As described in the PR, unless the UTF-8 encoding of \ufeff is padded with a NUL byte, the iconv_init test fails for the native Tru64 UNIX libiconv, iconv_byte_swap isn't set correctly and later iconv calls fail. This patch fixes this, allowing the arraycopy testcase in the PR and the actual md5test and shatest testcases to succeed. This way, I get clean libjava testsuite results.
Tested by rebuilding libgcj.so with the fixed natIconv.cc. A full rebuild and test of libgcj is currently running on alpha-dec-osf5.1b. Ok for mainline if it passes? Thanks. Rainer 2011-06-21 Rainer Orth <r...@cebitec.uni-bielefeld.de> PR libgcj/49314 * gnu/gcj/convert/natIconv.cc (iconv_init): Pad in[] with NUL byte. diff --git a/libjava/gnu/gcj/convert/natIconv.cc b/libjava/gnu/gcj/convert/natIconv.cc --- a/libjava/gnu/gcj/convert/natIconv.cc +++ b/libjava/gnu/gcj/convert/natIconv.cc @@ -1,6 +1,6 @@ // natIconv.cc -- Java side of iconv() reader. -/* Copyright (C) 2000, 2001, 2003, 2006 Free Software Foundation +/* Copyright (C) 2000, 2001, 2003, 2006, 2011 Free Software Foundation This file is part of libgcj. @@ -264,17 +264,19 @@ gnu::gcj::convert::IOConverter::iconv_in if (handle != (iconv_t) -1) { jchar c; - unsigned char in[3]; + unsigned char in[4]; char *inp, *outp; size_t inc, outc, r; - // This is the UTF-8 encoding of \ufeff. + // This is the UTF-8 encoding of \ufeff. At least Tru64 UNIX libiconv + // needs the trailing NUL byte, otherwise iconv fails with EINVAL. in[0] = 0xef; in[1] = 0xbb; in[2] = 0xbf; + in[3] = 0x00; inp = (char *) in; - inc = 3; + inc = 4; outp = (char *) &c; outc = 2; -- ----------------------------------------------------------------------------- Rainer Orth, Center for Biotechnology, Bielefeld University