feat(math): Minimal support for accents in MathML and TeX-like commands

packages/math/atoms.lua

@@ -14,7 +14,7 @@ local atomType = {
    over = 8, -- Unused for now (used for overlines etc. in The TeXbook)
    under = 9, -- Unused for now (used for underlines etc. in The TeXbook)
    accent = 10,
-   botaccent = 11, -- Unused for now but botaccent is encoded in our dictionary
+   botaccent = 11,
 }
 
 return { types = atomType }

packages/math/base-elements.lua

@@ -732,6 +732,18 @@ local function isNotEmpty (element)
    return element and (element:is_a(elements.terminal) or #element.children > 0)
 end
 
+local function getAccentMode (mode)
+   -- Size unchanged but leave display mode
+   -- See MathML Core §3.4.3
+   if mode == mathMode.display then
+      return mathMode.text
+   end
+   if mode == mathMode.displayCramped then
+      return mathMode.textCramped
+   end
+   return mode
+end
+
 local function unwrapSingleElementMrow (elt)
    -- CODE SMELL.
    -- For \overset or \underset in LaTeX, MathML would use <mover> or <munder>.
@@ -748,10 +760,13 @@ local function unwrapSingleElementMrow (elt)
    end
 end
 
-function elements.underOver:_init (base, sub, sup)
+function elements.underOver:_init (attributes, base, sub, sup)
    elements.mbox._init(self)
    base = unwrapSingleElementMrow(base)
    self.atom = base.atom
+   self.attributes = attributes or {}
+   self.attributes.accent = SU.boolean(self.attributes.accent, false)
+   self.attributes.accentunder = SU.boolean(self.attributes.accentunder, false)
    self.base = base
    self.sub = isNotEmpty(sub) and sub or nil
    self.sup = isNotEmpty(sup) and sup or nil
@@ -771,10 +786,10 @@ function elements.underOver:styleChildren ()
       self.base.mode = self.mode
    end
    if self.sub then
-      self.sub.mode = getSubscriptMode(self.mode)
+      self.sub.mode = self.attributes.accentunder and getAccentMode(self.mode) or getSubscriptMode(self.mode)
    end
    if self.sup then
-      self.sup.mode = getSuperscriptMode(self.mode)
+      self.sup.mode = self.attributes.accent and getAccentMode(self.mode) or getSuperscriptMode(self.mode)
    end
 end
 
@@ -816,7 +831,10 @@ function elements.underOver:_stretchyReshapeToBase (part)
 end
 
 function elements.underOver:shape ()
+   local constants = self:getMathMetrics().constants
+   local scaleDown = self:getScaleDown()
    local isMovableLimits = SU.boolean(self.base and self.base.movablelimits, false)
+   local itCorr = self:calculateItalicsCorrection() * scaleDown
    if not (self.mode == mathMode.display or self.mode == mathMode.displayCramped) and isMovableLimits then
       -- When the base is a movable limit, the under/over scripts are not placed under/over the base,
       -- but other to the right of it, when display mode is not used.
@@ -827,32 +845,54 @@ function elements.underOver:shape ()
       elements.subscript.shape(self)
       return
    end
-   local constants = self:getMathMetrics().constants
-   local scaleDown = self:getScaleDown()
    -- Determine relative Ys
    if self.base then
       self.base.relY = SILE.types.length(0)
    end
    if self.sub then
       self:_stretchyReshapeToBase(self.sub)
-      self.sub.relY = self.base.depth
-         + SILE.types.length(
-            math.max(
-               (self.sub.height + constants.lowerLimitGapMin * scaleDown):tonumber(),
-               constants.lowerLimitBaselineDropMin * scaleDown
+      -- TODO These rules are incomplete and even wrong if we were to fully implement MathML Core.
+      if self.attributes.accentunder then
+         self.sub.relY = self.base.depth
+            + SILE.types.length(
+               (self.sub.height + constants.lowerLimitGapMin * scaleDown):tonumber()
+               -- We assume that the accent is aligned on the base.
             )
-         )
+      else
+         self.sub.relY = self.base.depth
+            + SILE.types.length(
+               math.max(
+                  (self.sub.height + constants.lowerLimitGapMin * scaleDown):tonumber(),
+                  constants.lowerLimitBaselineDropMin * scaleDown
+               )
+            )
+      end
    end
    if self.sup then
       self:_stretchyReshapeToBase(self.sup)
-      self.sup.relY = 0
-         - self.base.height
-         - SILE.types.length(
-            math.max(
-               (constants.upperLimitGapMin * scaleDown + self.sup.depth):tonumber(),
-               constants.upperLimitBaselineRiseMin * scaleDown
+      -- TODO These rules are incomplete if we were to fully implement MathML Core.
+      if self.attributes.accent then
+         self.sup.relY = 0 - self.base.height
+         -- MathML Core wants to align on the accentBaseHeight...
+         local overShift = math.max(0, constants.accentBaseHeight * scaleDown - self.base.height:tonumber())
+         self.sup.relY = self.sup.relY - SILE.types.length(overShift)
+         -- HACK: .... but improperly dimensioned accents can overshoot the base glyph.
+         -- So we try some guesswork to correct this.
+         -- Typically some non-combining symbols are in this case...
+         local heuristics = 0.5 * constants.flattenedAccentBaseHeight + 0.5 * constants.accentBaseHeight
+         if self.sup.height > SILE.types.length(heuristics * scaleDown) then
+            self.sup.relY = self.sup.relY + SILE.types.length(constants.accentBaseHeight * scaleDown)
+         end
+      else
+         self.sup.relY = 0
+            - self.base.height
+            - SILE.types.length(
+               math.max(
+                  (constants.upperLimitGapMin * scaleDown + self.sup.depth):tonumber(),
+                  constants.upperLimitBaselineRiseMin * scaleDown
+               )
             )
-         )
+      end
    end
    -- Determine relative Xs based on widest symbol
    local widest, a, b
@@ -893,7 +933,6 @@ function elements.underOver:shape ()
    if b then
       b.relX = c - b.width / 2
    end
-   local itCorr = self:calculateItalicsCorrection() * scaleDown
    if self.sup then
       self.sup.relX = self.sup.relX + itCorr / 2
    end
@@ -1201,7 +1240,10 @@ end
 function elements.text:_vertStretchyReshape (depth, height)
    local hasStretched = self:_stretchyReshape(depth + height, true)
    if hasStretched then
-      -- HACK: see output routine
+      -- RESCALING HACK: see output routine
+      -- We only do it if the scaling logic found constructions on the vertical block axis.
+      -- It's a dirty hack until we properly implement assembly of glyphs in the case we couldn't
+      -- find a big enough variant.
       self.vertExpectedSz = height + depth
       self.vertScalingRatio = (depth + height):tonumber() / (self.height:tonumber() + self.depth:tonumber())
       self.height = height
@@ -1212,12 +1254,21 @@ end
 
 function elements.text:_horizStretchyReshape (width)
    local hasStretched = self:_stretchyReshape(width, false)
-   if hasStretched then
-      -- HACK: see output routine
-      self.horizScalingRatio = width:tonumber() / self.width:tonumber()
-      self.width = width
-   end
-   return hasStretched
+   if not hasStretched and width:tonumber() < self.width:tonumber() then
+      -- Never shrink glyphs, it looks ugly
+      return false
+   end
+   -- But if stretching couldn't be done, it will be ugly anyway, so we will force
+   -- a re-scaling of the glyph.
+   -- (So it slightly different from the vertical case, 'cause MathML just has one stretchy
+   -- attribute, whether for stretching on the vertical (block) or horizontal (inline) axis,
+   -- and we cannot know which axis is meant unless we implement yet another mapping table
+   -- as the one in the MathML Core appendices. Frankly, how many non-normative appendices
+   -- do we need to implement MathML correctly?)
+   -- RESCALING HACK: see output routine
+   self.horizScalingRatio = width:tonumber() / self.width:tonumber()
+   self.width = width
+   return true
 end
 
 function elements.text:output (x, y, line)
@@ -1356,7 +1407,7 @@ local function newSubscript (spec)
 end
 
 local function newUnderOver (spec)
-   return elements.underOver(spec.base, spec.sub, spec.sup)
+   return elements.underOver(spec.attributes, spec.base, spec.sub, spec.sup)
 end
 
 -- TODO replace with penlight equivalent

packages/math/init.lua

@@ -427,7 +427,7 @@ The \code{counter} or the direct value \code{number} is passed as a parameter to
 
 \paragraph{Missing features}
 This package still lacks support for some mathematical constructs, but hopefully we’ll get there.
-Among unsupported constructs are: decorating symbols with so-called accents, such as arrows or hats, “over” or “under” braces, and line breaking inside a formula.
+Among unsupported features, we can mention line breaking inside a formula.
 
 \font:remove-fallback
 \end{document}

packages/math/texlike.lua

@@ -396,9 +396,24 @@ local function isOperatorKind (tree, typeOfAtom)
    return false
 end
 
-local function isMoveableLimits (tree)
+local function isMoveableLimitsOrAlwaysStacked (tree)
+   if not tree then
+      return false -- safeguard
+   end
+   if tree.is_always_stacked then
+      -- We use an internal flag to mark commands that are always stacking
+      -- their sup/sub arguments, such as brace-like commands.
+      return true
+   end
    if tree.command ~= "mo" then
-      return false
+      -- On the recursion:
+      -- MathML allows movablelimits on <mo> elements, but "embellished operators"
+      -- can be other elements inheriting the property from their "core operator",
+      -- see MathML Core §3.2.4.1, which is full of intricacies so we are probably
+      -- not even doing the right thing here.
+      -- On the hack:
+      -- See variant commands for limits further down.
+      return SU.boolean(tree.is_hacked_movablelimits, false) or isMoveableLimitsOrAlwaysStacked(tree[1])
    end
    if tree.options and SU.boolean(tree.options.movablelimits, false) then
       return true
@@ -430,6 +445,9 @@ end
 local function isAccentSymbol (symbol)
    return operatorDict[symbol] and operatorDict[symbol].atom == atoms.types.accent
 end
+local function isBottomAccentSymbol (symbol)
+   return operatorDict[symbol] and operatorDict[symbol].atom == atoms.types.botaccent
+end
 
 local function compileToMathML_aux (_, arg_env, tree)
    if type(tree) == "string" then
@@ -565,14 +583,15 @@ local function compileToMathML_aux (_, arg_env, tree)
       end
       tree.options = {}
    -- Translate TeX-like sub/superscripts to `munderover` or `msubsup`,
-   -- depending on whether the base is an operator with moveable limits.
-   elseif tree.id == "sup" and isMoveableLimits(tree[1]) then
+   -- depending on whether the base is an operator with moveable limits,
+   -- or a brace-like command.
+   elseif tree.id == "sup" and isMoveableLimitsOrAlwaysStacked(tree[1]) then
       tree.command = "mover"
-   elseif tree.id == "sub" and isMoveableLimits(tree[1]) then
+   elseif tree.id == "sub" and isMoveableLimitsOrAlwaysStacked(tree[1]) then
       tree.command = "munder"
-   elseif tree.id == "subsup" and isMoveableLimits(tree[1]) then
+   elseif tree.id == "subsup" and isMoveableLimitsOrAlwaysStacked(tree[1]) then
       tree.command = "munderover"
-   elseif tree.id == "supsub" and isMoveableLimits(tree[1]) then
+   elseif tree.id == "supsub" and isMoveableLimitsOrAlwaysStacked(tree[1]) then
       tree.command = "munderover"
       local tmp = tree[2]
       tree[2] = tree[3]
@@ -638,7 +657,7 @@ local function compileToMathML_aux (_, arg_env, tree)
    elseif tree.id == "command" and symbols[tree.command] then
       local atom = { id = "atom", [1] = symbols[tree.command] }
       if isAccentSymbol(symbols[tree.command]) and #tree > 0 then
-         -- LaTeX-style accents \vec{v} = <mover accent="true"><mi>v</mi><mo>→</mo></mover>
+         -- LaTeX-style accents \overrightarrow{v} = <mover accent="true"><mi>v</mi><mo>&#x20D7;</mo></mover>
          local accent = {
             id = "command",
             command = "mover",
@@ -649,6 +668,18 @@ local function compileToMathML_aux (_, arg_env, tree)
          accent[1] = compileToMathML_aux(nil, arg_env, tree[1])
          accent[2] = compileToMathML_aux(nil, arg_env, atom)
          tree = accent
+      elseif isBottomAccentSymbol(symbols[tree.command]) and #tree > 0 then
+         -- LaTeX-style bottom accents \underleftarrow{v} = <munder accent="true"><mi>v</mi><mo>&#x20EE;</mo></munder>
+         local accent = {
+            id = "command",
+            command = "munder",
+            options = {
+               accentunder = "true",
+            },
+         }
+         accent[1] = compileToMathML_aux(nil, arg_env, tree[1])
+         accent[2] = compileToMathML_aux(nil, arg_env, atom)
+         tree = accent
       elseif #tree > 0 then
          -- Play cool with LaTeX-style commands that don't take arguments:
          -- Edge case for non-accent symbols so we don't loose bracketed groups
@@ -728,6 +759,80 @@ registerCommand("mn", { [1] = objType.str }, function (x)
    return x
 end)
 
+-- Register a limit-like variant command
+-- Variants of superior, inferior, projective and injective limits are special:
+-- They accept a sub/sup behaving as a movablelimits, but also have a symbol
+-- on top of the limit symbol, which is not a movablelimits.
+-- I can't see in the MathML specification how to do this properly: MathML Core
+-- seems to only allow movablelimits on <mo> elements, and <mover>/<munder> may
+-- inherit that property from their "core operator", but in this case we do not
+-- want the accent to be movable, only the limit sup/sub.
+-- So we use a hack, and also avoid "\def" here to prevent unwanted mrows.
+-- @tparam string name    TeX command name
+-- @tparam string command MathML command (mover or munder)
+-- @tparam number symbol  Unicode codepoint for the accent symbol
+-- @tparam string text    Text representation
+local function registerVarLimits (name, command, symbol, text)
+   registerCommand(name, {}, function ()
+      local options = command == "mover" and { accent = "true" } or { accentunder = "true" }
+      return {
+         command = command,
+         is_hacked_movablelimits = true, -- Internal flag to mark this as a hack
+         options = options,
+         {
+            command = "mo",
+            options = { atom = "op", movablelimits = false },
+            text,
+         },
+         {
+            command = "mo",
+            options = { accentunder = "true" },
+            luautf8.char(symbol),
+         },
+      }
+   end)
+end
+registerVarLimits("varlimsup", "mover", 0x203E, "lim") -- U+203E OVERLINE
+registerVarLimits("varliminf", "munder", 0x203E, "lim") -- U+203E OVERLINE
+registerVarLimits("varprojlim", "munder", 0x2190, "lim") -- U+2190 LEFTWARDS ARROW
+registerVarLimits("varinjlim", "munder", 0x2192, "lim") -- U+2192 RIGHTWARDS ARROW
+
+-- Register a brace-like commands.
+-- Those symbols are accents per-se in MathML, and are non-combining in Unicode.
+-- But TeX treats them as "pseudo-accent" stretchy symbols.
+-- Moreover, they accept a sub/sup which is always stacked, and not movable.
+-- So we use an internal flag.
+-- We also avoid "\def" here to prevent unwanted mrows resulting from the
+-- compilation of the argument.
+-- @tparam string name    TeX command name
+-- @tparam string command MathML command (mover or munder)
+-- @tparam number symbol  Unicode codepoint for the brace symbol
+local function registerBraceLikeCommands (name, command, symbol)
+   registerCommand(name, {
+      [1] = objType.tree,
+   }, function (tree)
+      local options = command == "mover" and { accent = "true" } or { accentunder = "true" }
+      return {
+         command = command,
+         is_always_stacked = true, -- Internal flag to mark this as a brace-like command
+         options = options,
+         tree[1],
+         {
+            command = "mo",
+            options = { stretchy = "true" },
+            luautf8.char(symbol),
+         },
+      }
+   end)
+end
+-- Note: the following overriddes the default commands from xml-entities / unicode-math.
+registerBraceLikeCommands("overbrace", "mover", 0x23DE) -- U+23DE TOP CURLY BRACKET
+registerBraceLikeCommands("underbrace", "munder", 0x23DF) -- U+23DF BOTTOM CURLY BRACKET
+registerBraceLikeCommands("overparen", "mover", 0x23DC) -- U+23DC TOP PARENTHESIS
+registerBraceLikeCommands("underparen", "munder", 0x23DD) -- U+23DD BOTTOM PARENTHESIS
+registerBraceLikeCommands("overbracket", "mover", 0x23B4) -- U+23B4 TOP SQUARE BRACKET
+registerBraceLikeCommands("underbracket", "munder", 0x23B5) -- U+23B5 BOTTOM SQUARE BRACKET
+
 compileToMathML(
    nil,
    {},
@@ -737,7 +842,6 @@ compileToMathML(
   \def{sqrt}{\msqrt{#1}}
   \def{bi}{\mi[mathvariant=bold-italic]{#1}}
   \def{dsi}{\mi[mathvariant=double-struck]{#1}}
-  \def{vec}{\mover[accent=true]{#1}{\rightarrow}}
 
   % From amsmath:
   \def{to}{\mo[atom=bin]{→}}