Use explicit Word equation markers

src/ai_typewriter/math_format.py

@@ -9,36 +9,14 @@ GREEK_AND_SYMBOLS = {
     r"\gamma": "γ",
     r"\delta": "δ",
     r"\epsilon": "ε",
-    r"\varepsilon": "ε",
-    r"\zeta": "ζ",
-    r"\eta": "η",
     r"\theta": "θ",
-    r"\vartheta": "ϑ",
-    r"\iota": "ι",
-    r"\kappa": "κ",
     r"\lambda": "λ",
     r"\mu": "μ",
-    r"\nu": "ν",
-    r"\xi": "ξ",
     r"\pi": "π",
-    r"\rho": "ρ",
     r"\sigma": "σ",
-    r"\tau": "τ",
-    r"\upsilon": "υ",
     r"\phi": "φ",
-    r"\varphi": "φ",
-    r"\chi": "χ",
-    r"\psi": "ψ",
     r"\omega": "ω",
-    r"\Gamma": "Γ",
     r"\Delta": "Δ",
-    r"\Theta": "Θ",
-    r"\Lambda": "Λ",
-    r"\Xi": "Ξ",
-    r"\Pi": "Π",
-    r"\Sigma": "Σ",
-    r"\Phi": "Φ",
-    r"\Psi": "Ψ",
     r"\Omega": "Ω",
     r"\infty": "∞",
     r"\leq": "≤",
@@ -51,31 +29,11 @@ GREEK_AND_SYMBOLS = {
     r"\times": "×",
     r"\cdot": "·",
     r"\pm": "±",
-    r"\mp": "∓",
     r"\to": "→",
     r"\rightarrow": "→",
-    r"\leftarrow": "←",
-    r"\Rightarrow": "⇒",
-    r"\Leftrightarrow": "⇔",
-    r"\forall": "∀",
-    r"\exists": "∃",
-    r"\in": "∈",
-    r"\notin": "∉",
-    r"\subset": "⊂",
-    r"\subseteq": "⊆",
-    r"\cup": "∪",
-    r"\cap": "∩",
-    r"\emptyset": "∅",
-    r"\mathbb{R}": "ℝ",
-    r"\mathbb{C}": "ℂ",
-    r"\mathbb{N}": "ℕ",
-    r"\mathbb{Z}": "ℤ",
-    r"\mathbb{Q}": "ℚ",
     r"\int": "∫",
     r"\sum": "∑",
-    r"\prod": "∏",
-    r"\partial": "∂",
-    r"\nabla": "∇",
+    r"\sqrt": "√",
 }
 
 SUPERSCRIPT = str.maketrans({
@@ -94,18 +52,9 @@ SUBSCRIPT = str.maketrans({
     "s": "ₛ", "t": "ₜ", "u": "ᵤ", "v": "ᵥ", "x": "ₓ",
 })
 
-_GROUP_PATTERN = re.compile(r"([_^])\(([^()]+)\)|([_^])\{([^{}]+)\}|([_^])([A-Za-z0-9+\-=])")
-_MATH_START_PATTERN = re.compile(
-    r"(\\[A-Za-z]+|[A-Za-zΑ-Ωα-ω]+(?:[_^](?:\([^()]+\)|\{[^{}]+\}|[A-Za-z0-9+\-=]))+|[∫∑∞≤≥≠≈α-ωΑ-Ω]|=)"
-)
-_MATH_TOKEN_PATTERN = re.compile(
-    r"\s*(?:"
-    r"\\[A-Za-z]+(?:\([^()]+\)|\{[^{}]+\})?"
-    r"|[A-Za-zΑ-Ωα-ω0-9]+(?:[_^](?:\([^()]+\)|\{[^{}]+\}|[A-Za-z0-9+\-=]))*"
-    r"|[=+\-*/×·≤≥≠≈<>]"
-    r"|[(){}]"
-    r")"
-)
+_SCRIPT_PATTERN = re.compile(r"([_^])\(([^()]+)\)|([_^])\{([^{}]+)\}|([_^])([A-Za-z0-9+\-=])")
+EQUATION_OPEN = "[EQ]"
+EQUATION_CLOSE = "[/EQ]"
 
 
 @dataclass(frozen=True)
@@ -127,75 +76,54 @@ def _replace_script(match: re.Match[str]) -> str:
 
 
 def format_math_text(text: str, mode: str = "plain") -> str:
-    r"""Convert common UnicodeMath/LaTeX-like linear math to Unicode plain text.
-
-    This is meant for Word normal text insertion. It makes expressions such as
-    z_1, x^2, \alpha and \infty render as z₁, x², α and ∞ without requiring the
-    user to manually open Word's equation editor for every expression.
-    """
+    """Optional legacy text-only math formatting for non-Word targets."""
     if mode == "plain":
         return text
     if mode not in {"unicode", "unicode_math"}:
         raise ValueError("math_text_format doit être 'plain', 'unicode' ou 'word_equation'")
 
     formatted = text
-    # Replace longer commands first so \subseteq wins before \subset.
     for command, replacement in sorted(GREEK_AND_SYMBOLS.items(), key=lambda item: len(item[0]), reverse=True):
         formatted = formatted.replace(command, replacement)
 
-    formatted = re.sub(r"\\vec\(([^()]+)\)", lambda match: match.group(1) + "⃗", formatted)
-    formatted = formatted.replace(r"\sqrt", "√")
     previous = None
     while previous != formatted:
         previous = formatted
-        formatted = _GROUP_PATTERN.sub(_replace_script, formatted)
+        formatted = _SCRIPT_PATTERN.sub(_replace_script, formatted)
     return formatted
 
 
-def _consume_math_sequence(text: str, start: int) -> int:
-    end = start
-    while end < len(text):
-        if text[end] in "\n.,;:!?":
-            break
-        match = _MATH_TOKEN_PATTERN.match(text, end)
-        if match is None:
-            break
-        end = match.end()
-    return end
-
-
 def split_word_equation_segments(text: str) -> list[MathSegment]:
-    r"""Split text into normal text and Word UnicodeMath equation insertions.
+    """Split explicit [EQ]...[/EQ] blocks into Word equation actions.
 
-    Equation segments keep Word's linear UnicodeMath syntax (z_1, x^2, \alpha),
-    because Word converts that syntax inside the Alt+= equation editor.
+    No automatic math guessing is done here: the model decides where an equation
+    starts and ends by returning [EQ] before the expression and [/EQ] after it.
     """
     segments: list[MathSegment] = []
     cursor = 0
+
     while cursor < len(text):
-        match = _MATH_START_PATTERN.search(text, cursor)
-        if match is None:
+        start = text.find(EQUATION_OPEN, cursor)
+        if start == -1:
             if cursor < len(text):
                 segments.append(MathSegment("text", text[cursor:]))
             break
 
-        start = match.start()
         if start > cursor:
             segments.append(MathSegment("text", text[cursor:start]))
 
-        end = _consume_math_sequence(text, start)
-        expression = text[start:end].strip()
-        leading = text[start : start + len(text[start:end]) - len(text[start:end].lstrip())]
-        trailing = text[start:end][len(text[start:end].rstrip()):]
-        if leading:
-            segments.append(MathSegment("text", leading))
+        expression_start = start + len(EQUATION_OPEN)
+        end = text.find(EQUATION_CLOSE, expression_start)
+        if end == -1:
+            expression = text[expression_start:].strip()
+            cursor = len(text)
+        else:
+            expression = text[expression_start:end].strip()
+            cursor = end + len(EQUATION_CLOSE)
+
         if expression:
             segments.append(MathSegment("equation", expression))
-        if trailing:
-            segments.append(MathSegment("text", trailing))
-        cursor = end
 
-    # Merge adjacent text segments for cleaner action counts.
     merged: list[MathSegment] = []
     for segment in segments:
         if merged and segment.kind == "text" and merged[-1].kind == "text":