diff --git a/benchmark/string_coderange_scan.yml b/benchmark/string_coderange_scan.yml new file mode 100644 index 00000000000000..2fca47362804de --- /dev/null +++ b/benchmark/string_coderange_scan.yml @@ -0,0 +1,26 @@ +prelude: | + ascii1 = [*"a".."m",*"N".."Z",*"0".."9"].join("") + ascii10 = ascii1 * 10 + ascii100 = ascii10 * 10 + ascii1000 = ascii100 * 10 + + utf81 = [*"a".."m",*"N".."Z", "éà€‹›Ç☃"].join("") + utf810 = utf81 * 10 + utf8100 = utf810 * 10 + utf81000 = utf8100 * 10 + + invalid_ascii_last = ascii1000 + "☃" + invalid_ascii_first = "☃" + ascii1000 +benchmark: + ascii-coderange-scan-1: ascii1.force_encoding(Encoding::UTF_8).valid_encoding? + ascii-coderange-scan-10: ascii10.force_encoding(Encoding::UTF_8).valid_encoding? + ascii-coderange-scan-100: ascii100.force_encoding(Encoding::UTF_8).valid_encoding? + ascii-coderange-scan-1000: ascii1000.force_encoding(Encoding::UTF_8).valid_encoding? + + utf8-coderange-scan-1: utf81.force_encoding(Encoding::UTF_8).valid_encoding? + utf8-coderange-scan-10: utf810.force_encoding(Encoding::UTF_8).valid_encoding? + utf8-coderange-scan-100: utf8100.force_encoding(Encoding::UTF_8).valid_encoding? + utf8-coderange-scan-1000: utf81000.force_encoding(Encoding::UTF_8).valid_encoding? + + ascii-coderange-scan-first-1: invalid_ascii_first.force_encoding(Encoding::ASCII).valid_encoding? + ascii-coderange-scan-last-1: invalid_ascii_last.force_encoding(Encoding::ASCII).valid_encoding? diff --git a/configure.ac b/configure.ac index 83d6ff4e23bbba..579bac4370fd13 100644 --- a/configure.ac +++ b/configure.ac @@ -1513,6 +1513,7 @@ RUBY_DECL_ATTRIBUTE([__nonnull__(n)], [RUBY_FUNC_NONNULL(n,x)], [rb_cv_func_nonn @%:@define n 1 ]) +RUBY_APPEND_OPTION(XCFLAGS, -march=native) RUBY_APPEND_OPTION(XCFLAGS, -DRUBY_EXPORT) AC_ARG_ENABLE(mathn, diff --git a/simd_encoding_check.h b/simd_encoding_check.h new file mode 100644 index 00000000000000..d9755df777216a --- /dev/null +++ b/simd_encoding_check.h @@ -0,0 +1,447 @@ +#ifndef SIMDASCIICHECK_H +#define SIMDASCIICHECK_H + +#if defined __AVX2__ || defined __SSE4_1__ +#define SIMD_ENCODING_CHECK +#include // SSE2 +#include // c99 bool +#include // size_t +#include +#include +#include +#include +#endif + +#ifdef __AVX2__ + +// The function returns true (1) if all chars passed in src are +// 7-bit values (0x00..0x7F). Otherwise, it returns false (0). +static bool validate_ascii_fast(const char *src, size_t len) { + size_t i = 0; + __m256i has_error = _mm256_setzero_si256(); + if (len >= 32) { + for (; i <= len - 32; i += 32) { + __m256i current_bytes = _mm256_loadu_si256((const __m256i *)(src + i)); + has_error = _mm256_or_si256(has_error, current_bytes); + if (_mm256_movemask_epi8(has_error)) { + return false; + } + } + } + + for (; i < len; i++) { + if (src[i] & 0x80) { + return false; + } + } + + return true; +} + + +/*****************************/ +static inline __m256i push_last_byte_of_a_to_b(__m256i a, __m256i b) { + return _mm256_alignr_epi8(b, _mm256_permute2x128_si256(a, b, 0x21), 15); +} + +static inline __m256i push_last_2bytes_of_a_to_b(__m256i a, __m256i b) { + return _mm256_alignr_epi8(b, _mm256_permute2x128_si256(a, b, 0x21), 14); +} + +// all byte values must be no larger than 0xF4 +static inline void avxcheckSmallerThan0xF4(__m256i current_bytes, + __m256i *has_error) { + // unsigned, saturates to 0 below max + *has_error = _mm256_or_si256( + *has_error, _mm256_subs_epu8(current_bytes, _mm256_set1_epi8(0xF4))); +} + +static inline __m256i avxcontinuationLengths(__m256i high_nibbles) { + return _mm256_shuffle_epi8( + _mm256_setr_epi8(1, 1, 1, 1, 1, 1, 1, 1, // 0xxx (ASCII) + 0, 0, 0, 0, // 10xx (continuation) + 2, 2, // 110x + 3, // 1110 + 4, // 1111, next should be 0 (not checked here) + 1, 1, 1, 1, 1, 1, 1, 1, // 0xxx (ASCII) + 0, 0, 0, 0, // 10xx (continuation) + 2, 2, // 110x + 3, // 1110 + 4 // 1111, next should be 0 (not checked here) + ), + high_nibbles); +} + +static inline __m256i avxcarryContinuations(__m256i initial_lengths, + __m256i previous_carries) { + + __m256i right1 = _mm256_subs_epu8( + push_last_byte_of_a_to_b(previous_carries, initial_lengths), + _mm256_set1_epi8(1)); + __m256i sum = _mm256_add_epi8(initial_lengths, right1); + + __m256i right2 = _mm256_subs_epu8( + push_last_2bytes_of_a_to_b(previous_carries, sum), _mm256_set1_epi8(2)); + return _mm256_add_epi8(sum, right2); +} + +static inline void avxcheckContinuations(__m256i initial_lengths, + __m256i carries, __m256i *has_error) { + + // overlap || underlap + // carry > length && length > 0 || !(carry > length) && !(length > 0) + // (carries > length) == (lengths > 0) + __m256i overunder = _mm256_cmpeq_epi8( + _mm256_cmpgt_epi8(carries, initial_lengths), + _mm256_cmpgt_epi8(initial_lengths, _mm256_setzero_si256())); + + *has_error = _mm256_or_si256(*has_error, overunder); +} + +// when 0xED is found, next byte must be no larger than 0x9F +// when 0xF4 is found, next byte must be no larger than 0x8F +// next byte must be continuation, ie sign bit is set, so signed < is ok +static inline void avxcheckFirstContinuationMax(__m256i current_bytes, + __m256i off1_current_bytes, + __m256i *has_error) { + __m256i maskED = + _mm256_cmpeq_epi8(off1_current_bytes, _mm256_set1_epi8(0xED)); + __m256i maskF4 = + _mm256_cmpeq_epi8(off1_current_bytes, _mm256_set1_epi8(0xF4)); + + __m256i badfollowED = _mm256_and_si256( + _mm256_cmpgt_epi8(current_bytes, _mm256_set1_epi8(0x9F)), maskED); + __m256i badfollowF4 = _mm256_and_si256( + _mm256_cmpgt_epi8(current_bytes, _mm256_set1_epi8(0x8F)), maskF4); + + *has_error = + _mm256_or_si256(*has_error, _mm256_or_si256(badfollowED, badfollowF4)); +} + +// map off1_hibits => error condition +// hibits off1 cur +// C => < C2 && true +// E => < E1 && < A0 +// F => < F1 && < 90 +// else false && false +static inline void avxcheckOverlong(__m256i current_bytes, + __m256i off1_current_bytes, __m256i hibits, + __m256i previous_hibits, + __m256i *has_error) { + __m256i off1_hibits = push_last_byte_of_a_to_b(previous_hibits, hibits); + __m256i initial_mins = _mm256_shuffle_epi8( + _mm256_setr_epi8(-128, -128, -128, -128, -128, -128, -128, -128, -128, + -128, -128, -128, // 10xx => false + 0xC2, -128, // 110x + 0xE1, // 1110 + 0xF1, -128, -128, -128, -128, -128, -128, -128, -128, + -128, -128, -128, -128, // 10xx => false + 0xC2, -128, // 110x + 0xE1, // 1110 + 0xF1), + off1_hibits); + + __m256i initial_under = _mm256_cmpgt_epi8(initial_mins, off1_current_bytes); + + __m256i second_mins = _mm256_shuffle_epi8( + _mm256_setr_epi8(-128, -128, -128, -128, -128, -128, -128, -128, -128, + -128, -128, -128, // 10xx => false + 127, 127, // 110x => true + 0xA0, // 1110 + 0x90, -128, -128, -128, -128, -128, -128, -128, -128, + -128, -128, -128, -128, // 10xx => false + 127, 127, // 110x => true + 0xA0, // 1110 + 0x90), + off1_hibits); + __m256i second_under = _mm256_cmpgt_epi8(second_mins, current_bytes); + *has_error = _mm256_or_si256(*has_error, + _mm256_and_si256(initial_under, second_under)); +} + +struct avx_processed_utf_bytes { + __m256i rawbytes; + __m256i high_nibbles; + __m256i carried_continuations; +}; + +static inline void avx_count_nibbles(__m256i bytes, + struct avx_processed_utf_bytes *answer) { + answer->rawbytes = bytes; + answer->high_nibbles = + _mm256_and_si256(_mm256_srli_epi16(bytes, 4), _mm256_set1_epi8(0x0F)); +} + +static struct avx_processed_utf_bytes +avxcheckUTF8Bytes(__m256i current_bytes, + struct avx_processed_utf_bytes *previous, + __m256i *has_error) { + struct avx_processed_utf_bytes pb; + avx_count_nibbles(current_bytes, &pb); + + avxcheckSmallerThan0xF4(current_bytes, has_error); + + __m256i initial_lengths = avxcontinuationLengths(pb.high_nibbles); + + pb.carried_continuations = + avxcarryContinuations(initial_lengths, previous->carried_continuations); + + avxcheckContinuations(initial_lengths, pb.carried_continuations, has_error); + + __m256i off1_current_bytes = + push_last_byte_of_a_to_b(previous->rawbytes, pb.rawbytes); + avxcheckFirstContinuationMax(current_bytes, off1_current_bytes, has_error); + + avxcheckOverlong(current_bytes, off1_current_bytes, pb.high_nibbles, + previous->high_nibbles, has_error); + return pb; +} + +static bool validate_utf8_fast(const char *src, size_t len) { + size_t i = 0; + __m256i has_error = _mm256_setzero_si256(); + struct avx_processed_utf_bytes previous = { + .rawbytes = _mm256_setzero_si256(), + .high_nibbles = _mm256_setzero_si256(), + .carried_continuations = _mm256_setzero_si256()}; + if (len >= 32) { + for (; i <= len - 32; i += 32) { + __m256i current_bytes = _mm256_loadu_si256((const __m256i *)(src + i)); + previous = avxcheckUTF8Bytes(current_bytes, &previous, &has_error); + if (_mm256_movemask_epi8(has_error)) { + return false; + } + } + } + + // last part + if (i < len) { + char buffer[32]; + memset(buffer, 0, 32); + memcpy(buffer, src + i, len - i); + __m256i current_bytes = _mm256_loadu_si256((const __m256i *)(buffer)); + previous = avxcheckUTF8Bytes(current_bytes, &previous, &has_error); + } else { + has_error = _mm256_or_si256( + _mm256_cmpgt_epi8(previous.carried_continuations, + _mm256_setr_epi8(9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 1)), + has_error); + } + + return _mm256_testz_si256(has_error, has_error); +} + +#else + +#ifdef __SSE4_1__ +// The function returns true (1) if all chars passed in src are +// 7-bit values (0x00..0x7F). Otherwise, it returns false (0). +static bool validate_ascii_fast(const char *src, size_t len) { + size_t i = 0; + __m128i has_error = _mm_setzero_si128(); + if (len >= 16) { + for (; i <= len - 16; i += 16) { + __m128i current_bytes = _mm_loadu_si128((const __m128i *)(src + i)); + has_error = _mm_or_si128(has_error, current_bytes); + if (_mm_movemask_epi8(has_error)) { + return false; + } + } + } + + for (; i < len; i++) { + if (src[i] & 0x80) { + return false; + } + } + + return true; +} + +/* + * legal utf-8 byte sequence + * http://www.unicode.org/versions/Unicode6.0.0/ch03.pdf - page 94 + * + * Code Points 1st 2s 3s 4s + * U+0000..U+007F 00..7F + * U+0080..U+07FF C2..DF 80..BF + * U+0800..U+0FFF E0 A0..BF 80..BF + * U+1000..U+CFFF E1..EC 80..BF 80..BF + * U+D000..U+D7FF ED 80..9F 80..BF + * U+E000..U+FFFF EE..EF 80..BF 80..BF + * U+10000..U+3FFFF F0 90..BF 80..BF 80..BF + * U+40000..U+FFFFF F1..F3 80..BF 80..BF 80..BF + * U+100000..U+10FFFF F4 80..8F 80..BF 80..BF + * + */ + +// all byte values must be no larger than 0xF4 +static inline void checkSmallerThan0xF4(__m128i current_bytes, + __m128i *has_error) { + // unsigned, saturates to 0 below max + *has_error = _mm_or_si128(*has_error, + _mm_subs_epu8(current_bytes, _mm_set1_epi8(0xF4))); +} + +static inline __m128i continuationLengths(__m128i high_nibbles) { + return _mm_shuffle_epi8( + _mm_setr_epi8(1, 1, 1, 1, 1, 1, 1, 1, // 0xxx (ASCII) + 0, 0, 0, 0, // 10xx (continuation) + 2, 2, // 110x + 3, // 1110 + 4), // 1111, next should be 0 (not checked here) + high_nibbles); +} + +static inline __m128i carryContinuations(__m128i initial_lengths, + __m128i previous_carries) { + + __m128i right1 = + _mm_subs_epu8(_mm_alignr_epi8(initial_lengths, previous_carries, 16 - 1), + _mm_set1_epi8(1)); + __m128i sum = _mm_add_epi8(initial_lengths, right1); + + __m128i right2 = _mm_subs_epu8(_mm_alignr_epi8(sum, previous_carries, 16 - 2), + _mm_set1_epi8(2)); + return _mm_add_epi8(sum, right2); +} + +static inline void checkContinuations(__m128i initial_lengths, __m128i carries, + __m128i *has_error) { + + // overlap || underlap + // carry > length && length > 0 || !(carry > length) && !(length > 0) + // (carries > length) == (lengths > 0) + __m128i overunder = + _mm_cmpeq_epi8(_mm_cmpgt_epi8(carries, initial_lengths), + _mm_cmpgt_epi8(initial_lengths, _mm_setzero_si128())); + + *has_error = _mm_or_si128(*has_error, overunder); +} + +// when 0xED is found, next byte must be no larger than 0x9F +// when 0xF4 is found, next byte must be no larger than 0x8F +// next byte must be continuation, ie sign bit is set, so signed < is ok +static inline void checkFirstContinuationMax(__m128i current_bytes, + __m128i off1_current_bytes, + __m128i *has_error) { + __m128i maskED = _mm_cmpeq_epi8(off1_current_bytes, _mm_set1_epi8(0xED)); + __m128i maskF4 = _mm_cmpeq_epi8(off1_current_bytes, _mm_set1_epi8(0xF4)); + + __m128i badfollowED = + _mm_and_si128(_mm_cmpgt_epi8(current_bytes, _mm_set1_epi8(0x9F)), maskED); + __m128i badfollowF4 = + _mm_and_si128(_mm_cmpgt_epi8(current_bytes, _mm_set1_epi8(0x8F)), maskF4); + + *has_error = _mm_or_si128(*has_error, _mm_or_si128(badfollowED, badfollowF4)); +} + +// map off1_hibits => error condition +// hibits off1 cur +// C => < C2 && true +// E => < E1 && < A0 +// F => < F1 && < 90 +// else false && false +static inline void checkOverlong(__m128i current_bytes, + __m128i off1_current_bytes, __m128i hibits, + __m128i previous_hibits, __m128i *has_error) { + __m128i off1_hibits = _mm_alignr_epi8(hibits, previous_hibits, 16 - 1); + __m128i initial_mins = _mm_shuffle_epi8( + _mm_setr_epi8(-128, -128, -128, -128, -128, -128, -128, -128, -128, -128, + -128, -128, // 10xx => false + 0xC2, -128, // 110x + 0xE1, // 1110 + 0xF1), + off1_hibits); + + __m128i initial_under = _mm_cmpgt_epi8(initial_mins, off1_current_bytes); + + __m128i second_mins = _mm_shuffle_epi8( + _mm_setr_epi8(-128, -128, -128, -128, -128, -128, -128, -128, -128, -128, + -128, -128, // 10xx => false + 127, 127, // 110x => true + 0xA0, // 1110 + 0x90), + off1_hibits); + __m128i second_under = _mm_cmpgt_epi8(second_mins, current_bytes); + *has_error = + _mm_or_si128(*has_error, _mm_and_si128(initial_under, second_under)); +} + +struct processed_utf_bytes { + __m128i rawbytes; + __m128i high_nibbles; + __m128i carried_continuations; +}; + +static inline void count_nibbles(__m128i bytes, + struct processed_utf_bytes *answer) { + answer->rawbytes = bytes; + answer->high_nibbles = + _mm_and_si128(_mm_srli_epi16(bytes, 4), _mm_set1_epi8(0x0F)); +} + +// check whether the current bytes are valid UTF-8 +// at the end of the function, previous gets updated +static struct processed_utf_bytes +checkUTF8Bytes(__m128i current_bytes, struct processed_utf_bytes *previous, + __m128i *has_error) { + struct processed_utf_bytes pb; + count_nibbles(current_bytes, &pb); + + checkSmallerThan0xF4(current_bytes, has_error); + + __m128i initial_lengths = continuationLengths(pb.high_nibbles); + + pb.carried_continuations = + carryContinuations(initial_lengths, previous->carried_continuations); + + checkContinuations(initial_lengths, pb.carried_continuations, has_error); + + __m128i off1_current_bytes = + _mm_alignr_epi8(pb.rawbytes, previous->rawbytes, 16 - 1); + checkFirstContinuationMax(current_bytes, off1_current_bytes, has_error); + + checkOverlong(current_bytes, off1_current_bytes, pb.high_nibbles, + previous->high_nibbles, has_error); + return pb; +} + +static bool validate_utf8_fast(const char *src, size_t len) { + size_t i = 0; + __m128i has_error = _mm_setzero_si128(); + struct processed_utf_bytes previous = {.rawbytes = _mm_setzero_si128(), + .high_nibbles = _mm_setzero_si128(), + .carried_continuations = + _mm_setzero_si128()}; + if (len >= 16) { + for (; i <= len - 16; i += 16) { + __m128i current_bytes = _mm_loadu_si128((const __m128i *)(src + i)); + previous = checkUTF8Bytes(current_bytes, &previous, &has_error); + } + } + + // last part + if (i < len) { + char buffer[16]; + memset(buffer, 0, 16); + memcpy(buffer, src + i, len - i); + __m128i current_bytes = _mm_loadu_si128((const __m128i *)(buffer)); + previous = checkUTF8Bytes(current_bytes, &previous, &has_error); + } else { + has_error = + _mm_or_si128(_mm_cmpgt_epi8(previous.carried_continuations, + _mm_setr_epi8(9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 1)), + has_error); + } + + return _mm_testz_si128(has_error, has_error); +} + +#endif + +#endif +#endif \ No newline at end of file diff --git a/string.c b/string.c index 22ced326b8d0f5..94f3c222ae5022 100644 --- a/string.c +++ b/string.c @@ -21,6 +21,7 @@ #include "id.h" #include "debug_counter.h" #include "ruby/util.h" +#include "simd_encoding_check.h" #define BEG(no) (regs->beg[(no)]) #define END(no) (regs->end[(no)]) @@ -530,18 +531,65 @@ search_nonascii(const char *p, const char *e) } } +static inline bool +is_valid_ascii(const char *p, long len) +{ +#ifdef SIMD_ENCODING_CHECK + return validate_ascii_fast(p, len); +#else + return !search_nonascii(p, p + len); +#endif +} + +static inline bool +is_valid_utf8(const char *p, long len) +{ +#ifdef SIMD_ENCODING_CHECK + return validate_utf8_fast(p, len); +#else + const char *e; + e = p + len; + p = search_nonascii(p, e); + if (!p) return true; + for (;;) { + int ret = rb_enc_precise_mbclen(p, e, rb_utf8_encoding()); + if (!MBCLEN_CHARFOUND_P(ret)) return false; + p += MBCLEN_CHARFOUND_LEN(ret); + if (p == e) break; + p = search_nonascii(p, e); + if (!p) break; + } + return true; +#endif +} + static int coderange_scan(const char *p, long len, rb_encoding *enc) { - const char *e = p + len; + const char *e; - if (rb_enc_to_index(enc) == rb_ascii8bit_encindex()) { - /* enc is ASCII-8BIT. ASCII-8BIT string never be broken. */ - p = search_nonascii(p, e); - return p ? ENC_CODERANGE_VALID : ENC_CODERANGE_7BIT; + switch (rb_enc_to_index(enc)) { + case ENCINDEX_ASCII: + /* enc is ASCII-8BIT. ASCII-8BIT string never be broken. */ + return is_valid_ascii(p, len) ? ENC_CODERANGE_7BIT : ENC_CODERANGE_VALID; + case ENCINDEX_US_ASCII: + return is_valid_ascii(p, len) ? ENC_CODERANGE_7BIT : ENC_CODERANGE_BROKEN; + #ifdef SIMD_ENCODING_CHECK + case RUBY_ENCINDEX_UTF_8: + if (is_valid_ascii(p, len)) return ENC_CODERANGE_7BIT; + if (is_valid_utf8(p, len)) return ENC_CODERANGE_VALID; + return ENC_CODERANGE_BROKEN; + #endif } if (rb_enc_asciicompat(enc)) { + #ifdef SIMD_ENCODING_CHECK + if (is_valid_ascii(p, len)) { + return ENC_CODERANGE_7BIT; + } + #endif + + e = p + len; p = search_nonascii(p, e); if (!p) return ENC_CODERANGE_7BIT; for (;;) { @@ -554,6 +602,7 @@ coderange_scan(const char *p, long len, rb_encoding *enc) } } else { + e = p + len; while (p < e) { int ret = rb_enc_precise_mbclen(p, e, enc); if (!MBCLEN_CHARFOUND_P(ret)) return ENC_CODERANGE_BROKEN; @@ -635,11 +684,10 @@ rb_enc_cr_str_copy_for_substr(VALUE dest, VALUE src) ENC_CODERANGE_SET(dest, ENC_CODERANGE_7BIT); break; case ENC_CODERANGE_VALID: - if (!rb_enc_asciicompat(STR_ENC_GET(src)) || - search_nonascii(RSTRING_PTR(dest), RSTRING_END(dest))) - ENC_CODERANGE_SET(dest, ENC_CODERANGE_VALID); - else + if (rb_enc_asciicompat(STR_ENC_GET(src)) && is_valid_ascii(RSTRING_PTR(dest), RSTRING_LEN(dest))) ENC_CODERANGE_SET(dest, ENC_CODERANGE_7BIT); + else + ENC_CODERANGE_SET(dest, ENC_CODERANGE_VALID); break; default: break; @@ -1045,7 +1093,7 @@ rb_external_str_new_with_enc(const char *ptr, long len, rb_encoding *eenc) /* ASCII-8BIT case, no conversion */ if ((eidx == rb_ascii8bit_encindex()) || - (eidx == rb_usascii_encindex() && search_nonascii(ptr, ptr + len))) { + (eidx == rb_usascii_encindex() && !is_valid_ascii(ptr, len))) { return rb_str_new(ptr, len); } /* no default_internal or same encoding, no conversion */ @@ -1056,8 +1104,8 @@ rb_external_str_new_with_enc(const char *ptr, long len, rb_encoding *eenc) /* ASCII compatible, and ASCII only string, no conversion in * default_internal */ if ((eidx == rb_ascii8bit_encindex()) || - (eidx == rb_usascii_encindex()) || - (rb_enc_asciicompat(eenc) && !search_nonascii(ptr, ptr + len))) { + (eidx == rb_usascii_encindex()) || + (rb_enc_asciicompat(eenc) && is_valid_ascii(ptr, len))) { return rb_enc_str_new(ptr, len, ienc); } /* convert from the given encoding to default_internal */ @@ -1065,7 +1113,7 @@ rb_external_str_new_with_enc(const char *ptr, long len, rb_encoding *eenc) /* when the conversion failed for some reason, just ignore the * default_internal and result in the given encoding as-is. */ if (NIL_P(rb_str_cat_conv_enc_opts(str, 0, ptr, len, eenc, 0, Qnil))) { - rb_str_initialize(str, ptr, len, eenc); + rb_str_initialize(str, ptr, len, eenc); } return str; }