum + S"+-.!?" --- sic!
@@ -644,22 +674,6 @@ local feature_list      = Cf(Ct""
                            * (featuresep * option^-1)^0
                            , rawset)
                         * featuresep^-1
-
---- other -------------------------------------------------------------
---- This rule is present in the original parser. It sets the “sub”
---- field of the specification which allows addressing a specific
---- font inside a TTC container. Neither in Luatex-Fonts nor in
---- Luaotfload is this documented, so we might as well silently drop
---- it. However, as backward compatibility is one of our prime goals we
---- just insert it here and leave it undocumented until someone cares
---- to ask. (Note: afair subfonts are numbered, but this rule matches a
---- string; I won’t mess with it though until someone reports a
---- problem.)
---- local subvalue   = P("(") * (C(P(1-S("()"))^1)/issub) * P(")") -- for Kim
---- Note to self: subfonts apparently start at index 0. Tested with
---- Cambria.ttc that includes “Cambria Math” at 0 and “Cambria” at 1.
---- Other values cause luatex to segfault.
-local subfont           = P"(" * Cg((1 - S"()")^1, "sub") * P")"
 --- top-level rules ---------------------------------------------------
 --- \font\foo=<specification>:<features>
 local features          = Cg(feature_list, "features")