From 2a34b90d1e47dfdcdf513f0bfc3baa914baaffd8 Mon Sep 17 00:00:00 2001
From: Bryce Kwon <bryce@brycekwon.com>
Date: Fri, 16 Oct 2026 18:04:13 -1000
Subject: Keep underscores inside words out of emphasis

---
 custom/extensions/about-render.lua     | 13 +++++++++----
 tests/extensions/test-about-render.lua |  6 ++++++
 2 files changed, 15 insertions(+), 4 deletions(-)

diff --git a/custom/extensions/about-render.lua b/custom/extensions/about-render.lua
index 4e4b50b..73b1336 100644
--- a/custom/extensions/about-render.lua
+++ b/custom/extensions/about-render.lua
@@ -291,7 +291,7 @@ end
 local render_markdown, render_man
 
 if has_lpeg then
-	local P, R, S, C, Ct = lpeg.P, lpeg.R, lpeg.S, lpeg.C, lpeg.Ct
+	local P, R, S, B, C, Ct = lpeg.P, lpeg.R, lpeg.S, lpeg.B, lpeg.C, lpeg.Ct
 
 	local space = S(" \t")
 	local whitespace = S(" \t\r\n\f\v")
@@ -345,6 +345,11 @@ if has_lpeg then
 	local address = (alnum + S(".!#$%&'*+/=?^_`{|}~-")) ^ 1
 		* P("@") * alnum * (alnum + S(".-")) ^ 0
 
+	-- An underscore inside a word belongs to the word, as in snake_case,
+	-- so it neither opens nor closes emphasis.
+	local word = alnum + R("\128\255")
+	local inner = B(word) * P("_") ^ 1 * #word
+
 	-- A character reference passes through for the browser to decode,
 	-- within CommonMark's lengths. Its shape leaves no room for markup, and
 	-- a name the browser does not know shows as written.
@@ -370,10 +375,10 @@ if has_lpeg then
 		+ (P("!") * label * target) / node_image
 		+ (label * target) / node_link
 		+ (P("**") * C((1 - P("**")) ^ 1) * P("**")) / node_strong
-		+ (P("__") * C((1 - P("__")) ^ 1) * P("__")) / node_strong
+		+ (-B(word) * P("__") * C((inner + (1 - P("__"))) ^ 1) * P("__") * -word) / node_strong
 		+ (P("*") * C((1 - whitespace) * (1 - P("*")) ^ 0) * P("*")) / node_em
-		+ (P("_") * C((1 - whitespace) * (1 - P("_")) ^ 0) * P("_")) / node_em
-		+ C((1 - marker) ^ 1) / node_text
+		+ (-B(word) * P("_") * C((1 - whitespace) * (inner + (1 - P("_"))) ^ 0) * P("_") * -word) / node_em
+		+ C((1 - marker + inner) ^ 1) / node_text
 		+ C(P(1)) / node_text
 	) ^ 0)
 
diff --git a/tests/extensions/test-about-render.lua b/tests/extensions/test-about-render.lua
index e84c854..3925dfd 100644
--- a/tests/extensions/test-about-render.lua
+++ b/tests/extensions/test-about-render.lua
@@ -108,6 +108,12 @@ out = render("README.md", "an __up__ and _down_ word")
 h.contains("double underscores embolden", out, "<strong>up</strong>")
 h.contains("single underscores lean", out, "<em>down</em>")
 
+out = render("README.md", "snake_case_name and _lean_ and foo__bar__")
+h.contains("underscores inside words stay literal", out, "<p>snake_case_name and <em>lean</em> and foo__bar__</p>")
+
+out = render("README.md", "_a_b_ and __c_d__")
+h.contains("emphasis spans an underscore inside a word", out, "<p><em>a_b</em> and <strong>c_d</strong></p>")
+
 out = render("README.md", "\\*not leaning\\* and \\# and \\q")
 h.contains("a backslash makes punctuation literal", out, "<p>*not leaning* and # and \\q</p>")
 
-- 
cgit