From 04e63a5b3ebcb8ff64a327575d3ec5c11f98a1df Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Daniel=20Mart=C3=AD?= <mvdan@mvdan.cc>
Date: Sun, 29 Sep 2024 14:01:41 +0200
Subject: [PATCH] syntax: parse all CRLF line endings as LF

We treat CR as whitespace, so this already happened in practice
for regular line endings. However, since heredocs are a quoted state
where white space is not skipped, the CRLF line endings caused errors
where we couldn't match `<<EOF` with a CRLF EOF line as we got `EOF\r`.

Fixes #1088.
---
 syntax/filetests_test.go | 17 ++++++++---------
 syntax/lexer.go          | 15 ++++++++++-----
 2 files changed, 18 insertions(+), 14 deletions(-)

diff --git a/syntax/filetests_test.go b/syntax/filetests_test.go
index f33ef951..2ae0a49c 100644
--- a/syntax/filetests_test.go
+++ b/syntax/filetests_test.go
@@ -932,6 +932,7 @@ var fileTests = []testCase{
 			"foo <<EOF\nbar\nEOF",
 			"foo <<EOF \nbar\nEOF",
 			"foo <<EOF\t\nbar\nEOF",
+			"foo <<EOF\r\nbar\r\nEOF\r\n",
 		},
 		common: &Stmt{
 			Cmd: litCall("foo"),
@@ -1389,6 +1390,7 @@ var fileTests = []testCase{
 		Strs: []string{
 			"foo <<'EOF'\nbar\\\nEOF",
 			"foo <<'EOF'\nbar\\\r\nEOF",
+			"foo <<'EOF'\nbar\\\r\nEOF\r\n",
 		},
 		common: &Stmt{
 			Cmd: litCall("foo"),
@@ -1400,7 +1402,10 @@ var fileTests = []testCase{
 		},
 	},
 	{
-		Strs: []string{"foo <<-EOF\n\tbar\nEOF"},
+		Strs: []string{
+			"foo <<-EOF\n\tbar\nEOF",
+			"foo <<-EOF\r\n\tbar\r\nEOF\r\n",
+		},
 		common: &Stmt{
 			Cmd: litCall("foo"),
 			Redirs: []*Redirect{{
@@ -4592,18 +4597,12 @@ func recursiveSanityCheck(tb testing.TB, src string, v any) {
 			if i == 0 {
 				gotErr = got
 			}
+			got = strings.ReplaceAll(got, "\x00", "")
+			got = strings.ReplaceAll(got, "\r\n", "\n")
 			if !strings.Contains(want, "\\\n") {
 				// Hack to let "foobar" match the input "foo\\\nbar".
 				got = strings.ReplaceAll(got, "\\\n", "")
-			} else {
-				// Hack to let "\\\n" match the input "\\\r\n".
-				got = strings.ReplaceAll(got, "\\\r\n", "\\\n")
-			}
-			if !strings.Contains(want, "\\\r\n") {
-				// Hack to let "foobar" match the input "foo\\\r\nbar".
-				got = strings.ReplaceAll(got, "\\\r\n", "")
 			}
-			got = strings.ReplaceAll(got, "\x00", "")
 			if strings.HasPrefix(got, want) {
 				return
 			}
diff --git a/syntax/lexer.go b/syntax/lexer.go
index 9fa4d14d..b90e8e1c 100644
--- a/syntax/lexer.go
+++ b/syntax/lexer.go
@@ -70,18 +70,24 @@ retry:
 	if p.bsp < uint(len(p.bs)) {
 		if b := p.bs[p.bsp]; b < utf8.RuneSelf {
 			p.bsp++
-			if b == '\x00' {
+			switch b {
+			case '\x00':
 				// Ignore null bytes while parsing, like bash.
 				p.col++
 				goto retry
-			}
-			if b == '\\' {
+			case '\r':
+				if p.peekByte('\n') { // \r\n turns into \n
+					p.col++
+					goto retry
+				}
+			case '\\':
 				if p.r == '\\' {
 				} else if p.peekByte('\n') {
 					p.bsp++
 					p.w, p.r = 1, escNewl
 					return escNewl
-				} else if p.peekBytes("\r\n") {
+				} else if p.peekBytes("\r\n") { // \\\r\n turns into \\\n
+					p.col++
 					p.bsp += 2
 					p.w, p.r = 2, escNewl
 					return escNewl
@@ -90,7 +96,6 @@ retry:
 					p.bsp < uint(len(p.bs)) && bquoteEscaped(p.bs[p.bsp]) {
 					// We turn backquote command substitutions into $(),
 					// so we remove the extra backslashes needed by the backquotes.
-					// For good position information, we still include them in p.w.
 					bquotes++
 					p.col++
 					goto retry