Initial commit: add .gitignore and README

.venv/lib/python3.12/site-packages/mypyc/lib-rt/base64/arch/generic/32/dec_loop.c

@@ -0,0 +1,86 @@
+static BASE64_FORCE_INLINE int
+dec_loop_generic_32_inner (const uint8_t **s, uint8_t **o, size_t *rounds)
+{
+	const uint32_t str
+		= base64_table_dec_32bit_d0[(*s)[0]]
+		| base64_table_dec_32bit_d1[(*s)[1]]
+		| base64_table_dec_32bit_d2[(*s)[2]]
+		| base64_table_dec_32bit_d3[(*s)[3]];
+
+#if BASE64_LITTLE_ENDIAN
+
+	// LUTs for little-endian set MSB in case of invalid character:
+	if (str & UINT32_C(0x80000000)) {
+		return 0;
+	}
+#else
+	// LUTs for big-endian set LSB in case of invalid character:
+	if (str & UINT32_C(1)) {
+		return 0;
+	}
+#endif
+	// Store the output:
+	memcpy(*o, &str, sizeof (str));
+
+	*s += 4;
+	*o += 3;
+	*rounds -= 1;
+
+	return 1;
+}
+
+static inline void
+dec_loop_generic_32 (const uint8_t **s, size_t *slen, uint8_t **o, size_t *olen)
+{
+	if (*slen < 8) {
+		return;
+	}
+
+	// Process blocks of 4 bytes per round. Because one extra zero byte is
+	// written after the output, ensure that there will be at least 4 bytes
+	// of input data left to cover the gap. (Two data bytes and up to two
+	// end-of-string markers.)
+	size_t rounds = (*slen - 4) / 4;
+
+	*slen -= rounds * 4;	// 4 bytes consumed per round
+	*olen += rounds * 3;	// 3 bytes produced per round
+
+	do {
+		if (rounds >= 8) {
+			if (dec_loop_generic_32_inner(s, o, &rounds) &&
+			    dec_loop_generic_32_inner(s, o, &rounds) &&
+			    dec_loop_generic_32_inner(s, o, &rounds) &&
+			    dec_loop_generic_32_inner(s, o, &rounds) &&
+			    dec_loop_generic_32_inner(s, o, &rounds) &&
+			    dec_loop_generic_32_inner(s, o, &rounds) &&
+			    dec_loop_generic_32_inner(s, o, &rounds) &&
+			    dec_loop_generic_32_inner(s, o, &rounds)) {
+				continue;
+			}
+			break;
+		}
+		if (rounds >= 4) {
+			if (dec_loop_generic_32_inner(s, o, &rounds) &&
+			    dec_loop_generic_32_inner(s, o, &rounds) &&
+			    dec_loop_generic_32_inner(s, o, &rounds) &&
+			    dec_loop_generic_32_inner(s, o, &rounds)) {
+				continue;
+			}
+			break;
+		}
+		if (rounds >= 2) {
+			if (dec_loop_generic_32_inner(s, o, &rounds) &&
+			    dec_loop_generic_32_inner(s, o, &rounds)) {
+				continue;
+			}
+			break;
+		}
+		dec_loop_generic_32_inner(s, o, &rounds);
+		break;
+
+	} while (rounds > 0);
+
+	// Adjust for any rounds that were skipped:
+	*slen += rounds * 4;
+	*olen -= rounds * 3;
+}

.venv/lib/python3.12/site-packages/mypyc/lib-rt/base64/arch/generic/32/enc_loop.c

@@ -0,0 +1,73 @@
+static BASE64_FORCE_INLINE void
+enc_loop_generic_32_inner (const uint8_t **s, uint8_t **o)
+{
+	uint32_t src;
+
+	// Load input:
+	memcpy(&src, *s, sizeof (src));
+
+	// Reorder to 32-bit big-endian, if not already in that format. The
+	// workset must be in big-endian, otherwise the shifted bits do not
+	// carry over properly among adjacent bytes:
+	src = BASE64_HTOBE32(src);
+
+	// Two indices for the 12-bit lookup table:
+	const size_t index0 = (src >> 20) & 0xFFFU;
+	const size_t index1 = (src >>  8) & 0xFFFU;
+
+	// Table lookup and store:
+	memcpy(*o + 0, base64_table_enc_12bit + index0, 2);
+	memcpy(*o + 2, base64_table_enc_12bit + index1, 2);
+
+	*s += 3;
+	*o += 4;
+}
+
+static inline void
+enc_loop_generic_32 (const uint8_t **s, size_t *slen, uint8_t **o, size_t *olen)
+{
+	if (*slen < 4) {
+		return;
+	}
+
+	// Process blocks of 3 bytes at a time. Because blocks are loaded 4
+	// bytes at a time, ensure that there will be at least one remaining
+	// byte after the last round, so that the final read will not pass
+	// beyond the bounds of the input buffer:
+	size_t rounds = (*slen - 1) / 3;
+
+	*slen -= rounds * 3;	// 3 bytes consumed per round
+	*olen += rounds * 4;	// 4 bytes produced per round
+
+	do {
+		if (rounds >= 8) {
+			enc_loop_generic_32_inner(s, o);
+			enc_loop_generic_32_inner(s, o);
+			enc_loop_generic_32_inner(s, o);
+			enc_loop_generic_32_inner(s, o);
+			enc_loop_generic_32_inner(s, o);
+			enc_loop_generic_32_inner(s, o);
+			enc_loop_generic_32_inner(s, o);
+			enc_loop_generic_32_inner(s, o);
+			rounds -= 8;
+			continue;
+		}
+		if (rounds >= 4) {
+			enc_loop_generic_32_inner(s, o);
+			enc_loop_generic_32_inner(s, o);
+			enc_loop_generic_32_inner(s, o);
+			enc_loop_generic_32_inner(s, o);
+			rounds -= 4;
+			continue;
+		}
+		if (rounds >= 2) {
+			enc_loop_generic_32_inner(s, o);
+			enc_loop_generic_32_inner(s, o);
+			rounds -= 2;
+			continue;
+		}
+		enc_loop_generic_32_inner(s, o);
+		break;
+
+	} while (rounds > 0);
+}

.venv/lib/python3.12/site-packages/mypyc/lib-rt/base64/arch/generic/64/enc_loop.c

@@ -0,0 +1,77 @@
+static BASE64_FORCE_INLINE void
+enc_loop_generic_64_inner (const uint8_t **s, uint8_t **o)
+{
+	uint64_t src;
+
+	// Load input:
+	memcpy(&src, *s, sizeof (src));
+
+	// Reorder to 64-bit big-endian, if not already in that format. The
+	// workset must be in big-endian, otherwise the shifted bits do not
+	// carry over properly among adjacent bytes:
+	src = BASE64_HTOBE64(src);
+
+	// Four indices for the 12-bit lookup table:
+	const size_t index0 = (src >> 52) & 0xFFFU;
+	const size_t index1 = (src >> 40) & 0xFFFU;
+	const size_t index2 = (src >> 28) & 0xFFFU;
+	const size_t index3 = (src >> 16) & 0xFFFU;
+
+	// Table lookup and store:
+	memcpy(*o + 0, base64_table_enc_12bit + index0, 2);
+	memcpy(*o + 2, base64_table_enc_12bit + index1, 2);
+	memcpy(*o + 4, base64_table_enc_12bit + index2, 2);
+	memcpy(*o + 6, base64_table_enc_12bit + index3, 2);
+
+	*s += 6;
+	*o += 8;
+}
+
+static inline void
+enc_loop_generic_64 (const uint8_t **s, size_t *slen, uint8_t **o, size_t *olen)
+{
+	if (*slen < 8) {
+		return;
+	}
+
+	// Process blocks of 6 bytes at a time. Because blocks are loaded 8
+	// bytes at a time, ensure that there will be at least 2 remaining
+	// bytes after the last round, so that the final read will not pass
+	// beyond the bounds of the input buffer:
+	size_t rounds = (*slen - 2) / 6;
+
+	*slen -= rounds * 6;	// 6 bytes consumed per round
+	*olen += rounds * 8;	// 8 bytes produced per round
+
+	do {
+		if (rounds >= 8) {
+			enc_loop_generic_64_inner(s, o);
+			enc_loop_generic_64_inner(s, o);
+			enc_loop_generic_64_inner(s, o);
+			enc_loop_generic_64_inner(s, o);
+			enc_loop_generic_64_inner(s, o);
+			enc_loop_generic_64_inner(s, o);
+			enc_loop_generic_64_inner(s, o);
+			enc_loop_generic_64_inner(s, o);
+			rounds -= 8;
+			continue;
+		}
+		if (rounds >= 4) {
+			enc_loop_generic_64_inner(s, o);
+			enc_loop_generic_64_inner(s, o);
+			enc_loop_generic_64_inner(s, o);
+			enc_loop_generic_64_inner(s, o);
+			rounds -= 4;
+			continue;
+		}
+		if (rounds >= 2) {
+			enc_loop_generic_64_inner(s, o);
+			enc_loop_generic_64_inner(s, o);
+			rounds -= 2;
+			continue;
+		}
+		enc_loop_generic_64_inner(s, o);
+		break;
+
+	} while (rounds > 0);
+}

.venv/lib/python3.12/site-packages/mypyc/lib-rt/base64/arch/generic/codec.c

@@ -0,0 +1,41 @@
+#include <stdint.h>
+#include <stddef.h>
+#include <string.h>
+
+#include "libbase64.h"
+#include "../../tables/tables.h"
+#include "../../codecs.h"
+#include "config.h"
+#include "../../env.h"
+
+#if BASE64_WORDSIZE == 32
+#  include "32/enc_loop.c"
+#elif BASE64_WORDSIZE == 64
+#  include "64/enc_loop.c"
+#endif
+
+#if BASE64_WORDSIZE >= 32
+#  include "32/dec_loop.c"
+#endif
+
+void
+base64_stream_encode_plain BASE64_ENC_PARAMS
+{
+	#include "enc_head.c"
+#if BASE64_WORDSIZE == 32
+	enc_loop_generic_32(&s, &slen, &o, &olen);
+#elif BASE64_WORDSIZE == 64
+	enc_loop_generic_64(&s, &slen, &o, &olen);
+#endif
+	#include "enc_tail.c"
+}
+
+int
+base64_stream_decode_plain BASE64_DEC_PARAMS
+{
+	#include "dec_head.c"
+#if BASE64_WORDSIZE >= 32
+	dec_loop_generic_32(&s, &slen, &o, &olen);
+#endif
+	#include "dec_tail.c"
+}

.venv/lib/python3.12/site-packages/mypyc/lib-rt/base64/arch/generic/dec_head.c

@@ -0,0 +1,37 @@
+int ret = 0;
+const uint8_t *s = (const uint8_t *) src;
+uint8_t *o = (uint8_t *) out;
+uint8_t q;
+
+// Use local temporaries to avoid cache thrashing:
+size_t olen = 0;
+size_t slen = srclen;
+struct base64_state st;
+st.eof = state->eof;
+st.bytes = state->bytes;
+st.carry = state->carry;
+
+// If we previously saw an EOF or an invalid character, bail out:
+if (st.eof) {
+	*outlen = 0;
+	ret = 0;
+	// If there was a trailing '=' to check, check it:
+	if (slen && (st.eof == BASE64_AEOF)) {
+		state->bytes = 0;
+		state->eof = BASE64_EOF;
+		ret = ((base64_table_dec_8bit[*s++] == 254) && (slen == 1)) ? 1 : 0;
+	}
+	return ret;
+}
+
+// Turn four 6-bit numbers into three bytes:
+// out[0] = 11111122
+// out[1] = 22223333
+// out[2] = 33444444
+
+// Duff's device again:
+switch (st.bytes)
+{
+	for (;;)
+	{
+	case 0:

.venv/lib/python3.12/site-packages/mypyc/lib-rt/base64/arch/generic/dec_tail.c

@@ -0,0 +1,91 @@
+		if (slen-- == 0) {
+			ret = 1;
+			break;
+		}
+		if ((q = base64_table_dec_8bit[*s++]) >= 254) {
+			st.eof = BASE64_EOF;
+			// Treat character '=' as invalid for byte 0:
+			break;
+		}
+		st.carry = q << 2;
+		st.bytes++;
+
+		// Deliberate fallthrough:
+		BASE64_FALLTHROUGH
+
+	case 1:	if (slen-- == 0) {
+			ret = 1;
+			break;
+		}
+		if ((q = base64_table_dec_8bit[*s++]) >= 254) {
+			st.eof = BASE64_EOF;
+			// Treat character '=' as invalid for byte 1:
+			break;
+		}
+		*o++ = st.carry | (q >> 4);
+		st.carry = q << 4;
+		st.bytes++;
+		olen++;
+
+		// Deliberate fallthrough:
+		BASE64_FALLTHROUGH
+
+	case 2:	if (slen-- == 0) {
+			ret = 1;
+			break;
+		}
+		if ((q = base64_table_dec_8bit[*s++]) >= 254) {
+			st.bytes++;
+			// When q == 254, the input char is '='.
+			// Check if next byte is also '=':
+			if (q == 254) {
+				if (slen-- != 0) {
+					st.bytes = 0;
+					// EOF:
+					st.eof = BASE64_EOF;
+					q = base64_table_dec_8bit[*s++];
+					ret = ((q == 254) && (slen == 0)) ? 1 : 0;
+					break;
+				}
+				else {
+					// Almost EOF
+					st.eof = BASE64_AEOF;
+					ret = 1;
+					break;
+				}
+			}
+			// If we get here, there was an error:
+			break;
+		}
+		*o++ = st.carry | (q >> 2);
+		st.carry = q << 6;
+		st.bytes++;
+		olen++;
+
+		// Deliberate fallthrough:
+		BASE64_FALLTHROUGH
+
+	case 3:	if (slen-- == 0) {
+			ret = 1;
+			break;
+		}
+		if ((q = base64_table_dec_8bit[*s++]) >= 254) {
+			st.bytes = 0;
+			st.eof = BASE64_EOF;
+			// When q == 254, the input char is '='. Return 1 and EOF.
+			// When q == 255, the input char is invalid. Return 0 and EOF.
+			ret = ((q == 254) && (slen == 0)) ? 1 : 0;
+			break;
+		}
+		*o++ = st.carry | q;
+		st.carry = 0;
+		st.bytes = 0;
+		olen++;
+	}
+}
+
+state->eof = st.eof;
+state->bytes = st.bytes;
+state->carry = st.carry;
+*outlen = olen;
+return ret;

.venv/lib/python3.12/site-packages/mypyc/lib-rt/base64/arch/generic/enc_head.c

@@ -0,0 +1,24 @@
+// Assume that *out is large enough to contain the output.
+// Theoretically it should be 4/3 the length of src.
+const uint8_t *s = (const uint8_t *) src;
+uint8_t *o = (uint8_t *) out;
+
+// Use local temporaries to avoid cache thrashing:
+size_t olen = 0;
+size_t slen = srclen;
+struct base64_state st;
+st.bytes = state->bytes;
+st.carry = state->carry;
+
+// Turn three bytes into four 6-bit numbers:
+// in[0] = 00111111
+// in[1] = 00112222
+// in[2] = 00222233
+// in[3] = 00333333
+
+// Duff's device, a for() loop inside a switch() statement. Legal!
+switch (st.bytes)
+{
+	for (;;)
+	{
+	case 0:

.venv/lib/python3.12/site-packages/mypyc/lib-rt/base64/arch/generic/enc_tail.c

@@ -0,0 +1,34 @@
+		if (slen-- == 0) {
+			break;
+		}
+		*o++ = base64_table_enc_6bit[*s >> 2];
+		st.carry = (*s++ << 4) & 0x30;
+		st.bytes++;
+		olen += 1;
+
+		// Deliberate fallthrough:
+		BASE64_FALLTHROUGH
+
+	case 1:	if (slen-- == 0) {
+			break;
+		}
+		*o++ = base64_table_enc_6bit[st.carry | (*s >> 4)];
+		st.carry = (*s++ << 2) & 0x3C;
+		st.bytes++;
+		olen += 1;
+
+		// Deliberate fallthrough:
+		BASE64_FALLTHROUGH
+
+	case 2:	if (slen-- == 0) {
+			break;
+		}
+		*o++ = base64_table_enc_6bit[st.carry | (*s >> 6)];
+		*o++ = base64_table_enc_6bit[*s++ & 0x3F];
+		st.bytes = 0;
+		olen += 2;
+	}
+}
+state->bytes = st.bytes;
+state->carry = st.carry;
+*outlen = olen;