Implement shellSplitString for proper handling of NIX_SSHOPTS with spaces and quotes

src/libutil/strings.cc

@@ -4,6 +4,7 @@
 
 #include "strings-inline.hh"
 #include "os-string.hh"
+#include "error.hh"
 
 namespace nix {
 
@@ -48,4 +49,107 @@ template std::string dropEmptyInitThenConcatStringsSep(std::string_view, const s
 template std::string dropEmptyInitThenConcatStringsSep(std::string_view, const std::set<std::string> &);
 template std::string dropEmptyInitThenConcatStringsSep(std::string_view, const std::vector<std::string> &);
 
+/**
+ * Shell split string: split a string into shell arguments, respecting quotes and backslashes.
+ *
+ * Used for NIX_SSHOPTS handling, which previously used `tokenizeString` and was broken by
+ * Arguments that need to be passed to ssh with spaces in them.
+ *
+ * Read https://pubs.opengroup.org/onlinepubs/9699919799/utilities/V3_chap02.html for the
+ * POSIX shell specification, which is technically what we are implementing here.
+ */
+std::list<std::string> shellSplitString(std::string_view s)
+{
+    std::list<std::string> result;
+    std::string current;
+    bool startedCurrent = false;
+    bool escaping = false;
+
+    auto pushCurrent = [&]() {
+        if (startedCurrent) {
+            result.push_back(current);
+            current.clear();
+            startedCurrent = false;
+        }
+    };
+
+    auto pushChar = [&](char c) {
+        current.push_back(c);
+        startedCurrent = true;
+    };
+
+    auto pop = [&]() {
+        auto c = s[0];
+        s.remove_prefix(1);
+        return c;
+    };
+
+    auto inDoubleQuotes = [&]() {
+        startedCurrent = true;
+        // in double quotes, escaping with backslash is only effective for $, `, ", and backslash
+        while (!s.empty()) {
+            auto c = pop();
+            if (escaping) {
+                switch (c) {
+                case '$':
+                case '`':
+                case '"':
+                case '\\':
+                    pushChar(c);
+                    break;
+                default:
+                    pushChar('\\');
+                    pushChar(c);
+                    break;
+                }
+                escaping = false;
+            } else if (c == '\\') {
+                escaping = true;
+            } else if (c == '"') {
+                return;
+            } else {
+                pushChar(c);
+            }
+        }
+        if (s.empty()) {
+            throw Error("unterminated double quote");
+        }
+    };
+
+    auto inSingleQuotes = [&]() {
+        startedCurrent = true;
+        while (!s.empty()) {
+            auto c = pop();
+            if (c == '\'') {
+                return;
+            }
+            pushChar(c);
+        }
+        if (s.empty()) {
+            throw Error("unterminated single quote");
+        }
+    };
+
+    while (!s.empty()) {
+        auto c = pop();
+        if (escaping) {
+            pushChar(c);
+            escaping = false;
+        } else if (c == '\\') {
+            escaping = true;
+        } else if (c == ' ' || c == '\t') {
+            pushCurrent();
+        } else if (c == '"') {
+            inDoubleQuotes();
+        } else if (c == '\'') {
+            inSingleQuotes();
+        } else {
+            pushChar(c);
+        }
+    }
+
+    pushCurrent();
+
+    return result;
+}
 } // namespace nix

src/libutil/strings.hh

@@ -71,4 +71,11 @@ extern template std::string dropEmptyInitThenConcatStringsSep(std::string_view,
 extern template std::string dropEmptyInitThenConcatStringsSep(std::string_view, const std::set<std::string> &);
 extern template std::string dropEmptyInitThenConcatStringsSep(std::string_view, const std::vector<std::string> &);
 
+/**
+ * Shell split string: split a string into shell arguments, respecting quotes and backslashes.
+ *
+ * Used for NIX_SSHOPTS handling, which previously used `tokenizeString` and was broken by
+ * Arguments that need to be passed to ssh with spaces in them.
+ */
+std::list<std::string> shellSplitString(std::string_view s);
 }