use safer text2mecab (r9y9#10)

Hiroshiba · Apr 16, 2022 · 07f242e · 07f242e
1 parent a85521a
commit 07f242e
Show file tree

Hide file tree

Showing 3 changed files with 10 additions and 3 deletions.
diff --git a/lib/open_jtalk b/lib/open_jtalk
diff --git a/pyopenjtalk/openjtalk.pyx b/pyopenjtalk/openjtalk.pyx
@@ -2,6 +2,7 @@
 # cython: boundscheck=True, wraparound=True
 # cython: c_string_type=unicode, c_string_encoding=ascii
 
+import errno
 import numpy as np
 
 cimport numpy as np
@@ -165,7 +166,13 @@ cdef class OpenJTalk(object):
         if isinstance(text, str):
           text = text.encode("utf-8")
         cdef char buff[8192]
-        text2mecab(buff, text)
+        cdef int result = text2mecab(buff, 8192, text)
+        if result != 0:
+            if result == errno.ERANGE:
+                raise RuntimeError("Text is too long")
+            if result == errno.EINVAL:
+                raise RuntimeError("Invalid input for text2mecab")
+            raise RuntimeError("Unknown error: " + str(result))
         Mecab_analysis(self.mecab, buff)
         mecab2njd(self.njd, Mecab_get_feature(self.mecab), Mecab_get_size(self.mecab))
         _njd.njd_set_pronunciation(self.njd)

diff --git a/pyopenjtalk/openjtalk/text2mecab.pxd b/pyopenjtalk/openjtalk/text2mecab.pxd
@@ -1,4 +1,4 @@
 # distutils: language = c++
 
 cdef extern from "text2mecab.h":
-    void text2mecab(char *output, const char *input)
+    int text2mecab(char *output, size_t sizeOfOutput, const char *input)
+1 −1		src/CMakeLists.txt
+4 −1		src/bin/open_jtalk.c
+1 −1		src/mecab/src/string_buffer.h
+13 −1		src/text2mecab/text2mecab.c
+1 −1		src/text2mecab/text2mecab.h