1 | #ifndef DATA_LANG_J8_H
|
2 | #define DATA_LANG_J8_H
|
3 |
|
4 | #include <stdio.h> // sprintf
|
5 | #include <string.h> // memcmp, memcpy, strlen
|
6 |
|
7 | #include "data_lang/utf8_impls/bjoern_dfa.h"
|
8 | #include "data_lang/utf8.h"
|
9 |
|
10 | #define J8_OUT(ch) \
|
11 | **p_out = (ch); \
|
12 | (*p_out)++
|
13 |
|
14 | static inline int J8EncodeOne(unsigned char** p_in, unsigned char** p_out,
|
15 | int j8_escape) {
|
16 | // We use a slightly weird double pointer style because
|
17 | // *p_in may be advanced by 1 to 4 bytes (depending on whether it's UTF-8)
|
18 | // *p_out may be advanced by 1 to 6 bytes (depending on escaping)
|
19 |
|
20 | // IMPORTANT: J8EncodeOne(), BourneShellEncodeOne(), BashDollarEncodeOne()
|
21 | // all call Bjoern DFA decode(), and there's a subtle issue where p_in MUST
|
22 | // have a NUL terminator is required. This is so INCOMPLETE UTF-8 sequences
|
23 | // are terminated with an INVALID byte that the state machine can accept, and
|
24 | // 0x00 can only be ITSELF, never part of a sequence. An alternative would be
|
25 | // to do more bounds checks in these functions.
|
26 |
|
27 | // CALLER MUST CHECK that we are able to write up to 6 bytes!
|
28 | // Because the longest output is \u001f or \u{1f} for control chars, since
|
29 | // we don't escapes like \u{1f926} right now
|
30 | //
|
31 | // j8_escape: Whether to use j8 escapes, i.e. LOSSLESS encoding of data
|
32 | // \yff instead of Unicode replacement char
|
33 | // \u{1} instead of \u0001 for unprintable low chars
|
34 |
|
35 | // Returns:
|
36 | // 0 wrote valid UTF-8 (encoded or not)
|
37 | // 1 wrote byte that's invalid UTF-8
|
38 |
|
39 | unsigned char ch = **p_in;
|
40 |
|
41 | //
|
42 | // Handle \\ \b \f \n \r \t
|
43 | //
|
44 |
|
45 | // clang-format off
|
46 | switch (ch) {
|
47 | case '\\': J8_OUT('\\'); J8_OUT('\\'); (*p_in)++; return 0;
|
48 | case '\b': J8_OUT('\\'); J8_OUT('b'); (*p_in)++; return 0;
|
49 | case '\f': J8_OUT('\\'); J8_OUT('f'); (*p_in)++; return 0;
|
50 | case '\n': J8_OUT('\\'); J8_OUT('n'); (*p_in)++; return 0;
|
51 | case '\r': J8_OUT('\\'); J8_OUT('r'); (*p_in)++; return 0;
|
52 | case '\t': J8_OUT('\\'); J8_OUT('t'); (*p_in)++; return 0;
|
53 | }
|
54 | // clang-format on
|
55 |
|
56 | //
|
57 | // Conditionally handle \' and \"
|
58 | //
|
59 | if (ch == '\'' && j8_escape) { // J8-style strings \'
|
60 | J8_OUT('\\');
|
61 | J8_OUT('\'');
|
62 | (*p_in)++;
|
63 | return 0;
|
64 | }
|
65 | if (ch == '"' && !j8_escape) { // JSON-style strings \"
|
66 | J8_OUT('\\');
|
67 | J8_OUT('"');
|
68 | (*p_in)++;
|
69 | return 0;
|
70 | }
|
71 |
|
72 | //
|
73 | // Unprintable ASCII control codes
|
74 | //
|
75 | if (ch < 0x20) {
|
76 | if (j8_escape) {
|
77 | // printf("Writing for %04x %p\n", ch, *p_out);
|
78 | int n = sprintf((char*)*p_out, "\\u{%x}", ch);
|
79 | // printf("! Wrote %d bytes for %04x\n", n, ch);
|
80 | *p_out += n;
|
81 | } else {
|
82 | // printf("Writing for %04x %p\n", ch, *p_out);
|
83 | int n = sprintf((char*)*p_out, "\\u%04x", ch);
|
84 | *p_out += n;
|
85 | // printf("Wrote %d bytes for %04x\n", n, ch);
|
86 | }
|
87 | (*p_in)++;
|
88 | return 0;
|
89 | }
|
90 |
|
91 | //
|
92 | // UTF-8 encoded runes and invalid bytes
|
93 | //
|
94 | Utf8Result_t result;
|
95 | utf8_decode(*p_in, &result);
|
96 |
|
97 | if (result.error == UTF8_OK) {
|
98 | memcpy(*p_out, *p_in, result.bytes_read);
|
99 | *p_in += result.bytes_read;
|
100 | *p_out += result.bytes_read;
|
101 | return 0;
|
102 | }
|
103 |
|
104 | // We have a UTF-8 decoding error. This is handled one of three ways:
|
105 | // 1. Losslessly encode as J8 byte literals (only applicable in J8)
|
106 | // 2. Try to encode a lone surrogate
|
107 | // 3. Insert a Unicode replacement char
|
108 |
|
109 | if (j8_escape) {
|
110 | int n = sprintf((char*)*p_out, "\\y%02x", ch);
|
111 | *p_in += 1;
|
112 | *p_out += n;
|
113 | } else if (result.error == UTF8_ERR_SURROGATE) {
|
114 | int n = sprintf((char*)*p_out, "\\u%04x", result.codepoint);
|
115 | *p_in += result.bytes_read;
|
116 | *p_out += n;
|
117 | return 1;
|
118 | } else {
|
119 | // Unicode replacement char is U+FFFD, so write encoded form
|
120 | // >>> '\ufffd'.encode('utf-8')
|
121 | // b'\xef\xbf\xbd'
|
122 | J8_OUT('\xef');
|
123 | J8_OUT('\xbf');
|
124 | J8_OUT('\xbd');
|
125 | *p_in += 1; // Advance past the byte we wrote
|
126 | }
|
127 |
|
128 | return 1;
|
129 | }
|
130 |
|
131 | // Like the above, but
|
132 | //
|
133 | // \xff instead of \yff
|
134 | // \u001f always, never \u{1f}
|
135 | // No JSON vs. J8
|
136 | // No \" escape ever
|
137 | // No errors -- it can encode everything
|
138 |
|
139 | static inline void BashDollarEncodeOne(unsigned char** p_in,
|
140 | unsigned char** p_out) {
|
141 | unsigned char ch = **p_in;
|
142 |
|
143 | //
|
144 | // Handle \\ \b \f \n \r \t \'
|
145 | //
|
146 |
|
147 | // clang-format off
|
148 | switch (ch) {
|
149 | case '\\': J8_OUT('\\'); J8_OUT('\\'); (*p_in)++; return;
|
150 | case '\b': J8_OUT('\\'); J8_OUT('b'); (*p_in)++; return;
|
151 | case '\f': J8_OUT('\\'); J8_OUT('f'); (*p_in)++; return;
|
152 | case '\n': J8_OUT('\\'); J8_OUT('n'); (*p_in)++; return;
|
153 | case '\r': J8_OUT('\\'); J8_OUT('r'); (*p_in)++; return;
|
154 | case '\t': J8_OUT('\\'); J8_OUT('t'); (*p_in)++; return;
|
155 | case '\'': J8_OUT('\\'); J8_OUT('\''); (*p_in)++; return;
|
156 | }
|
157 | // clang-format on
|
158 |
|
159 | //
|
160 | // Unprintable ASCII control codes
|
161 | //
|
162 | if (ch < 0x20) {
|
163 | // printf("Writing for %04x %p\n", ch, *p_out);
|
164 | int n = sprintf((char*)*p_out, "\\u%04x", ch);
|
165 | *p_out += n;
|
166 | // printf("Wrote %d bytes for %04x\n", n, ch);
|
167 | (*p_in)++;
|
168 | return;
|
169 | }
|
170 |
|
171 | //
|
172 | // UTF-8 encoded runes and invalid bytes
|
173 | //
|
174 | unsigned char* start = *p_in; // save start position
|
175 | uint32_t codepoint = 0;
|
176 | uint32_t state = UTF8_ACCEPT;
|
177 |
|
178 | while (1) {
|
179 | // unsigned char byte = **p_in;
|
180 | decode(&state, &codepoint, ch);
|
181 | // printf(" state %d ch %d\n", state, ch);
|
182 | switch (state) {
|
183 | // BUG: we don't reject IMMEDIATELY
|
184 | //
|
185 | // We could be in another state for up to 4 chars
|
186 | // And then we hit REJECT
|
187 | // And then we need to output \yff\yff\yff\yff
|
188 | // OK that's actually SIXTEEN at once?
|
189 |
|
190 | case UTF8_REJECT: {
|
191 | int n = sprintf((char*)*p_out, "\\x%02x", *start);
|
192 | *p_out += n;
|
193 | (*p_in) = start; // REWIND because we might have consumed NUL terminator!
|
194 | (*p_in)++; // Advance past the byte we wrote
|
195 | return;
|
196 | }
|
197 | case UTF8_ACCEPT: {
|
198 | (*p_in)++;
|
199 | // printf("start %p p_in %p\n", start, *p_in);
|
200 | while (start < *p_in) {
|
201 | J8_OUT(*start);
|
202 | start++;
|
203 | }
|
204 | return;
|
205 | }
|
206 | default:
|
207 | (*p_in)++; // advance, next UTF8_ACCEPT will write it
|
208 | ch = **p_in;
|
209 | // printf(" => ch %d\n", ch);
|
210 | break;
|
211 | }
|
212 | }
|
213 | // Unreachable
|
214 | }
|
215 |
|
216 | // BourneShellEncodeOne rules:
|
217 | //
|
218 | // must be valid UTF-8
|
219 | // no control chars
|
220 | // no ' is required
|
221 | // no \ -- not required, but avoids ambiguous '\n'
|
222 | //
|
223 | // For example we write $'\\' or b'\\' not '\'
|
224 | // The latter should be written r'\', but we're not outputing
|
225 |
|
226 | static inline int BourneShellEncodeOne(unsigned char** p_in,
|
227 | unsigned char** p_out) {
|
228 | unsigned char ch = **p_in;
|
229 |
|
230 | if (ch == '\'' || ch == '\\') { // can't encode these in Bourne shell ''
|
231 | return 1;
|
232 | }
|
233 | if (ch < 0x20) { // Unprintable ASCII control codes
|
234 | return 1;
|
235 | }
|
236 |
|
237 | // UTF-8 encoded runes and invalid bytes
|
238 | unsigned char* start = *p_in; // save start position
|
239 | uint32_t codepoint = 0;
|
240 | uint32_t state = UTF8_ACCEPT;
|
241 |
|
242 | while (1) {
|
243 | decode(&state, &codepoint, ch);
|
244 | // printf(" state %d\n", state);
|
245 | switch (state) {
|
246 | case UTF8_REJECT: {
|
247 | return 1;
|
248 | }
|
249 | case UTF8_ACCEPT: {
|
250 | (*p_in)++;
|
251 | // printf("start %p p_in %p\n", start, *p_in);
|
252 | while (start < *p_in) {
|
253 | J8_OUT(*start);
|
254 | start++;
|
255 | }
|
256 | return 0;
|
257 | }
|
258 | default:
|
259 | (*p_in)++; // advance, next UTF8_ACCEPT will write it
|
260 | ch = **p_in;
|
261 | break;
|
262 | }
|
263 | }
|
264 | // Unreachable
|
265 | }
|
266 |
|
267 | // Right now \u001f and \u{1f} are the longest output sequences for a byte.
|
268 | // Bug fix: we need 6 + 1 for the NUL terminator that sprintf() writes! (Even
|
269 | // though we don't technically need it)
|
270 |
|
271 | // Bug: we may need up to 16 bytes: \yaa\yaa\yaa\yaa
|
272 | // If this is too small, we would enter an infinite loop
|
273 | // +1 for NUL terminator
|
274 |
|
275 | #define J8_MAX_BYTES_PER_INPUT_BYTE 7
|
276 |
|
277 | // The minimum capacity must be more than the number above.
|
278 | // TODO: Tune this for our allocator? We call buf->EnsureMoreSpace(capacity);
|
279 | #define J8_MIN_CAPACITY 16
|
280 |
|
281 | static inline int J8EncodeChunk(unsigned char** p_in, unsigned char* in_end,
|
282 | unsigned char** p_out, unsigned char* out_end,
|
283 | int j8_escape) {
|
284 | while (*p_in < in_end && (*p_out + J8_MAX_BYTES_PER_INPUT_BYTE) <= out_end) {
|
285 | // printf("iter %d %p < %p \n", i++, *p_out, out_end);
|
286 | int invalid_utf8 = J8EncodeOne(p_in, p_out, j8_escape);
|
287 | if (invalid_utf8 && !j8_escape) { // first JSON pass got binary data?
|
288 | return invalid_utf8; // early return
|
289 | }
|
290 | }
|
291 | return 0;
|
292 | }
|
293 |
|
294 | static inline int BashDollarEncodeChunk(unsigned char** p_in,
|
295 | unsigned char* in_end,
|
296 | unsigned char** p_out,
|
297 | unsigned char* out_end) {
|
298 | while (*p_in < in_end && (*p_out + J8_MAX_BYTES_PER_INPUT_BYTE) <= out_end) {
|
299 | BashDollarEncodeOne(p_in, p_out);
|
300 | }
|
301 | return 0;
|
302 | }
|
303 |
|
304 | static inline int BourneShellEncodeChunk(unsigned char** p_in,
|
305 | unsigned char* in_end,
|
306 | unsigned char** p_out,
|
307 | unsigned char* out_end) {
|
308 | while (*p_in < in_end && (*p_out + J8_MAX_BYTES_PER_INPUT_BYTE) <= out_end) {
|
309 | int cannot_encode = BourneShellEncodeOne(p_in, p_out);
|
310 | if (cannot_encode) { // we need escaping, e.g. \u0001 or \'
|
311 | return cannot_encode; // early return
|
312 | }
|
313 | }
|
314 | return 0;
|
315 | }
|
316 |
|
317 | static inline int CanOmitQuotes(unsigned char* s, int len) {
|
318 | if (len == 0) { // empty string has to be quoted
|
319 | return 0;
|
320 | }
|
321 |
|
322 | // 3 special case keywords
|
323 | if (len == 4) {
|
324 | if (memcmp(s, "null", 4) == 0) {
|
325 | return 0;
|
326 | }
|
327 | if (memcmp(s, "true", 4) == 0) {
|
328 | return 0;
|
329 | }
|
330 | }
|
331 | if (len == 5) {
|
332 | if (memcmp(s, "false", 5) == 0) {
|
333 | return 0;
|
334 | }
|
335 | }
|
336 |
|
337 | for (int i = 0; i < len; ++i) {
|
338 | unsigned char ch = s[i];
|
339 |
|
340 | // Corresponds to regex [a-zA-Z0-9./_-]
|
341 | if ('a' <= ch && ch <= 'z') {
|
342 | continue;
|
343 | }
|
344 | if ('A' <= ch && ch <= 'Z') {
|
345 | continue;
|
346 | }
|
347 | if ('0' <= ch && ch <= '9') {
|
348 | continue;
|
349 | }
|
350 | if (ch == '.' || ch == '/' || ch == '_' || ch == '-') {
|
351 | continue;
|
352 | }
|
353 | // some byte requires quotes
|
354 | // Not including UTF-8 here because it can have chars that look like space
|
355 | // or quotes
|
356 | return 0;
|
357 | }
|
358 | return 1; // everything OK
|
359 | }
|
360 |
|
361 | #endif // DATA_LANG_J8_H
|