[src] Make arpa2fst robust against ARPA files without <s>.

.gitignore

@@ -1,6 +1,7 @@
 # Compiled extensionless executable files in /src/*/
 # This stanza must precede wildcard patterns below!
 /src/*/*
+!/src/lm/test_data/
 !/src/*/?*.*
 !/src/doc/*
 !/src/*/Makefile

src/lm/arpa-lm-compiler-test.cc

@@ -204,6 +204,17 @@ bool ScoringTest(bool seps, const string &infile, const string& sentence,
   return ok;
 }
 
+bool ThrowsExceptionTest(bool seps, const string &infile) {
+  try {
+    // Make memory cleanup easy in both cases of try-catch block.
+    std::unique_ptr<ArpaLmCompiler> compiler(Compile(seps, infile));
+    return false;
+  } catch (const std::runtime_error&) {
+    // Kaldi throws only std::runtime_error in kaldi-error.cc
+    return true;
+  }
+}
+
 }  // namespace kaldi
 
 bool RunAllTests(bool seps) {
@@ -214,6 +225,9 @@ bool RunAllTests(bool seps) {
 
   ok &= kaldi::ScoringTest(seps, "test_data/input.arpa", "b b b a", 59.2649);
   ok &= kaldi::ScoringTest(seps, "test_data/input.arpa", "a b", 4.36082);
+
+  ok &= kaldi::ThrowsExceptionTest(seps, "test_data/missing_bos.arpa");
+
   if (!ok) {
     KALDI_WARN << "Tests " << (seps ? "with" : "without")
                << " epsilon substitution FAILED";

src/lm/arpa-lm-compiler.cc

@@ -360,10 +360,18 @@ void ArpaLmCompiler::RemoveRedundantStates() {
             << fst_.NumStates();
 }
 
+void ArpaLmCompiler::Check() const {
+  if (fst_.Start() == fst::kNoStateId) {
+    KALDI_ERR << "Arpa file did not contain the beginning-of-sentence symbol "
+              << Symbols()->Find(Options().bos_symbol) << ".";
+  }
+}
+
 void ArpaLmCompiler::ReadComplete() {
   fst_.SetInputSymbols(Symbols());
   fst_.SetOutputSymbols(Symbols());
   RemoveRedundantStates();
+  Check();
 }
 
 }  // namespace kaldi

src/lm/arpa-lm-compiler.h

@@ -52,6 +52,7 @@ class ArpaLmCompiler : public ArpaFileParser {
   // this function removes states that only have a backoff arc coming
   // out of them.
   void RemoveRedundantStates();
+  void Check() const;
 
   int sub_eps_;
   ArpaLmCompilerImplInterface* impl_;  // Owned.

src/lm/test_data/missing_bos.arpa

@@ -0,0 +1,18 @@
+
+\data\
+ngram 1=3
+ngram 2=1
+ngram 3=1
+
+\1-grams:
+-5.234679	a -3.3
+-3.456783	b -3.0
+-4.333333	</s>
+
+\2-grams:
+-1.45678	a b -3.23
+
+\3-grams:
+-0.23940	a b </s>
+
+\end\