Validate regular expressions before reading input

src/Compiler.cpp

@@ -1,4 +1,5 @@
 #include "sedpp/Compiler.h"
+#include "sedpp/Regex.h"
 
 #include <cctype>
 #include <iostream>
@@ -103,6 +104,28 @@ Program Compiler::compile(const std::vector<std::string> &scripts) {
   if (braceDepth_ > 0) {
     throw std::runtime_error("-e expression #1, char 0: unmatched '{'");
   }
+
+  // Validate every literal regular expression before Runner opens or reads an
+  // input stream.  Besides matching GNU sed's diagnostic timing, this matters
+  // for portability probes such as `(sed non_gnu_re || sed fallback_re)`: the
+  // rejected first expression must leave the shared pipe untouched so the
+  // fallback process can consume it.  Empty expressions are runtime references
+  // to the previously used regex and therefore cannot be compiled here.
+  auto validateAddress = [this](const std::optional<Address> &address) {
+    if (address && address->kind == AddressKind::Regex &&
+        !address->text.empty()) {
+      Regex regex(address->text, options_);
+    }
+  };
+  for (const Command &command : program_.commands) {
+    validateAddress(command.firstAddress);
+    validateAddress(command.secondAddress);
+    if (command.substitute && !command.substitute->pattern.empty()) {
+      Regex regex(command.substitute->pattern, options_,
+                  command.substitute->ignoreCase,
+                  command.substitute->multiline);
+    }
+  }
   return program_;
 }
 

src/Options.cpp

@@ -7,7 +7,7 @@
 #include <cstdlib>
 
 namespace sedpp {
-int SEDPP_VERSION = 1;
+constexpr const char *SEDPP_VERSION = "1.0.1";
 
 void OptionParser::printVersion() {
   // Version output intentionally names this implementation, not GNU sed.