xref: /netbsd-src/external/gpl3/gcc.old/dist/libcpp/lex.c (revision 404ee5b9334f618040b6cdef96a0ff35a6fc4636)
1 /* CPP Library - lexical analysis.
2    Copyright (C) 2000-2017 Free Software Foundation, Inc.
3    Contributed by Per Bothner, 1994-95.
4    Based on CCCP program by Paul Rubin, June 1986
5    Adapted to ANSI C, Richard Stallman, Jan 1987
6    Broken out to separate file, Zack Weinberg, Mar 2000
7 
8 This program is free software; you can redistribute it and/or modify it
9 under the terms of the GNU General Public License as published by the
10 Free Software Foundation; either version 3, or (at your option) any
11 later version.
12 
13 This program is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16 GNU General Public License for more details.
17 
18 You should have received a copy of the GNU General Public License
19 along with this program; see the file COPYING3.  If not see
20 <http://www.gnu.org/licenses/>.  */
21 
22 #include "config.h"
23 #include "system.h"
24 #include "cpplib.h"
25 #include "internal.h"
26 
27 enum spell_type
28 {
29   SPELL_OPERATOR = 0,
30   SPELL_IDENT,
31   SPELL_LITERAL,
32   SPELL_NONE
33 };
34 
35 struct token_spelling
36 {
37   enum spell_type category;
38   const unsigned char *name;
39 };
40 
41 static const unsigned char *const digraph_spellings[] =
42 { UC"%:", UC"%:%:", UC"<:", UC":>", UC"<%", UC"%>" };
43 
44 #define OP(e, s) { SPELL_OPERATOR, UC s  },
45 #define TK(e, s) { SPELL_ ## s,    UC #e },
46 static const struct token_spelling token_spellings[N_TTYPES] = { TTYPE_TABLE };
47 #undef OP
48 #undef TK
49 
50 #define TOKEN_SPELL(token) (token_spellings[(token)->type].category)
51 #define TOKEN_NAME(token) (token_spellings[(token)->type].name)
52 
53 static void add_line_note (cpp_buffer *, const uchar *, unsigned int);
54 static int skip_line_comment (cpp_reader *);
55 static void skip_whitespace (cpp_reader *, cppchar_t);
56 static void lex_string (cpp_reader *, cpp_token *, const uchar *);
57 static void save_comment (cpp_reader *, cpp_token *, const uchar *, cppchar_t);
58 static void store_comment (cpp_reader *, cpp_token *);
59 static void create_literal (cpp_reader *, cpp_token *, const uchar *,
60 			    unsigned int, enum cpp_ttype);
61 static bool warn_in_comment (cpp_reader *, _cpp_line_note *);
62 static int name_p (cpp_reader *, const cpp_string *);
63 static tokenrun *next_tokenrun (tokenrun *);
64 
65 static _cpp_buff *new_buff (size_t);
66 
67 
68 /* Utility routine:
69 
70    Compares, the token TOKEN to the NUL-terminated string STRING.
71    TOKEN must be a CPP_NAME.  Returns 1 for equal, 0 for unequal.  */
72 int
73 cpp_ideq (const cpp_token *token, const char *string)
74 {
75   if (token->type != CPP_NAME)
76     return 0;
77 
78   return !ustrcmp (NODE_NAME (token->val.node.node), (const uchar *) string);
79 }
80 
81 /* Record a note TYPE at byte POS into the current cleaned logical
82    line.  */
83 static void
84 add_line_note (cpp_buffer *buffer, const uchar *pos, unsigned int type)
85 {
86   if (buffer->notes_used == buffer->notes_cap)
87     {
88       buffer->notes_cap = buffer->notes_cap * 2 + 200;
89       buffer->notes = XRESIZEVEC (_cpp_line_note, buffer->notes,
90                                   buffer->notes_cap);
91     }
92 
93   buffer->notes[buffer->notes_used].pos = pos;
94   buffer->notes[buffer->notes_used].type = type;
95   buffer->notes_used++;
96 }
97 
98 
99 /* Fast path to find line special characters using optimized character
100    scanning algorithms.  Anything complicated falls back to the slow
101    path below.  Since this loop is very hot it's worth doing these kinds
102    of optimizations.
103 
104    One of the paths through the ifdefs should provide
105 
106      const uchar *search_line_fast (const uchar *s, const uchar *end);
107 
108    Between S and END, search for \n, \r, \\, ?.  Return a pointer to
109    the found character.
110 
111    Note that the last character of the buffer is *always* a newline,
112    as forced by _cpp_convert_input.  This fact can be used to avoid
113    explicitly looking for the end of the buffer.  */
114 
115 /* Configure gives us an ifdef test.  */
116 #ifndef WORDS_BIGENDIAN
117 #define WORDS_BIGENDIAN 0
118 #endif
119 
120 /* We'd like the largest integer that fits into a register.  There's nothing
121    in <stdint.h> that gives us that.  For most hosts this is unsigned long,
122    but MS decided on an LLP64 model.  Thankfully when building with GCC we
123    can get the "real" word size.  */
124 #ifdef __GNUC__
125 typedef unsigned int word_type __attribute__((__mode__(__word__)));
126 #else
127 typedef unsigned long word_type;
128 #endif
129 
130 /* The code below is only expecting sizes 4 or 8.
131    Die at compile-time if this expectation is violated.  */
132 typedef char check_word_type_size
133   [(sizeof(word_type) == 8 || sizeof(word_type) == 4) * 2 - 1];
134 
135 /* Return X with the first N bytes forced to values that won't match one
136    of the interesting characters.  Note that NUL is not interesting.  */
137 
138 static inline word_type
139 acc_char_mask_misalign (word_type val, unsigned int n)
140 {
141   word_type mask = -1;
142   if (WORDS_BIGENDIAN)
143     mask >>= n * 8;
144   else
145     mask <<= n * 8;
146   return val & mask;
147 }
148 
149 /* Return X replicated to all byte positions within WORD_TYPE.  */
150 
151 static inline word_type
152 acc_char_replicate (uchar x)
153 {
154   word_type ret;
155 
156   ret = (x << 24) | (x << 16) | (x << 8) | x;
157   if (sizeof(word_type) == 8)
158     ret = (ret << 16 << 16) | ret;
159   return ret;
160 }
161 
162 /* Return non-zero if some byte of VAL is (probably) C.  */
163 
164 static inline word_type
165 acc_char_cmp (word_type val, word_type c)
166 {
167 #if defined(__GNUC__) && defined(__alpha__)
168   /* We can get exact results using a compare-bytes instruction.
169      Get (val == c) via (0 >= (val ^ c)).  */
170   return __builtin_alpha_cmpbge (0, val ^ c);
171 #else
172   word_type magic = 0x7efefefeU;
173   if (sizeof(word_type) == 8)
174     magic = (magic << 16 << 16) | 0xfefefefeU;
175   magic |= 1;
176 
177   val ^= c;
178   return ((val + magic) ^ ~val) & ~magic;
179 #endif
180 }
181 
182 /* Given the result of acc_char_cmp is non-zero, return the index of
183    the found character.  If this was a false positive, return -1.  */
184 
185 static inline int
186 acc_char_index (word_type cmp ATTRIBUTE_UNUSED,
187 		word_type val ATTRIBUTE_UNUSED)
188 {
189 #if defined(__GNUC__) && defined(__alpha__) && !WORDS_BIGENDIAN
190   /* The cmpbge instruction sets *bits* of the result corresponding to
191      matches in the bytes with no false positives.  */
192   return __builtin_ctzl (cmp);
193 #else
194   unsigned int i;
195 
196   /* ??? It would be nice to force unrolling here,
197      and have all of these constants folded.  */
198   for (i = 0; i < sizeof(word_type); ++i)
199     {
200       uchar c;
201       if (WORDS_BIGENDIAN)
202 	c = (val >> (sizeof(word_type) - i - 1) * 8) & 0xff;
203       else
204 	c = (val >> i * 8) & 0xff;
205 
206       if (c == '\n' || c == '\r' || c == '\\' || c == '?')
207 	return i;
208     }
209 
210   return -1;
211 #endif
212 }
213 
214 /* A version of the fast scanner using bit fiddling techniques.
215 
216    For 32-bit words, one would normally perform 16 comparisons and
217    16 branches.  With this algorithm one performs 24 arithmetic
218    operations and one branch.  Whether this is faster with a 32-bit
219    word size is going to be somewhat system dependent.
220 
221    For 64-bit words, we eliminate twice the number of comparisons
222    and branches without increasing the number of arithmetic operations.
223    It's almost certainly going to be a win with 64-bit word size.  */
224 
225 static const uchar * search_line_acc_char (const uchar *, const uchar *)
226   ATTRIBUTE_UNUSED;
227 
228 static const uchar *
229 search_line_acc_char (const uchar *s, const uchar *end ATTRIBUTE_UNUSED)
230 {
231   const word_type repl_nl = acc_char_replicate ('\n');
232   const word_type repl_cr = acc_char_replicate ('\r');
233   const word_type repl_bs = acc_char_replicate ('\\');
234   const word_type repl_qm = acc_char_replicate ('?');
235 
236   unsigned int misalign;
237   const word_type *p;
238   word_type val, t;
239 
240   /* Align the buffer.  Mask out any bytes from before the beginning.  */
241   p = (word_type *)((uintptr_t)s & -sizeof(word_type));
242   val = *p;
243   misalign = (uintptr_t)s & (sizeof(word_type) - 1);
244   if (misalign)
245     val = acc_char_mask_misalign (val, misalign);
246 
247   /* Main loop.  */
248   while (1)
249     {
250       t  = acc_char_cmp (val, repl_nl);
251       t |= acc_char_cmp (val, repl_cr);
252       t |= acc_char_cmp (val, repl_bs);
253       t |= acc_char_cmp (val, repl_qm);
254 
255       if (__builtin_expect (t != 0, 0))
256 	{
257 	  int i = acc_char_index (t, val);
258 	  if (i >= 0)
259 	    return (const uchar *)p + i;
260 	}
261 
262       val = *++p;
263     }
264 }
265 
266 /* Disable on Solaris 2/x86 until the following problem can be properly
267    autoconfed:
268 
269    The Solaris 10+ assembler tags objects with the instruction set
270    extensions used, so SSE4.2 executables cannot run on machines that
271    don't support that extension.  */
272 
273 #if (GCC_VERSION >= 4005) && (__GNUC__ >= 5 || !defined(__PIC__)) && (defined(__i386__) || defined(__x86_64__)) && !(defined(__sun__) && defined(__svr4__))
274 
275 /* Replicated character data to be shared between implementations.
276    Recall that outside of a context with vector support we can't
277    define compatible vector types, therefore these are all defined
278    in terms of raw characters.  */
279 static const char repl_chars[4][16] __attribute__((aligned(16))) = {
280   { '\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n',
281     '\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n' },
282   { '\r', '\r', '\r', '\r', '\r', '\r', '\r', '\r',
283     '\r', '\r', '\r', '\r', '\r', '\r', '\r', '\r' },
284   { '\\', '\\', '\\', '\\', '\\', '\\', '\\', '\\',
285     '\\', '\\', '\\', '\\', '\\', '\\', '\\', '\\' },
286   { '?', '?', '?', '?', '?', '?', '?', '?',
287     '?', '?', '?', '?', '?', '?', '?', '?' },
288 };
289 
290 /* A version of the fast scanner using MMX vectorized byte compare insns.
291 
292    This uses the PMOVMSKB instruction which was introduced with "MMX2",
293    which was packaged into SSE1; it is also present in the AMD MMX
294    extension.  Mark the function as using "sse" so that we emit a real
295    "emms" instruction, rather than the 3dNOW "femms" instruction.  */
296 
297 static const uchar *
298 #ifndef __SSE__
299 __attribute__((__target__("sse")))
300 #endif
301 search_line_mmx (const uchar *s, const uchar *end ATTRIBUTE_UNUSED)
302 {
303   typedef char v8qi __attribute__ ((__vector_size__ (8)));
304   typedef int __m64 __attribute__ ((__vector_size__ (8), __may_alias__));
305 
306   const v8qi repl_nl = *(const v8qi *)repl_chars[0];
307   const v8qi repl_cr = *(const v8qi *)repl_chars[1];
308   const v8qi repl_bs = *(const v8qi *)repl_chars[2];
309   const v8qi repl_qm = *(const v8qi *)repl_chars[3];
310 
311   unsigned int misalign, found, mask;
312   const v8qi *p;
313   v8qi data, t, c;
314 
315   /* Align the source pointer.  While MMX doesn't generate unaligned data
316      faults, this allows us to safely scan to the end of the buffer without
317      reading beyond the end of the last page.  */
318   misalign = (uintptr_t)s & 7;
319   p = (const v8qi *)((uintptr_t)s & -8);
320   data = *p;
321 
322   /* Create a mask for the bytes that are valid within the first
323      16-byte block.  The Idea here is that the AND with the mask
324      within the loop is "free", since we need some AND or TEST
325      insn in order to set the flags for the branch anyway.  */
326   mask = -1u << misalign;
327 
328   /* Main loop processing 8 bytes at a time.  */
329   goto start;
330   do
331     {
332       data = *++p;
333       mask = -1;
334 
335     start:
336       t = __builtin_ia32_pcmpeqb(data, repl_nl);
337       c = __builtin_ia32_pcmpeqb(data, repl_cr);
338       t = (v8qi) __builtin_ia32_por ((__m64)t, (__m64)c);
339       c = __builtin_ia32_pcmpeqb(data, repl_bs);
340       t = (v8qi) __builtin_ia32_por ((__m64)t, (__m64)c);
341       c = __builtin_ia32_pcmpeqb(data, repl_qm);
342       t = (v8qi) __builtin_ia32_por ((__m64)t, (__m64)c);
343       found = __builtin_ia32_pmovmskb (t);
344       found &= mask;
345     }
346   while (!found);
347 
348   __builtin_ia32_emms ();
349 
350   /* FOUND contains 1 in bits for which we matched a relevant
351      character.  Conversion to the byte index is trivial.  */
352   found = __builtin_ctz(found);
353   return (const uchar *)p + found;
354 }
355 
356 /* A version of the fast scanner using SSE2 vectorized byte compare insns.  */
357 
358 static const uchar *
359 #ifndef __SSE2__
360 __attribute__((__target__("sse2")))
361 #endif
362 search_line_sse2 (const uchar *s, const uchar *end ATTRIBUTE_UNUSED)
363 {
364   typedef char v16qi __attribute__ ((__vector_size__ (16)));
365 
366   const v16qi repl_nl = *(const v16qi *)repl_chars[0];
367   const v16qi repl_cr = *(const v16qi *)repl_chars[1];
368   const v16qi repl_bs = *(const v16qi *)repl_chars[2];
369   const v16qi repl_qm = *(const v16qi *)repl_chars[3];
370 
371   unsigned int misalign, found, mask;
372   const v16qi *p;
373   v16qi data, t;
374 
375   /* Align the source pointer.  */
376   misalign = (uintptr_t)s & 15;
377   p = (const v16qi *)((uintptr_t)s & -16);
378   data = *p;
379 
380   /* Create a mask for the bytes that are valid within the first
381      16-byte block.  The Idea here is that the AND with the mask
382      within the loop is "free", since we need some AND or TEST
383      insn in order to set the flags for the branch anyway.  */
384   mask = -1u << misalign;
385 
386   /* Main loop processing 16 bytes at a time.  */
387   goto start;
388   do
389     {
390       data = *++p;
391       mask = -1;
392 
393     start:
394       t  = __builtin_ia32_pcmpeqb128(data, repl_nl);
395       t |= __builtin_ia32_pcmpeqb128(data, repl_cr);
396       t |= __builtin_ia32_pcmpeqb128(data, repl_bs);
397       t |= __builtin_ia32_pcmpeqb128(data, repl_qm);
398       found = __builtin_ia32_pmovmskb128 (t);
399       found &= mask;
400     }
401   while (!found);
402 
403   /* FOUND contains 1 in bits for which we matched a relevant
404      character.  Conversion to the byte index is trivial.  */
405   found = __builtin_ctz(found);
406   return (const uchar *)p + found;
407 }
408 
409 #ifdef HAVE_SSE4
410 /* A version of the fast scanner using SSE 4.2 vectorized string insns.  */
411 
412 static const uchar *
413 #ifndef __SSE4_2__
414 __attribute__((__target__("sse4.2")))
415 #endif
416 search_line_sse42 (const uchar *s, const uchar *end)
417 {
418   typedef char v16qi __attribute__ ((__vector_size__ (16)));
419   static const v16qi search = { '\n', '\r', '?', '\\' };
420 
421   uintptr_t si = (uintptr_t)s;
422   uintptr_t index;
423 
424   /* Check for unaligned input.  */
425   if (si & 15)
426     {
427       v16qi sv;
428 
429       if (__builtin_expect (end - s < 16, 0)
430 	  && __builtin_expect ((si & 0xfff) > 0xff0, 0))
431 	{
432 	  /* There are less than 16 bytes left in the buffer, and less
433 	     than 16 bytes left on the page.  Reading 16 bytes at this
434 	     point might generate a spurious page fault.  Defer to the
435 	     SSE2 implementation, which already handles alignment.  */
436 	  return search_line_sse2 (s, end);
437 	}
438 
439       /* ??? The builtin doesn't understand that the PCMPESTRI read from
440 	 memory need not be aligned.  */
441       sv = __builtin_ia32_loaddqu ((const char *) s);
442       index = __builtin_ia32_pcmpestri128 (search, 4, sv, 16, 0);
443 
444       if (__builtin_expect (index < 16, 0))
445 	goto found;
446 
447       /* Advance the pointer to an aligned address.  We will re-scan a
448 	 few bytes, but we no longer need care for reading past the
449 	 end of a page, since we're guaranteed a match.  */
450       s = (const uchar *)((si + 15) & -16);
451     }
452 
453   /* Main loop, processing 16 bytes at a time.  */
454 #ifdef __GCC_ASM_FLAG_OUTPUTS__
455   while (1)
456     {
457       char f;
458 
459       /* By using inline assembly instead of the builtin,
460 	 we can use the result, as well as the flags set.  */
461       __asm ("%vpcmpestri\t$0, %2, %3"
462 	     : "=c"(index), "=@ccc"(f)
463 	     : "m"(*s), "x"(search), "a"(4), "d"(16));
464       if (f)
465 	break;
466 
467       s += 16;
468     }
469 #else
470   s -= 16;
471   /* By doing the whole loop in inline assembly,
472      we can make proper use of the flags set.  */
473   __asm (      ".balign 16\n"
474 	"0:	add $16, %1\n"
475 	"	%vpcmpestri\t$0, (%1), %2\n"
476 	"	jnc 0b"
477 	: "=&c"(index), "+r"(s)
478 	: "x"(search), "a"(4), "d"(16));
479 #endif
480 
481  found:
482   return s + index;
483 }
484 
485 #else
486 /* Work around out-dated assemblers without sse4 support.  */
487 #define search_line_sse42 search_line_sse2
488 #endif
489 
490 /* Check the CPU capabilities.  */
491 
492 #include "../gcc/config/i386/cpuid.h"
493 
494 typedef const uchar * (*search_line_fast_type) (const uchar *, const uchar *);
495 static search_line_fast_type search_line_fast;
496 
497 #define HAVE_init_vectorized_lexer 1
498 static inline void
499 init_vectorized_lexer (void)
500 {
501   unsigned dummy, ecx = 0, edx = 0;
502   search_line_fast_type impl = search_line_acc_char;
503   int minimum = 0;
504 
505 #if defined(__SSE4_2__)
506   minimum = 3;
507 #elif defined(__SSE2__)
508   minimum = 2;
509 #elif defined(__SSE__)
510   minimum = 1;
511 #endif
512 
513   if (minimum == 3)
514     impl = search_line_sse42;
515   else if (__get_cpuid (1, &dummy, &dummy, &ecx, &edx) || minimum == 2)
516     {
517       if (minimum == 3 || (ecx & bit_SSE4_2))
518         impl = search_line_sse42;
519       else if (minimum == 2 || (edx & bit_SSE2))
520 	impl = search_line_sse2;
521       else if (minimum == 1 || (edx & bit_SSE))
522 	impl = search_line_mmx;
523     }
524   else if (__get_cpuid (0x80000001, &dummy, &dummy, &dummy, &edx))
525     {
526       if (minimum == 1
527 	  || (edx & (bit_MMXEXT | bit_CMOV)) == (bit_MMXEXT | bit_CMOV))
528 	impl = search_line_mmx;
529     }
530 
531   search_line_fast = impl;
532 }
533 
534 #elif defined(_ARCH_PWR8) && defined(__ALTIVEC__)
535 
536 /* A vection of the fast scanner using AltiVec vectorized byte compares
537    and VSX unaligned loads (when VSX is available).  This is otherwise
538    the same as the pre-GCC 5 version.  */
539 
540 ATTRIBUTE_NO_SANITIZE_UNDEFINED
541 static const uchar *
542 search_line_fast (const uchar *s, const uchar *end ATTRIBUTE_UNUSED)
543 {
544   typedef __attribute__((altivec(vector))) unsigned char vc;
545 
546   const vc repl_nl = {
547     '\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n',
548     '\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n'
549   };
550   const vc repl_cr = {
551     '\r', '\r', '\r', '\r', '\r', '\r', '\r', '\r',
552     '\r', '\r', '\r', '\r', '\r', '\r', '\r', '\r'
553   };
554   const vc repl_bs = {
555     '\\', '\\', '\\', '\\', '\\', '\\', '\\', '\\',
556     '\\', '\\', '\\', '\\', '\\', '\\', '\\', '\\'
557   };
558   const vc repl_qm = {
559     '?', '?', '?', '?', '?', '?', '?', '?',
560     '?', '?', '?', '?', '?', '?', '?', '?',
561   };
562   const vc zero = { 0 };
563 
564   vc data, t;
565 
566   /* Main loop processing 16 bytes at a time.  */
567   do
568     {
569       vc m_nl, m_cr, m_bs, m_qm;
570 
571       data = __builtin_vec_vsx_ld (0, s);
572       s += 16;
573 
574       m_nl = (vc) __builtin_vec_cmpeq(data, repl_nl);
575       m_cr = (vc) __builtin_vec_cmpeq(data, repl_cr);
576       m_bs = (vc) __builtin_vec_cmpeq(data, repl_bs);
577       m_qm = (vc) __builtin_vec_cmpeq(data, repl_qm);
578       t = (m_nl | m_cr) | (m_bs | m_qm);
579 
580       /* T now contains 0xff in bytes for which we matched one of the relevant
581 	 characters.  We want to exit the loop if any byte in T is non-zero.
582 	 Below is the expansion of vec_any_ne(t, zero).  */
583     }
584   while (!__builtin_vec_vcmpeq_p(/*__CR6_LT_REV*/3, t, zero));
585 
586   /* Restore s to to point to the 16 bytes we just processed.  */
587   s -= 16;
588 
589   {
590 #define N  (sizeof(vc) / sizeof(long))
591 
592     union {
593       vc v;
594       /* Statically assert that N is 2 or 4.  */
595       unsigned long l[(N == 2 || N == 4) ? N : -1];
596     } u;
597     unsigned long l, i = 0;
598 
599     u.v = t;
600 
601     /* Find the first word of T that is non-zero.  */
602     switch (N)
603       {
604       case 4:
605 	l = u.l[i++];
606 	if (l != 0)
607 	  break;
608 	s += sizeof(unsigned long);
609 	l = u.l[i++];
610 	if (l != 0)
611 	  break;
612 	s += sizeof(unsigned long);
613 	/* FALLTHRU */
614       case 2:
615 	l = u.l[i++];
616 	if (l != 0)
617 	  break;
618 	s += sizeof(unsigned long);
619 	l = u.l[i];
620       }
621 
622     /* L now contains 0xff in bytes for which we matched one of the
623        relevant characters.  We can find the byte index by finding
624        its bit index and dividing by 8.  */
625 #ifdef __BIG_ENDIAN__
626     l = __builtin_clzl(l) >> 3;
627 #else
628     l = __builtin_ctzl(l) >> 3;
629 #endif
630     return s + l;
631 
632 #undef N
633   }
634 }
635 
636 #elif (GCC_VERSION >= 4005) && defined(__ALTIVEC__) && defined (__BIG_ENDIAN__)
637 
638 /* A vection of the fast scanner using AltiVec vectorized byte compares.
639    This cannot be used for little endian because vec_lvsl/lvsr are
640    deprecated for little endian and the code won't work properly.  */
641 /* ??? Unfortunately, attribute(target("altivec")) is not yet supported,
642    so we can't compile this function without -maltivec on the command line
643    (or implied by some other switch).  */
644 
645 static const uchar *
646 search_line_fast (const uchar *s, const uchar *end ATTRIBUTE_UNUSED)
647 {
648   typedef __attribute__((altivec(vector))) unsigned char vc;
649 
650   const vc repl_nl = {
651     '\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n',
652     '\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n'
653   };
654   const vc repl_cr = {
655     '\r', '\r', '\r', '\r', '\r', '\r', '\r', '\r',
656     '\r', '\r', '\r', '\r', '\r', '\r', '\r', '\r'
657   };
658   const vc repl_bs = {
659     '\\', '\\', '\\', '\\', '\\', '\\', '\\', '\\',
660     '\\', '\\', '\\', '\\', '\\', '\\', '\\', '\\'
661   };
662   const vc repl_qm = {
663     '?', '?', '?', '?', '?', '?', '?', '?',
664     '?', '?', '?', '?', '?', '?', '?', '?',
665   };
666   const vc ones = {
667     -1, -1, -1, -1, -1, -1, -1, -1,
668     -1, -1, -1, -1, -1, -1, -1, -1,
669   };
670   const vc zero = { 0 };
671 
672   vc data, mask, t;
673 
674   /* Altivec loads automatically mask addresses with -16.  This lets us
675      issue the first load as early as possible.  */
676   data = __builtin_vec_ld(0, (const vc *)s);
677 
678   /* Discard bytes before the beginning of the buffer.  Do this by
679      beginning with all ones and shifting in zeros according to the
680      mis-alignment.  The LVSR instruction pulls the exact shift we
681      want from the address.  */
682   mask = __builtin_vec_lvsr(0, s);
683   mask = __builtin_vec_perm(zero, ones, mask);
684   data &= mask;
685 
686   /* While altivec loads mask addresses, we still need to align S so
687      that the offset we compute at the end is correct.  */
688   s = (const uchar *)((uintptr_t)s & -16);
689 
690   /* Main loop processing 16 bytes at a time.  */
691   goto start;
692   do
693     {
694       vc m_nl, m_cr, m_bs, m_qm;
695 
696       s += 16;
697       data = __builtin_vec_ld(0, (const vc *)s);
698 
699     start:
700       m_nl = (vc) __builtin_vec_cmpeq(data, repl_nl);
701       m_cr = (vc) __builtin_vec_cmpeq(data, repl_cr);
702       m_bs = (vc) __builtin_vec_cmpeq(data, repl_bs);
703       m_qm = (vc) __builtin_vec_cmpeq(data, repl_qm);
704       t = (m_nl | m_cr) | (m_bs | m_qm);
705 
706       /* T now contains 0xff in bytes for which we matched one of the relevant
707 	 characters.  We want to exit the loop if any byte in T is non-zero.
708 	 Below is the expansion of vec_any_ne(t, zero).  */
709     }
710   while (!__builtin_vec_vcmpeq_p(/*__CR6_LT_REV*/3, t, zero));
711 
712   {
713 #define N  (sizeof(vc) / sizeof(long))
714 
715     union {
716       vc v;
717       /* Statically assert that N is 2 or 4.  */
718       unsigned long l[(N == 2 || N == 4) ? N : -1];
719     } u;
720     unsigned long l, i = 0;
721 
722     u.v = t;
723 
724     /* Find the first word of T that is non-zero.  */
725     switch (N)
726       {
727       case 4:
728 	l = u.l[i++];
729 	if (l != 0)
730 	  break;
731 	s += sizeof(unsigned long);
732 	l = u.l[i++];
733 	if (l != 0)
734 	  break;
735 	s += sizeof(unsigned long);
736 	/* FALLTHROUGH */
737       case 2:
738 	l = u.l[i++];
739 	if (l != 0)
740 	  break;
741 	s += sizeof(unsigned long);
742 	l = u.l[i];
743       }
744 
745     /* L now contains 0xff in bytes for which we matched one of the
746        relevant characters.  We can find the byte index by finding
747        its bit index and dividing by 8.  */
748     l = __builtin_clzl(l) >> 3;
749     return s + l;
750 
751 #undef N
752   }
753 }
754 
755 #elif defined (__ARM_NEON) && defined (__ARM_64BIT_STATE)
756 #include "arm_neon.h"
757 
758 /* This doesn't have to be the exact page size, but no system may use
759    a size smaller than this.  ARMv8 requires a minimum page size of
760    4k.  The impact of being conservative here is a small number of
761    cases will take the slightly slower entry path into the main
762    loop.  */
763 
764 #define AARCH64_MIN_PAGE_SIZE 4096
765 
766 static const uchar *
767 search_line_fast (const uchar *s, const uchar *end ATTRIBUTE_UNUSED)
768 {
769   const uint8x16_t repl_nl = vdupq_n_u8 ('\n');
770   const uint8x16_t repl_cr = vdupq_n_u8 ('\r');
771   const uint8x16_t repl_bs = vdupq_n_u8 ('\\');
772   const uint8x16_t repl_qm = vdupq_n_u8 ('?');
773   const uint8x16_t xmask = (uint8x16_t) vdupq_n_u64 (0x8040201008040201ULL);
774 
775 #ifdef __ARM_BIG_ENDIAN
776   const int16x8_t shift = {8, 8, 8, 8, 0, 0, 0, 0};
777 #else
778   const int16x8_t shift = {0, 0, 0, 0, 8, 8, 8, 8};
779 #endif
780 
781   unsigned int found;
782   const uint8_t *p;
783   uint8x16_t data;
784   uint8x16_t t;
785   uint16x8_t m;
786   uint8x16_t u, v, w;
787 
788   /* Align the source pointer.  */
789   p = (const uint8_t *)((uintptr_t)s & -16);
790 
791   /* Assuming random string start positions, with a 4k page size we'll take
792      the slow path about 0.37% of the time.  */
793   if (__builtin_expect ((AARCH64_MIN_PAGE_SIZE
794 			 - (((uintptr_t) s) & (AARCH64_MIN_PAGE_SIZE - 1)))
795 			< 16, 0))
796     {
797       /* Slow path: the string starts near a possible page boundary.  */
798       uint32_t misalign, mask;
799 
800       misalign = (uintptr_t)s & 15;
801       mask = (-1u << misalign) & 0xffff;
802       data = vld1q_u8 (p);
803       t = vceqq_u8 (data, repl_nl);
804       u = vceqq_u8 (data, repl_cr);
805       v = vorrq_u8 (t, vceqq_u8 (data, repl_bs));
806       w = vorrq_u8 (u, vceqq_u8 (data, repl_qm));
807       t = vorrq_u8 (v, w);
808       t = vandq_u8 (t, xmask);
809       m = vpaddlq_u8 (t);
810       m = vshlq_u16 (m, shift);
811       found = vaddvq_u16 (m);
812       found &= mask;
813       if (found)
814 	return (const uchar*)p + __builtin_ctz (found);
815     }
816   else
817     {
818       data = vld1q_u8 ((const uint8_t *) s);
819       t = vceqq_u8 (data, repl_nl);
820       u = vceqq_u8 (data, repl_cr);
821       v = vorrq_u8 (t, vceqq_u8 (data, repl_bs));
822       w = vorrq_u8 (u, vceqq_u8 (data, repl_qm));
823       t = vorrq_u8 (v, w);
824       if (__builtin_expect (vpaddd_u64 ((uint64x2_t)t) != 0, 0))
825 	goto done;
826     }
827 
828   do
829     {
830       p += 16;
831       data = vld1q_u8 (p);
832       t = vceqq_u8 (data, repl_nl);
833       u = vceqq_u8 (data, repl_cr);
834       v = vorrq_u8 (t, vceqq_u8 (data, repl_bs));
835       w = vorrq_u8 (u, vceqq_u8 (data, repl_qm));
836       t = vorrq_u8 (v, w);
837     } while (!vpaddd_u64 ((uint64x2_t)t));
838 
839 done:
840   /* Now that we've found the terminating substring, work out precisely where
841      we need to stop.  */
842   t = vandq_u8 (t, xmask);
843   m = vpaddlq_u8 (t);
844   m = vshlq_u16 (m, shift);
845   found = vaddvq_u16 (m);
846   return (((((uintptr_t) p) < (uintptr_t) s) ? s : (const uchar *)p)
847 	  + __builtin_ctz (found));
848 }
849 
850 #elif defined (__ARM_NEON)
851 #include "arm_neon.h"
852 
853 static const uchar *
854 search_line_fast (const uchar *s, const uchar *end ATTRIBUTE_UNUSED)
855 {
856   const uint8x16_t repl_nl = vdupq_n_u8 ('\n');
857   const uint8x16_t repl_cr = vdupq_n_u8 ('\r');
858   const uint8x16_t repl_bs = vdupq_n_u8 ('\\');
859   const uint8x16_t repl_qm = vdupq_n_u8 ('?');
860   const uint8x16_t xmask = (uint8x16_t) vdupq_n_u64 (0x8040201008040201ULL);
861 
862   unsigned int misalign, found, mask;
863   const uint8_t *p;
864   uint8x16_t data;
865 
866   /* Align the source pointer.  */
867   misalign = (uintptr_t)s & 15;
868   p = (const uint8_t *)((uintptr_t)s & -16);
869   data = vld1q_u8 (p);
870 
871   /* Create a mask for the bytes that are valid within the first
872      16-byte block.  The Idea here is that the AND with the mask
873      within the loop is "free", since we need some AND or TEST
874      insn in order to set the flags for the branch anyway.  */
875   mask = (-1u << misalign) & 0xffff;
876 
877   /* Main loop, processing 16 bytes at a time.  */
878   goto start;
879 
880   do
881     {
882       uint8x8_t l;
883       uint16x4_t m;
884       uint32x2_t n;
885       uint8x16_t t, u, v, w;
886 
887       p += 16;
888       data = vld1q_u8 (p);
889       mask = 0xffff;
890 
891     start:
892       t = vceqq_u8 (data, repl_nl);
893       u = vceqq_u8 (data, repl_cr);
894       v = vorrq_u8 (t, vceqq_u8 (data, repl_bs));
895       w = vorrq_u8 (u, vceqq_u8 (data, repl_qm));
896       t = vandq_u8 (vorrq_u8 (v, w), xmask);
897       l = vpadd_u8 (vget_low_u8 (t), vget_high_u8 (t));
898       m = vpaddl_u8 (l);
899       n = vpaddl_u16 (m);
900 
901       found = vget_lane_u32 ((uint32x2_t) vorr_u64 ((uint64x1_t) n,
902 	      vshr_n_u64 ((uint64x1_t) n, 24)), 0);
903       found &= mask;
904     }
905   while (!found);
906 
907   /* FOUND contains 1 in bits for which we matched a relevant
908      character.  Conversion to the byte index is trivial.  */
909   found = __builtin_ctz (found);
910   return (const uchar *)p + found;
911 }
912 
913 #else
914 
915 /* We only have one accelerated alternative.  Use a direct call so that
916    we encourage inlining.  */
917 
918 #define search_line_fast  search_line_acc_char
919 
920 #endif
921 
922 /* Initialize the lexer if needed.  */
923 
924 void
925 _cpp_init_lexer (void)
926 {
927 #ifdef HAVE_init_vectorized_lexer
928   init_vectorized_lexer ();
929 #endif
930 }
931 
932 /* Returns with a logical line that contains no escaped newlines or
933    trigraphs.  This is a time-critical inner loop.  */
934 void
935 _cpp_clean_line (cpp_reader *pfile)
936 {
937   cpp_buffer *buffer;
938   const uchar *s;
939   uchar c, *d, *p;
940 
941   buffer = pfile->buffer;
942   buffer->cur_note = buffer->notes_used = 0;
943   buffer->cur = buffer->line_base = buffer->next_line;
944   buffer->need_line = false;
945   s = buffer->next_line;
946 
947   if (!buffer->from_stage3)
948     {
949       const uchar *pbackslash = NULL;
950 
951       /* Fast path.  This is the common case of an un-escaped line with
952 	 no trigraphs.  The primary win here is by not writing any
953 	 data back to memory until we have to.  */
954       while (1)
955 	{
956 	  /* Perform an optimized search for \n, \r, \\, ?.  */
957 	  s = search_line_fast (s, buffer->rlimit);
958 
959 	  c = *s;
960 	  if (c == '\\')
961 	    {
962 	      /* Record the location of the backslash and continue.  */
963 	      pbackslash = s++;
964 	    }
965 	  else if (__builtin_expect (c == '?', 0))
966 	    {
967 	      if (__builtin_expect (s[1] == '?', false)
968 		   && _cpp_trigraph_map[s[2]])
969 		{
970 		  /* Have a trigraph.  We may or may not have to convert
971 		     it.  Add a line note regardless, for -Wtrigraphs.  */
972 		  add_line_note (buffer, s, s[2]);
973 		  if (CPP_OPTION (pfile, trigraphs))
974 		    {
975 		      /* We do, and that means we have to switch to the
976 		         slow path.  */
977 		      d = (uchar *) s;
978 		      *d = _cpp_trigraph_map[s[2]];
979 		      s += 2;
980 		      goto slow_path;
981 		    }
982 		}
983 	      /* Not a trigraph.  Continue on fast-path.  */
984 	      s++;
985 	    }
986 	  else
987 	    break;
988 	}
989 
990       /* This must be \r or \n.  We're either done, or we'll be forced
991 	 to write back to the buffer and continue on the slow path.  */
992       d = (uchar *) s;
993 
994       if (__builtin_expect (s == buffer->rlimit, false))
995 	goto done;
996 
997       /* DOS line ending? */
998       if (__builtin_expect (c == '\r', false) && s[1] == '\n')
999 	{
1000 	  s++;
1001 	  if (s == buffer->rlimit)
1002 	    goto done;
1003 	}
1004 
1005       if (__builtin_expect (pbackslash == NULL, true))
1006 	goto done;
1007 
1008       /* Check for escaped newline.  */
1009       p = d;
1010       while (is_nvspace (p[-1]))
1011 	p--;
1012       if (p - 1 != pbackslash)
1013 	goto done;
1014 
1015       /* Have an escaped newline; process it and proceed to
1016 	 the slow path.  */
1017       add_line_note (buffer, p - 1, p != d ? ' ' : '\\');
1018       d = p - 2;
1019       buffer->next_line = p - 1;
1020 
1021     slow_path:
1022       while (1)
1023 	{
1024 	  c = *++s;
1025 	  *++d = c;
1026 
1027 	  if (c == '\n' || c == '\r')
1028 	    {
1029 	      /* Handle DOS line endings.  */
1030 	      if (c == '\r' && s != buffer->rlimit && s[1] == '\n')
1031 		s++;
1032 	      if (s == buffer->rlimit)
1033 		break;
1034 
1035 	      /* Escaped?  */
1036 	      p = d;
1037 	      while (p != buffer->next_line && is_nvspace (p[-1]))
1038 		p--;
1039 	      if (p == buffer->next_line || p[-1] != '\\')
1040 		break;
1041 
1042 	      add_line_note (buffer, p - 1, p != d ? ' ': '\\');
1043 	      d = p - 2;
1044 	      buffer->next_line = p - 1;
1045 	    }
1046 	  else if (c == '?' && s[1] == '?' && _cpp_trigraph_map[s[2]])
1047 	    {
1048 	      /* Add a note regardless, for the benefit of -Wtrigraphs.  */
1049 	      add_line_note (buffer, d, s[2]);
1050 	      if (CPP_OPTION (pfile, trigraphs))
1051 		{
1052 		  *d = _cpp_trigraph_map[s[2]];
1053 		  s += 2;
1054 		}
1055 	    }
1056 	}
1057     }
1058   else
1059     {
1060       while (*s != '\n' && *s != '\r')
1061 	s++;
1062       d = (uchar *) s;
1063 
1064       /* Handle DOS line endings.  */
1065       if (*s == '\r' && s != buffer->rlimit && s[1] == '\n')
1066 	s++;
1067     }
1068 
1069  done:
1070   *d = '\n';
1071   /* A sentinel note that should never be processed.  */
1072   add_line_note (buffer, d + 1, '\n');
1073   buffer->next_line = s + 1;
1074 }
1075 
1076 /* Return true if the trigraph indicated by NOTE should be warned
1077    about in a comment.  */
1078 static bool
1079 warn_in_comment (cpp_reader *pfile, _cpp_line_note *note)
1080 {
1081   const uchar *p;
1082 
1083   /* Within comments we don't warn about trigraphs, unless the
1084      trigraph forms an escaped newline, as that may change
1085      behavior.  */
1086   if (note->type != '/')
1087     return false;
1088 
1089   /* If -trigraphs, then this was an escaped newline iff the next note
1090      is coincident.  */
1091   if (CPP_OPTION (pfile, trigraphs))
1092     return note[1].pos == note->pos;
1093 
1094   /* Otherwise, see if this forms an escaped newline.  */
1095   p = note->pos + 3;
1096   while (is_nvspace (*p))
1097     p++;
1098 
1099   /* There might have been escaped newlines between the trigraph and the
1100      newline we found.  Hence the position test.  */
1101   return (*p == '\n' && p < note[1].pos);
1102 }
1103 
1104 /* Process the notes created by add_line_note as far as the current
1105    location.  */
1106 void
1107 _cpp_process_line_notes (cpp_reader *pfile, int in_comment)
1108 {
1109   cpp_buffer *buffer = pfile->buffer;
1110 
1111   for (;;)
1112     {
1113       _cpp_line_note *note = &buffer->notes[buffer->cur_note];
1114       unsigned int col;
1115 
1116       if (note->pos > buffer->cur)
1117 	break;
1118 
1119       buffer->cur_note++;
1120       col = CPP_BUF_COLUMN (buffer, note->pos + 1);
1121 
1122       if (note->type == '\\' || note->type == ' ')
1123 	{
1124 	  if (note->type == ' ' && !in_comment)
1125 	    cpp_error_with_line (pfile, CPP_DL_WARNING, pfile->line_table->highest_line, col,
1126 				 "backslash and newline separated by space");
1127 
1128 	  if (buffer->next_line > buffer->rlimit)
1129 	    {
1130 	      cpp_error_with_line (pfile, CPP_DL_PEDWARN, pfile->line_table->highest_line, col,
1131 				   "backslash-newline at end of file");
1132 	      /* Prevent "no newline at end of file" warning.  */
1133 	      buffer->next_line = buffer->rlimit;
1134 	    }
1135 
1136 	  buffer->line_base = note->pos;
1137 	  CPP_INCREMENT_LINE (pfile, 0);
1138 	}
1139       else if (_cpp_trigraph_map[note->type])
1140 	{
1141 	  if (CPP_OPTION (pfile, warn_trigraphs)
1142 	      && (!in_comment || warn_in_comment (pfile, note)))
1143 	    {
1144 	      if (CPP_OPTION (pfile, trigraphs))
1145 		cpp_warning_with_line (pfile, CPP_W_TRIGRAPHS,
1146                                        pfile->line_table->highest_line, col,
1147 				       "trigraph ??%c converted to %c",
1148 				       note->type,
1149 				       (int) _cpp_trigraph_map[note->type]);
1150 	      else
1151 		{
1152 		  cpp_warning_with_line
1153 		    (pfile, CPP_W_TRIGRAPHS,
1154                      pfile->line_table->highest_line, col,
1155 		     "trigraph ??%c ignored, use -trigraphs to enable",
1156 		     note->type);
1157 		}
1158 	    }
1159 	}
1160       else if (note->type == 0)
1161 	/* Already processed in lex_raw_string.  */;
1162       else
1163 	abort ();
1164     }
1165 }
1166 
1167 /* Skip a C-style block comment.  We find the end of the comment by
1168    seeing if an asterisk is before every '/' we encounter.  Returns
1169    nonzero if comment terminated by EOF, zero otherwise.
1170 
1171    Buffer->cur points to the initial asterisk of the comment.  */
1172 bool
1173 _cpp_skip_block_comment (cpp_reader *pfile)
1174 {
1175   cpp_buffer *buffer = pfile->buffer;
1176   const uchar *cur = buffer->cur;
1177   uchar c;
1178 
1179   cur++;
1180   if (*cur == '/')
1181     cur++;
1182 
1183   for (;;)
1184     {
1185       /* People like decorating comments with '*', so check for '/'
1186 	 instead for efficiency.  */
1187       c = *cur++;
1188 
1189       if (c == '/')
1190 	{
1191 	  if (cur[-2] == '*')
1192 	    break;
1193 
1194 	  /* Warn about potential nested comments, but not if the '/'
1195 	     comes immediately before the true comment delimiter.
1196 	     Don't bother to get it right across escaped newlines.  */
1197 	  if (CPP_OPTION (pfile, warn_comments)
1198 	      && cur[0] == '*' && cur[1] != '/')
1199 	    {
1200 	      buffer->cur = cur;
1201 	      cpp_warning_with_line (pfile, CPP_W_COMMENTS,
1202 				     pfile->line_table->highest_line,
1203 				     CPP_BUF_COL (buffer),
1204 				     "\"/*\" within comment");
1205 	    }
1206 	}
1207       else if (c == '\n')
1208 	{
1209 	  unsigned int cols;
1210 	  buffer->cur = cur - 1;
1211 	  _cpp_process_line_notes (pfile, true);
1212 	  if (buffer->next_line >= buffer->rlimit)
1213 	    return true;
1214 	  _cpp_clean_line (pfile);
1215 
1216 	  cols = buffer->next_line - buffer->line_base;
1217 	  CPP_INCREMENT_LINE (pfile, cols);
1218 
1219 	  cur = buffer->cur;
1220 	}
1221     }
1222 
1223   buffer->cur = cur;
1224   _cpp_process_line_notes (pfile, true);
1225   return false;
1226 }
1227 
1228 /* Skip a C++ line comment, leaving buffer->cur pointing to the
1229    terminating newline.  Handles escaped newlines.  Returns nonzero
1230    if a multiline comment.  */
1231 static int
1232 skip_line_comment (cpp_reader *pfile)
1233 {
1234   cpp_buffer *buffer = pfile->buffer;
1235   source_location orig_line = pfile->line_table->highest_line;
1236 
1237   while (*buffer->cur != '\n')
1238     buffer->cur++;
1239 
1240   _cpp_process_line_notes (pfile, true);
1241   return orig_line != pfile->line_table->highest_line;
1242 }
1243 
1244 /* Skips whitespace, saving the next non-whitespace character.  */
1245 static void
1246 skip_whitespace (cpp_reader *pfile, cppchar_t c)
1247 {
1248   cpp_buffer *buffer = pfile->buffer;
1249   bool saw_NUL = false;
1250 
1251   do
1252     {
1253       /* Horizontal space always OK.  */
1254       if (c == ' ' || c == '\t')
1255 	;
1256       /* Just \f \v or \0 left.  */
1257       else if (c == '\0')
1258 	saw_NUL = true;
1259       else if (pfile->state.in_directive && CPP_PEDANTIC (pfile))
1260 	cpp_error_with_line (pfile, CPP_DL_PEDWARN, pfile->line_table->highest_line,
1261 			     CPP_BUF_COL (buffer),
1262 			     "%s in preprocessing directive",
1263 			     c == '\f' ? "form feed" : "vertical tab");
1264 
1265       c = *buffer->cur++;
1266     }
1267   /* We only want non-vertical space, i.e. ' ' \t \f \v \0.  */
1268   while (is_nvspace (c));
1269 
1270   if (saw_NUL)
1271     cpp_error (pfile, CPP_DL_WARNING, "null character(s) ignored");
1272 
1273   buffer->cur--;
1274 }
1275 
1276 /* See if the characters of a number token are valid in a name (no
1277    '.', '+' or '-').  */
1278 static int
1279 name_p (cpp_reader *pfile, const cpp_string *string)
1280 {
1281   unsigned int i;
1282 
1283   for (i = 0; i < string->len; i++)
1284     if (!is_idchar (string->text[i]))
1285       return 0;
1286 
1287   return 1;
1288 }
1289 
1290 /* After parsing an identifier or other sequence, produce a warning about
1291    sequences not in NFC/NFKC.  */
1292 static void
1293 warn_about_normalization (cpp_reader *pfile,
1294 			  const cpp_token *token,
1295 			  const struct normalize_state *s)
1296 {
1297   if (CPP_OPTION (pfile, warn_normalize) < NORMALIZE_STATE_RESULT (s)
1298       && !pfile->state.skipping)
1299     {
1300       /* Make sure that the token is printed using UCNs, even
1301 	 if we'd otherwise happily print UTF-8.  */
1302       unsigned char *buf = XNEWVEC (unsigned char, cpp_token_len (token));
1303       size_t sz;
1304 
1305       sz = cpp_spell_token (pfile, token, buf, false) - buf;
1306       if (NORMALIZE_STATE_RESULT (s) == normalized_C)
1307 	cpp_warning_with_line (pfile, CPP_W_NORMALIZE, token->src_loc, 0,
1308 			       "`%.*s' is not in NFKC", (int) sz, buf);
1309       else
1310 	cpp_warning_with_line (pfile, CPP_W_NORMALIZE, token->src_loc, 0,
1311 			       "`%.*s' is not in NFC", (int) sz, buf);
1312       free (buf);
1313     }
1314 }
1315 
1316 /* Returns TRUE if the sequence starting at buffer->cur is invalid in
1317    an identifier.  FIRST is TRUE if this starts an identifier.  */
1318 static bool
1319 forms_identifier_p (cpp_reader *pfile, int first,
1320 		    struct normalize_state *state)
1321 {
1322   cpp_buffer *buffer = pfile->buffer;
1323 
1324   if (*buffer->cur == '$')
1325     {
1326       if (!CPP_OPTION (pfile, dollars_in_ident))
1327 	return false;
1328 
1329       buffer->cur++;
1330       if (CPP_OPTION (pfile, warn_dollars) && !pfile->state.skipping)
1331 	{
1332 	  CPP_OPTION (pfile, warn_dollars) = 0;
1333 	  cpp_error (pfile, CPP_DL_PEDWARN, "'$' in identifier or number");
1334 	}
1335 
1336       return true;
1337     }
1338 
1339   /* Is this a syntactically valid UCN?  */
1340   if (CPP_OPTION (pfile, extended_identifiers)
1341       && *buffer->cur == '\\'
1342       && (buffer->cur[1] == 'u' || buffer->cur[1] == 'U'))
1343     {
1344       cppchar_t s;
1345       buffer->cur += 2;
1346       if (_cpp_valid_ucn (pfile, &buffer->cur, buffer->rlimit, 1 + !first,
1347 			  state, &s, NULL, NULL))
1348 	return true;
1349       buffer->cur -= 2;
1350     }
1351 
1352   return false;
1353 }
1354 
1355 /* Helper function to get the cpp_hashnode of the identifier BASE.  */
1356 static cpp_hashnode *
1357 lex_identifier_intern (cpp_reader *pfile, const uchar *base)
1358 {
1359   cpp_hashnode *result;
1360   const uchar *cur;
1361   unsigned int len;
1362   unsigned int hash = HT_HASHSTEP (0, *base);
1363 
1364   cur = base + 1;
1365   while (ISIDNUM (*cur))
1366     {
1367       hash = HT_HASHSTEP (hash, *cur);
1368       cur++;
1369     }
1370   len = cur - base;
1371   hash = HT_HASHFINISH (hash, len);
1372   result = CPP_HASHNODE (ht_lookup_with_hash (pfile->hash_table,
1373 					      base, len, hash, HT_ALLOC));
1374 
1375   /* Rarely, identifiers require diagnostics when lexed.  */
1376   if (__builtin_expect ((result->flags & NODE_DIAGNOSTIC)
1377 			&& !pfile->state.skipping, 0))
1378     {
1379       /* It is allowed to poison the same identifier twice.  */
1380       if ((result->flags & NODE_POISONED) && !pfile->state.poisoned_ok)
1381 	cpp_error (pfile, CPP_DL_ERROR, "attempt to use poisoned \"%s\"",
1382 		   NODE_NAME (result));
1383 
1384       /* Constraint 6.10.3.5: __VA_ARGS__ should only appear in the
1385 	 replacement list of a variadic macro.  */
1386       if (result == pfile->spec_nodes.n__VA_ARGS__
1387 	  && !pfile->state.va_args_ok)
1388 	{
1389 	  if (CPP_OPTION (pfile, cplusplus))
1390 	    cpp_error (pfile, CPP_DL_PEDWARN,
1391 		       "__VA_ARGS__ can only appear in the expansion"
1392 		       " of a C++11 variadic macro");
1393 	  else
1394 	    cpp_error (pfile, CPP_DL_PEDWARN,
1395 		       "__VA_ARGS__ can only appear in the expansion"
1396 		       " of a C99 variadic macro");
1397 	}
1398 
1399       /* For -Wc++-compat, warn about use of C++ named operators.  */
1400       if (result->flags & NODE_WARN_OPERATOR)
1401 	cpp_warning (pfile, CPP_W_CXX_OPERATOR_NAMES,
1402 		     "identifier \"%s\" is a special operator name in C++",
1403 		     NODE_NAME (result));
1404     }
1405 
1406   return result;
1407 }
1408 
1409 /* Get the cpp_hashnode of an identifier specified by NAME in
1410    the current cpp_reader object.  If none is found, NULL is returned.  */
1411 cpp_hashnode *
1412 _cpp_lex_identifier (cpp_reader *pfile, const char *name)
1413 {
1414   cpp_hashnode *result;
1415   result = lex_identifier_intern (pfile, (uchar *) name);
1416   return result;
1417 }
1418 
1419 /* Lex an identifier starting at BUFFER->CUR - 1.  */
1420 static cpp_hashnode *
1421 lex_identifier (cpp_reader *pfile, const uchar *base, bool starts_ucn,
1422 		struct normalize_state *nst, cpp_hashnode **spelling)
1423 {
1424   cpp_hashnode *result;
1425   const uchar *cur;
1426   unsigned int len;
1427   unsigned int hash = HT_HASHSTEP (0, *base);
1428 
1429   cur = pfile->buffer->cur;
1430   if (! starts_ucn)
1431     {
1432       while (ISIDNUM (*cur))
1433 	{
1434 	  hash = HT_HASHSTEP (hash, *cur);
1435 	  cur++;
1436 	}
1437       NORMALIZE_STATE_UPDATE_IDNUM (nst, *(cur - 1));
1438     }
1439   pfile->buffer->cur = cur;
1440   if (starts_ucn || forms_identifier_p (pfile, false, nst))
1441     {
1442       /* Slower version for identifiers containing UCNs (or $).  */
1443       do {
1444 	while (ISIDNUM (*pfile->buffer->cur))
1445 	  {
1446 	    NORMALIZE_STATE_UPDATE_IDNUM (nst, *pfile->buffer->cur);
1447 	    pfile->buffer->cur++;
1448 	  }
1449       } while (forms_identifier_p (pfile, false, nst));
1450       result = _cpp_interpret_identifier (pfile, base,
1451 					  pfile->buffer->cur - base);
1452       *spelling = cpp_lookup (pfile, base, pfile->buffer->cur - base);
1453     }
1454   else
1455     {
1456       len = cur - base;
1457       hash = HT_HASHFINISH (hash, len);
1458 
1459       result = CPP_HASHNODE (ht_lookup_with_hash (pfile->hash_table,
1460 						  base, len, hash, HT_ALLOC));
1461       *spelling = result;
1462     }
1463 
1464   /* Rarely, identifiers require diagnostics when lexed.  */
1465   if (__builtin_expect ((result->flags & NODE_DIAGNOSTIC)
1466 			&& !pfile->state.skipping, 0))
1467     {
1468       /* It is allowed to poison the same identifier twice.  */
1469       if ((result->flags & NODE_POISONED) && !pfile->state.poisoned_ok)
1470 	cpp_error (pfile, CPP_DL_ERROR, "attempt to use poisoned \"%s\"",
1471 		   NODE_NAME (result));
1472 
1473       /* Constraint 6.10.3.5: __VA_ARGS__ should only appear in the
1474 	 replacement list of a variadic macro.  */
1475       if (result == pfile->spec_nodes.n__VA_ARGS__
1476 	  && !pfile->state.va_args_ok)
1477 	{
1478 	  if (CPP_OPTION (pfile, cplusplus))
1479 	    cpp_error (pfile, CPP_DL_PEDWARN,
1480 		       "__VA_ARGS__ can only appear in the expansion"
1481 		       " of a C++11 variadic macro");
1482 	  else
1483 	    cpp_error (pfile, CPP_DL_PEDWARN,
1484 		       "__VA_ARGS__ can only appear in the expansion"
1485 		       " of a C99 variadic macro");
1486 	}
1487 
1488       /* For -Wc++-compat, warn about use of C++ named operators.  */
1489       if (result->flags & NODE_WARN_OPERATOR)
1490 	cpp_warning (pfile, CPP_W_CXX_OPERATOR_NAMES,
1491 		     "identifier \"%s\" is a special operator name in C++",
1492 		     NODE_NAME (result));
1493     }
1494 
1495   return result;
1496 }
1497 
1498 /* Lex a number to NUMBER starting at BUFFER->CUR - 1.  */
1499 static void
1500 lex_number (cpp_reader *pfile, cpp_string *number,
1501 	    struct normalize_state *nst)
1502 {
1503   const uchar *cur;
1504   const uchar *base;
1505   uchar *dest;
1506 
1507   base = pfile->buffer->cur - 1;
1508   do
1509     {
1510       cur = pfile->buffer->cur;
1511 
1512       /* N.B. ISIDNUM does not include $.  */
1513       while (ISIDNUM (*cur) || *cur == '.' || DIGIT_SEP (*cur)
1514 	     || VALID_SIGN (*cur, cur[-1]))
1515 	{
1516 	  NORMALIZE_STATE_UPDATE_IDNUM (nst, *cur);
1517 	  cur++;
1518 	}
1519       /* A number can't end with a digit separator.  */
1520       while (cur > pfile->buffer->cur && DIGIT_SEP (cur[-1]))
1521 	--cur;
1522 
1523       pfile->buffer->cur = cur;
1524     }
1525   while (forms_identifier_p (pfile, false, nst));
1526 
1527   number->len = cur - base;
1528   dest = _cpp_unaligned_alloc (pfile, number->len + 1);
1529   memcpy (dest, base, number->len);
1530   dest[number->len] = '\0';
1531   number->text = dest;
1532 }
1533 
1534 /* Create a token of type TYPE with a literal spelling.  */
1535 static void
1536 create_literal (cpp_reader *pfile, cpp_token *token, const uchar *base,
1537 		unsigned int len, enum cpp_ttype type)
1538 {
1539   uchar *dest = _cpp_unaligned_alloc (pfile, len + 1);
1540 
1541   memcpy (dest, base, len);
1542   dest[len] = '\0';
1543   token->type = type;
1544   token->val.str.len = len;
1545   token->val.str.text = dest;
1546 }
1547 
1548 /* Subroutine of lex_raw_string: Append LEN chars from BASE to the buffer
1549    sequence from *FIRST_BUFF_P to LAST_BUFF_P.  */
1550 
1551 static void
1552 bufring_append (cpp_reader *pfile, const uchar *base, size_t len,
1553 		_cpp_buff **first_buff_p, _cpp_buff **last_buff_p)
1554 {
1555   _cpp_buff *first_buff = *first_buff_p;
1556   _cpp_buff *last_buff = *last_buff_p;
1557 
1558   if (first_buff == NULL)
1559     first_buff = last_buff = _cpp_get_buff (pfile, len);
1560   else if (len > BUFF_ROOM (last_buff))
1561     {
1562       size_t room = BUFF_ROOM (last_buff);
1563       memcpy (BUFF_FRONT (last_buff), base, room);
1564       BUFF_FRONT (last_buff) += room;
1565       base += room;
1566       len -= room;
1567       last_buff = _cpp_append_extend_buff (pfile, last_buff, len);
1568     }
1569 
1570   memcpy (BUFF_FRONT (last_buff), base, len);
1571   BUFF_FRONT (last_buff) += len;
1572 
1573   *first_buff_p = first_buff;
1574   *last_buff_p = last_buff;
1575 }
1576 
1577 
1578 /* Returns true if a macro has been defined.
1579    This might not work if compile with -save-temps,
1580    or preprocess separately from compilation.  */
1581 
1582 static bool
1583 is_macro(cpp_reader *pfile, const uchar *base)
1584 {
1585   const uchar *cur = base;
1586   if (! ISIDST (*cur))
1587     return false;
1588   unsigned int hash = HT_HASHSTEP (0, *cur);
1589   ++cur;
1590   while (ISIDNUM (*cur))
1591     {
1592       hash = HT_HASHSTEP (hash, *cur);
1593       ++cur;
1594     }
1595   hash = HT_HASHFINISH (hash, cur - base);
1596 
1597   cpp_hashnode *result = CPP_HASHNODE (ht_lookup_with_hash (pfile->hash_table,
1598 					base, cur - base, hash, HT_NO_INSERT));
1599 
1600   return !result ? false : (result->type == NT_MACRO);
1601 }
1602 
1603 
1604 /* Lexes a raw string.  The stored string contains the spelling, including
1605    double quotes, delimiter string, '(' and ')', any leading
1606    'L', 'u', 'U' or 'u8' and 'R' modifier.  It returns the type of the
1607    literal, or CPP_OTHER if it was not properly terminated.
1608 
1609    The spelling is NUL-terminated, but it is not guaranteed that this
1610    is the first NUL since embedded NULs are preserved.  */
1611 
1612 static void
1613 lex_raw_string (cpp_reader *pfile, cpp_token *token, const uchar *base,
1614 		const uchar *cur)
1615 {
1616   uchar raw_prefix[17];
1617   uchar temp_buffer[18];
1618   const uchar *orig_base;
1619   unsigned int raw_prefix_len = 0, raw_suffix_len = 0;
1620   enum raw_str_phase { RAW_STR_PREFIX, RAW_STR, RAW_STR_SUFFIX };
1621   raw_str_phase phase = RAW_STR_PREFIX;
1622   enum cpp_ttype type;
1623   size_t total_len = 0;
1624   /* Index into temp_buffer during phases other than RAW_STR,
1625      during RAW_STR phase 17 to tell BUF_APPEND that nothing should
1626      be appended to temp_buffer.  */
1627   size_t temp_buffer_len = 0;
1628   _cpp_buff *first_buff = NULL, *last_buff = NULL;
1629   size_t raw_prefix_start;
1630   _cpp_line_note *note = &pfile->buffer->notes[pfile->buffer->cur_note];
1631 
1632   type = (*base == 'L' ? CPP_WSTRING :
1633 	  *base == 'U' ? CPP_STRING32 :
1634 	  *base == 'u' ? (base[1] == '8' ? CPP_UTF8STRING : CPP_STRING16)
1635 	  : CPP_STRING);
1636 
1637 #define BUF_APPEND(STR,LEN)					\
1638       do {							\
1639 	bufring_append (pfile, (const uchar *)(STR), (LEN),	\
1640 			&first_buff, &last_buff);		\
1641 	total_len += (LEN);					\
1642 	if (__builtin_expect (temp_buffer_len < 17, 0)		\
1643 	    && (const uchar *)(STR) != base			\
1644 	    && (LEN) <= 2)					\
1645 	  {							\
1646 	    memcpy (temp_buffer + temp_buffer_len,		\
1647 		    (const uchar *)(STR), (LEN));		\
1648 	    temp_buffer_len += (LEN);				\
1649 	  }							\
1650       } while (0);
1651 
1652   orig_base = base;
1653   ++cur;
1654   raw_prefix_start = cur - base;
1655   for (;;)
1656     {
1657       cppchar_t c;
1658 
1659       /* If we previously performed any trigraph or line splicing
1660 	 transformations, undo them in between the opening and closing
1661 	 double quote.  */
1662       while (note->pos < cur)
1663 	++note;
1664       for (; note->pos == cur; ++note)
1665 	{
1666 	  switch (note->type)
1667 	    {
1668 	    case '\\':
1669 	    case ' ':
1670 	      /* Restore backslash followed by newline.  */
1671 	      BUF_APPEND (base, cur - base);
1672 	      base = cur;
1673 	      BUF_APPEND ("\\", 1);
1674 	    after_backslash:
1675 	      if (note->type == ' ')
1676 		{
1677 		  /* GNU backslash whitespace newline extension.  FIXME
1678 		     could be any sequence of non-vertical space.  When we
1679 		     can properly restore any such sequence, we should mark
1680 		     this note as handled so _cpp_process_line_notes
1681 		     doesn't warn.  */
1682 		  BUF_APPEND (" ", 1);
1683 		}
1684 
1685 	      BUF_APPEND ("\n", 1);
1686 	      break;
1687 
1688 	    case 0:
1689 	      /* Already handled.  */
1690 	      break;
1691 
1692 	    default:
1693 	      if (_cpp_trigraph_map[note->type])
1694 		{
1695 		  /* Don't warn about this trigraph in
1696 		     _cpp_process_line_notes, since trigraphs show up as
1697 		     trigraphs in raw strings.  */
1698 		  uchar type = note->type;
1699 		  note->type = 0;
1700 
1701 		  if (!CPP_OPTION (pfile, trigraphs))
1702 		    /* If we didn't convert the trigraph in the first
1703 		       place, don't do anything now either.  */
1704 		    break;
1705 
1706 		  BUF_APPEND (base, cur - base);
1707 		  base = cur;
1708 		  BUF_APPEND ("??", 2);
1709 
1710 		  /* ??/ followed by newline gets two line notes, one for
1711 		     the trigraph and one for the backslash/newline.  */
1712 		  if (type == '/' && note[1].pos == cur)
1713 		    {
1714 		      if (note[1].type != '\\'
1715 			  && note[1].type != ' ')
1716 			abort ();
1717 		      BUF_APPEND ("/", 1);
1718 		      ++note;
1719 		      goto after_backslash;
1720 		    }
1721 		  else
1722 		    {
1723 		      /* Skip the replacement character.  */
1724 		      base = ++cur;
1725 		      BUF_APPEND (&type, 1);
1726 		      c = type;
1727 		      goto check_c;
1728 		    }
1729 		}
1730 	      else
1731 		abort ();
1732 	      break;
1733 	    }
1734 	}
1735       c = *cur++;
1736       if (__builtin_expect (temp_buffer_len < 17, 0))
1737 	temp_buffer[temp_buffer_len++] = c;
1738 
1739      check_c:
1740       if (phase == RAW_STR_PREFIX)
1741 	{
1742 	  while (raw_prefix_len < temp_buffer_len)
1743 	    {
1744 	      raw_prefix[raw_prefix_len] = temp_buffer[raw_prefix_len];
1745 	      switch (raw_prefix[raw_prefix_len])
1746 		{
1747 		case ' ': case '(': case ')': case '\\': case '\t':
1748 		case '\v': case '\f': case '\n': default:
1749 		  break;
1750 		/* Basic source charset except the above chars.  */
1751 		case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
1752 		case 'g': case 'h': case 'i': case 'j': case 'k': case 'l':
1753 		case 'm': case 'n': case 'o': case 'p': case 'q': case 'r':
1754 		case 's': case 't': case 'u': case 'v': case 'w': case 'x':
1755 		case 'y': case 'z':
1756 		case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
1757 		case 'G': case 'H': case 'I': case 'J': case 'K': case 'L':
1758 		case 'M': case 'N': case 'O': case 'P': case 'Q': case 'R':
1759 		case 'S': case 'T': case 'U': case 'V': case 'W': case 'X':
1760 		case 'Y': case 'Z':
1761 		case '0': case '1': case '2': case '3': case '4': case '5':
1762 		case '6': case '7': case '8': case '9':
1763 		case '_': case '{': case '}': case '#': case '[': case ']':
1764 		case '<': case '>': case '%': case ':': case ';': case '.':
1765 		case '?': case '*': case '+': case '-': case '/': case '^':
1766 		case '&': case '|': case '~': case '!': case '=': case ',':
1767 		case '"': case '\'':
1768 		  if (raw_prefix_len < 16)
1769 		    {
1770 		      raw_prefix_len++;
1771 		      continue;
1772 		    }
1773 		  break;
1774 		}
1775 
1776 	      if (raw_prefix[raw_prefix_len] != '(')
1777 		{
1778 		  int col = CPP_BUF_COLUMN (pfile->buffer, cur) + 1;
1779 		  if (raw_prefix_len == 16)
1780 		    cpp_error_with_line (pfile, CPP_DL_ERROR, token->src_loc,
1781 					 col, "raw string delimiter longer "
1782 					      "than 16 characters");
1783 		  else if (raw_prefix[raw_prefix_len] == '\n')
1784 		    cpp_error_with_line (pfile, CPP_DL_ERROR, token->src_loc,
1785 					 col, "invalid new-line in raw "
1786 					      "string delimiter");
1787 		  else
1788 		    cpp_error_with_line (pfile, CPP_DL_ERROR, token->src_loc,
1789 					 col, "invalid character '%c' in "
1790 					      "raw string delimiter",
1791 					 (int) raw_prefix[raw_prefix_len]);
1792 		  pfile->buffer->cur = orig_base + raw_prefix_start - 1;
1793 		  create_literal (pfile, token, orig_base,
1794 				  raw_prefix_start - 1, CPP_OTHER);
1795 		  if (first_buff)
1796 		    _cpp_release_buff (pfile, first_buff);
1797 		  return;
1798 		}
1799 	      raw_prefix[raw_prefix_len] = '"';
1800 	      phase = RAW_STR;
1801 	      /* Nothing should be appended to temp_buffer during
1802 		 RAW_STR phase.  */
1803 	      temp_buffer_len = 17;
1804 	      break;
1805 	    }
1806 	  continue;
1807 	}
1808       else if (phase == RAW_STR_SUFFIX)
1809 	{
1810 	  while (raw_suffix_len <= raw_prefix_len
1811 		 && raw_suffix_len < temp_buffer_len
1812 		 && temp_buffer[raw_suffix_len] == raw_prefix[raw_suffix_len])
1813 	    raw_suffix_len++;
1814 	  if (raw_suffix_len > raw_prefix_len)
1815 	    break;
1816 	  if (raw_suffix_len == temp_buffer_len)
1817 	    continue;
1818 	  phase = RAW_STR;
1819 	  /* Nothing should be appended to temp_buffer during
1820 	     RAW_STR phase.  */
1821 	  temp_buffer_len = 17;
1822 	}
1823       if (c == ')')
1824 	{
1825 	  phase = RAW_STR_SUFFIX;
1826 	  raw_suffix_len = 0;
1827 	  temp_buffer_len = 0;
1828 	}
1829       else if (c == '\n')
1830 	{
1831 	  if (pfile->state.in_directive
1832 	      || (pfile->state.parsing_args
1833 		  && pfile->buffer->next_line >= pfile->buffer->rlimit))
1834 	    {
1835 	      cur--;
1836 	      type = CPP_OTHER;
1837 	      cpp_error_with_line (pfile, CPP_DL_ERROR, token->src_loc, 0,
1838 				   "unterminated raw string");
1839 	      break;
1840 	    }
1841 
1842 	  BUF_APPEND (base, cur - base);
1843 
1844 	  if (pfile->buffer->cur < pfile->buffer->rlimit)
1845 	    CPP_INCREMENT_LINE (pfile, 0);
1846 	  pfile->buffer->need_line = true;
1847 
1848 	  pfile->buffer->cur = cur-1;
1849 	  _cpp_process_line_notes (pfile, false);
1850 	  if (!_cpp_get_fresh_line (pfile))
1851 	    {
1852 	      source_location src_loc = token->src_loc;
1853 	      token->type = CPP_EOF;
1854 	      /* Tell the compiler the line number of the EOF token.  */
1855 	      token->src_loc = pfile->line_table->highest_line;
1856 	      token->flags = BOL;
1857 	      if (first_buff != NULL)
1858 		_cpp_release_buff (pfile, first_buff);
1859 	      cpp_error_with_line (pfile, CPP_DL_ERROR, src_loc, 0,
1860 				   "unterminated raw string");
1861 	      return;
1862 	    }
1863 
1864 	  cur = base = pfile->buffer->cur;
1865 	  note = &pfile->buffer->notes[pfile->buffer->cur_note];
1866 	}
1867     }
1868 
1869   if (CPP_OPTION (pfile, user_literals))
1870     {
1871       /* If a string format macro, say from inttypes.h, is placed touching
1872 	 a string literal it could be parsed as a C++11 user-defined string
1873 	 literal thus breaking the program.
1874 	 Try to identify macros with is_macro. A warning is issued. */
1875       if (is_macro (pfile, cur))
1876 	{
1877 	  /* Raise a warning, but do not consume subsequent tokens.  */
1878 	  if (CPP_OPTION (pfile, warn_literal_suffix) && !pfile->state.skipping)
1879 	    cpp_warning_with_line (pfile, CPP_W_LITERAL_SUFFIX,
1880 				   token->src_loc, 0,
1881 				   "invalid suffix on literal; C++11 requires "
1882 				   "a space between literal and string macro");
1883 	}
1884       /* Grab user defined literal suffix.  */
1885       else if (ISIDST (*cur))
1886 	{
1887 	  type = cpp_userdef_string_add_type (type);
1888 	  ++cur;
1889 
1890 	  while (ISIDNUM (*cur))
1891 	    ++cur;
1892 	}
1893     }
1894 
1895   pfile->buffer->cur = cur;
1896   if (first_buff == NULL)
1897     create_literal (pfile, token, base, cur - base, type);
1898   else
1899     {
1900       uchar *dest = _cpp_unaligned_alloc (pfile, total_len + (cur - base) + 1);
1901 
1902       token->type = type;
1903       token->val.str.len = total_len + (cur - base);
1904       token->val.str.text = dest;
1905       last_buff = first_buff;
1906       while (last_buff != NULL)
1907 	{
1908 	  memcpy (dest, last_buff->base,
1909 		  BUFF_FRONT (last_buff) - last_buff->base);
1910 	  dest += BUFF_FRONT (last_buff) - last_buff->base;
1911 	  last_buff = last_buff->next;
1912 	}
1913       _cpp_release_buff (pfile, first_buff);
1914       memcpy (dest, base, cur - base);
1915       dest[cur - base] = '\0';
1916     }
1917 }
1918 
1919 /* Lexes a string, character constant, or angle-bracketed header file
1920    name.  The stored string contains the spelling, including opening
1921    quote and any leading 'L', 'u', 'U' or 'u8' and optional
1922    'R' modifier.  It returns the type of the literal, or CPP_OTHER
1923    if it was not properly terminated, or CPP_LESS for an unterminated
1924    header name which must be relexed as normal tokens.
1925 
1926    The spelling is NUL-terminated, but it is not guaranteed that this
1927    is the first NUL since embedded NULs are preserved.  */
1928 static void
1929 lex_string (cpp_reader *pfile, cpp_token *token, const uchar *base)
1930 {
1931   bool saw_NUL = false;
1932   const uchar *cur;
1933   cppchar_t terminator;
1934   enum cpp_ttype type;
1935 
1936   cur = base;
1937   terminator = *cur++;
1938   if (terminator == 'L' || terminator == 'U')
1939     terminator = *cur++;
1940   else if (terminator == 'u')
1941     {
1942       terminator = *cur++;
1943       if (terminator == '8')
1944 	terminator = *cur++;
1945     }
1946   if (terminator == 'R')
1947     {
1948       lex_raw_string (pfile, token, base, cur);
1949       return;
1950     }
1951   if (terminator == '"')
1952     type = (*base == 'L' ? CPP_WSTRING :
1953 	    *base == 'U' ? CPP_STRING32 :
1954 	    *base == 'u' ? (base[1] == '8' ? CPP_UTF8STRING : CPP_STRING16)
1955 			 : CPP_STRING);
1956   else if (terminator == '\'')
1957     type = (*base == 'L' ? CPP_WCHAR :
1958 	    *base == 'U' ? CPP_CHAR32 :
1959 	    *base == 'u' ? (base[1] == '8' ? CPP_UTF8CHAR : CPP_CHAR16)
1960 			 : CPP_CHAR);
1961   else
1962     terminator = '>', type = CPP_HEADER_NAME;
1963 
1964   for (;;)
1965     {
1966       cppchar_t c = *cur++;
1967 
1968       /* In #include-style directives, terminators are not escapable.  */
1969       if (c == '\\' && !pfile->state.angled_headers && *cur != '\n')
1970 	cur++;
1971       else if (c == terminator)
1972 	break;
1973       else if (c == '\n')
1974 	{
1975 	  cur--;
1976 	  /* Unmatched quotes always yield undefined behavior, but
1977 	     greedy lexing means that what appears to be an unterminated
1978 	     header name may actually be a legitimate sequence of tokens.  */
1979 	  if (terminator == '>')
1980 	    {
1981 	      token->type = CPP_LESS;
1982 	      return;
1983 	    }
1984 	  type = CPP_OTHER;
1985 	  break;
1986 	}
1987       else if (c == '\0')
1988 	saw_NUL = true;
1989     }
1990 
1991   if (saw_NUL && !pfile->state.skipping)
1992     cpp_error (pfile, CPP_DL_WARNING,
1993 	       "null character(s) preserved in literal");
1994 
1995   if (type == CPP_OTHER && CPP_OPTION (pfile, lang) != CLK_ASM)
1996     cpp_error (pfile, CPP_DL_PEDWARN, "missing terminating %c character",
1997 	       (int) terminator);
1998 
1999   if (CPP_OPTION (pfile, user_literals))
2000     {
2001       /* If a string format macro, say from inttypes.h, is placed touching
2002 	 a string literal it could be parsed as a C++11 user-defined string
2003 	 literal thus breaking the program.
2004 	 Try to identify macros with is_macro. A warning is issued. */
2005       if (is_macro (pfile, cur))
2006 	{
2007 	  /* Raise a warning, but do not consume subsequent tokens.  */
2008 	  if (CPP_OPTION (pfile, warn_literal_suffix) && !pfile->state.skipping)
2009 	    cpp_warning_with_line (pfile, CPP_W_LITERAL_SUFFIX,
2010 				   token->src_loc, 0,
2011 				   "invalid suffix on literal; C++11 requires "
2012 				   "a space between literal and string macro");
2013 	}
2014       /* Grab user defined literal suffix.  */
2015       else if (ISIDST (*cur))
2016 	{
2017 	  type = cpp_userdef_char_add_type (type);
2018 	  type = cpp_userdef_string_add_type (type);
2019           ++cur;
2020 
2021 	  while (ISIDNUM (*cur))
2022 	    ++cur;
2023 	}
2024     }
2025   else if (CPP_OPTION (pfile, cpp_warn_cxx11_compat)
2026 	   && is_macro (pfile, cur)
2027 	   && !pfile->state.skipping)
2028     cpp_warning_with_line (pfile, CPP_W_CXX11_COMPAT,
2029 			   token->src_loc, 0, "C++11 requires a space "
2030 			   "between string literal and macro");
2031 
2032   pfile->buffer->cur = cur;
2033   create_literal (pfile, token, base, cur - base, type);
2034 }
2035 
2036 /* Return the comment table. The client may not make any assumption
2037    about the ordering of the table.  */
2038 cpp_comment_table *
2039 cpp_get_comments (cpp_reader *pfile)
2040 {
2041   return &pfile->comments;
2042 }
2043 
2044 /* Append a comment to the end of the comment table. */
2045 static void
2046 store_comment (cpp_reader *pfile, cpp_token *token)
2047 {
2048   int len;
2049 
2050   if (pfile->comments.allocated == 0)
2051     {
2052       pfile->comments.allocated = 256;
2053       pfile->comments.entries = (cpp_comment *) xmalloc
2054 	(pfile->comments.allocated * sizeof (cpp_comment));
2055     }
2056 
2057   if (pfile->comments.count == pfile->comments.allocated)
2058     {
2059       pfile->comments.allocated *= 2;
2060       pfile->comments.entries = (cpp_comment *) xrealloc
2061 	(pfile->comments.entries,
2062 	 pfile->comments.allocated * sizeof (cpp_comment));
2063     }
2064 
2065   len = token->val.str.len;
2066 
2067   /* Copy comment. Note, token may not be NULL terminated. */
2068   pfile->comments.entries[pfile->comments.count].comment =
2069     (char *) xmalloc (sizeof (char) * (len + 1));
2070   memcpy (pfile->comments.entries[pfile->comments.count].comment,
2071 	  token->val.str.text, len);
2072   pfile->comments.entries[pfile->comments.count].comment[len] = '\0';
2073 
2074   /* Set source location. */
2075   pfile->comments.entries[pfile->comments.count].sloc = token->src_loc;
2076 
2077   /* Increment the count of entries in the comment table. */
2078   pfile->comments.count++;
2079 }
2080 
2081 /* The stored comment includes the comment start and any terminator.  */
2082 static void
2083 save_comment (cpp_reader *pfile, cpp_token *token, const unsigned char *from,
2084 	      cppchar_t type)
2085 {
2086   unsigned char *buffer;
2087   unsigned int len, clen, i;
2088   int convert_to_c = (pfile->state.in_directive || pfile->state.parsing_args)
2089     && type == '/';
2090 
2091   len = pfile->buffer->cur - from + 1; /* + 1 for the initial '/'.  */
2092 
2093   /* C++ comments probably (not definitely) have moved past a new
2094      line, which we don't want to save in the comment.  */
2095   if (is_vspace (pfile->buffer->cur[-1]))
2096     len--;
2097 
2098   /* If we are currently in a directive or in argument parsing, then
2099      we need to store all C++ comments as C comments internally, and
2100      so we need to allocate a little extra space in that case.
2101 
2102      Note that the only time we encounter a directive here is
2103      when we are saving comments in a "#define".  */
2104   clen = convert_to_c ? len + 2 : len;
2105 
2106   buffer = _cpp_unaligned_alloc (pfile, clen);
2107 
2108   token->type = CPP_COMMENT;
2109   token->val.str.len = clen;
2110   token->val.str.text = buffer;
2111 
2112   buffer[0] = '/';
2113   memcpy (buffer + 1, from, len - 1);
2114 
2115   /* Finish conversion to a C comment, if necessary.  */
2116   if (convert_to_c)
2117     {
2118       buffer[1] = '*';
2119       buffer[clen - 2] = '*';
2120       buffer[clen - 1] = '/';
2121       /* As there can be in a C++ comments illegal sequences for C comments
2122          we need to filter them out.  */
2123       for (i = 2; i < (clen - 2); i++)
2124         if (buffer[i] == '/' && (buffer[i - 1] == '*' || buffer[i + 1] == '*'))
2125           buffer[i] = '|';
2126     }
2127 
2128   /* Finally store this comment for use by clients of libcpp. */
2129   store_comment (pfile, token);
2130 }
2131 
2132 /* Returns true if comment at COMMENT_START is a recognized FALLTHROUGH
2133    comment.  */
2134 
2135 static bool
2136 fallthrough_comment_p (cpp_reader *pfile, const unsigned char *comment_start)
2137 {
2138   const unsigned char *from = comment_start + 1;
2139 
2140   switch (CPP_OPTION (pfile, cpp_warn_implicit_fallthrough))
2141     {
2142       /* For both -Wimplicit-fallthrough=0 and -Wimplicit-fallthrough=5 we
2143 	 don't recognize any comments.  The latter only checks attributes,
2144 	 the former doesn't warn.  */
2145     case 0:
2146     default:
2147       return false;
2148       /* -Wimplicit-fallthrough=1 considers any comment, no matter what
2149 	 content it has.  */
2150     case 1:
2151       return true;
2152     case 2:
2153       /* -Wimplicit-fallthrough=2 looks for (case insensitive)
2154 	 .*falls?[ \t-]*thr(u|ough).* regex.  */
2155       for (; (size_t) (pfile->buffer->cur - from) >= sizeof "fallthru" - 1;
2156 	   from++)
2157 	{
2158 	  /* Is there anything like strpbrk with upper boundary, or
2159 	     memchr looking for 2 characters rather than just one?  */
2160 	  if (from[0] != 'f' && from[0] != 'F')
2161 	    continue;
2162 	  if (from[1] != 'a' && from[1] != 'A')
2163 	    continue;
2164 	  if (from[2] != 'l' && from[2] != 'L')
2165 	    continue;
2166 	  if (from[3] != 'l' && from[3] != 'L')
2167 	    continue;
2168 	  from += sizeof "fall" - 1;
2169 	  if (from[0] == 's' || from[0] == 'S')
2170 	    from++;
2171 	  while (*from == ' ' || *from == '\t' || *from == '-')
2172 	    from++;
2173 	  if (from[0] != 't' && from[0] != 'T')
2174 	    continue;
2175 	  if (from[1] != 'h' && from[1] != 'H')
2176 	    continue;
2177 	  if (from[2] != 'r' && from[2] != 'R')
2178 	    continue;
2179 	  if (from[3] == 'u' || from[3] == 'U')
2180 	    return true;
2181 	  if (from[3] != 'o' && from[3] != 'O')
2182 	    continue;
2183 	  if (from[4] != 'u' && from[4] != 'U')
2184 	    continue;
2185 	  if (from[5] != 'g' && from[5] != 'G')
2186 	    continue;
2187 	  if (from[6] != 'h' && from[6] != 'H')
2188 	    continue;
2189 	  return true;
2190 	}
2191       return false;
2192     case 3:
2193     case 4:
2194       break;
2195     }
2196 
2197   /* Whole comment contents:
2198      -fallthrough
2199      @fallthrough@
2200    */
2201   if (*from == '-' || *from == '@')
2202     {
2203       size_t len = sizeof "fallthrough" - 1;
2204       if ((size_t) (pfile->buffer->cur - from - 1) < len)
2205 	return false;
2206       if (memcmp (from + 1, "fallthrough", len))
2207 	return false;
2208       if (*from == '@')
2209 	{
2210 	  if (from[len + 1] != '@')
2211 	    return false;
2212 	  len++;
2213 	}
2214       from += 1 + len;
2215     }
2216   /* Whole comment contents (regex):
2217      lint -fallthrough[ \t]*
2218    */
2219   else if (*from == 'l')
2220     {
2221       size_t len = sizeof "int -fallthrough" - 1;
2222       if ((size_t) (pfile->buffer->cur - from - 1) < len)
2223 	return false;
2224       if (memcmp (from + 1, "int -fallthrough", len))
2225 	return false;
2226       from += 1 + len;
2227       while (*from == ' ' || *from == '\t')
2228 	from++;
2229     }
2230   /* Whole comment contents (regex):
2231      [ \t]*FALLTHR(U|OUGH)[ \t]*
2232    */
2233   else if (CPP_OPTION (pfile, cpp_warn_implicit_fallthrough) == 4)
2234     {
2235       while (*from == ' ' || *from == '\t')
2236 	from++;
2237       if ((size_t) (pfile->buffer->cur - from)  < sizeof "FALLTHRU" - 1)
2238 	return false;
2239       if (memcmp (from, "FALLTHR", sizeof "FALLTHR" - 1))
2240 	return false;
2241       from += sizeof "FALLTHR" - 1;
2242       if (*from == 'U')
2243 	from++;
2244       else if ((size_t) (pfile->buffer->cur - from)  < sizeof "OUGH" - 1)
2245 	return false;
2246       else if (memcmp (from, "OUGH", sizeof "OUGH" - 1))
2247 	return false;
2248       else
2249 	from += sizeof "OUGH" - 1;
2250       while (*from == ' ' || *from == '\t')
2251 	from++;
2252     }
2253   /* Whole comment contents (regex):
2254      [ \t.!]*(ELSE,? |INTENTIONAL(LY)? )?FALL(S | |-)?THR(OUGH|U)[ \t.!]*(-[^\n\r]*)?
2255      [ \t.!]*(Else,? |Intentional(ly)? )?Fall((s | |-)[Tt]|t)hr(ough|u)[ \t.!]*(-[^\n\r]*)?
2256      [ \t.!]*([Ee]lse,? |[Ii]ntentional(ly)? )?fall(s | |-)?thr(ough|u)[ \t.!]*(-[^\n\r]*)?
2257    */
2258   else
2259     {
2260       while (*from == ' ' || *from == '\t' || *from == '.' || *from == '!')
2261 	from++;
2262       unsigned char f = *from;
2263       bool all_upper = false;
2264       if (f == 'E' || f == 'e')
2265 	{
2266 	  if ((size_t) (pfile->buffer->cur - from)
2267 	      < sizeof "else fallthru" - 1)
2268 	    return false;
2269 	  if (f == 'E' && memcmp (from + 1, "LSE", sizeof "LSE" - 1) == 0)
2270 	    all_upper = true;
2271 	  else if (memcmp (from + 1, "lse", sizeof "lse" - 1))
2272 	    return false;
2273 	  from += sizeof "else" - 1;
2274 	  if (*from == ',')
2275 	    from++;
2276 	  if (*from != ' ')
2277 	    return false;
2278 	  from++;
2279 	  if (all_upper && *from == 'f')
2280 	    return false;
2281 	  if (f == 'e' && *from == 'F')
2282 	    return false;
2283 	  f = *from;
2284 	}
2285       else if (f == 'I' || f == 'i')
2286 	{
2287 	  if ((size_t) (pfile->buffer->cur - from)
2288 	      < sizeof "intentional fallthru" - 1)
2289 	    return false;
2290 	  if (f == 'I' && memcmp (from + 1, "NTENTIONAL",
2291 				  sizeof "NTENTIONAL" - 1) == 0)
2292 	    all_upper = true;
2293 	  else if (memcmp (from + 1, "ntentional",
2294 			   sizeof "ntentional" - 1))
2295 	    return false;
2296 	  from += sizeof "intentional" - 1;
2297 	  if (*from == ' ')
2298 	    {
2299 	      from++;
2300 	      if (all_upper && *from == 'f')
2301 		return false;
2302 	    }
2303 	  else if (all_upper)
2304 	    {
2305 	      if (memcmp (from, "LY F", sizeof "LY F" - 1))
2306 		return false;
2307 	      from += sizeof "LY " - 1;
2308 	    }
2309 	  else
2310 	    {
2311 	      if (memcmp (from, "ly ", sizeof "ly " - 1))
2312 		return false;
2313 	      from += sizeof "ly " - 1;
2314 	    }
2315 	  if (f == 'i' && *from == 'F')
2316 	    return false;
2317 	  f = *from;
2318 	}
2319       if (f != 'F' && f != 'f')
2320 	return false;
2321       if ((size_t) (pfile->buffer->cur - from) < sizeof "fallthru" - 1)
2322 	return false;
2323       if (f == 'F' && memcmp (from + 1, "ALL", sizeof "ALL" - 1) == 0)
2324 	all_upper = true;
2325       else if (all_upper)
2326 	return false;
2327       else if (memcmp (from + 1, "all", sizeof "all" - 1))
2328 	return false;
2329       from += sizeof "fall" - 1;
2330       if (*from == (all_upper ? 'S' : 's') && from[1] == ' ')
2331 	from += 2;
2332       else if (*from == ' ' || *from == '-')
2333 	from++;
2334       else if (*from != (all_upper ? 'T' : 't'))
2335 	return false;
2336       if ((f == 'f' || *from != 'T') && (all_upper || *from != 't'))
2337 	return false;
2338       if ((size_t) (pfile->buffer->cur - from) < sizeof "thru" - 1)
2339 	return false;
2340       if (memcmp (from + 1, all_upper ? "HRU" : "hru", sizeof "hru" - 1))
2341 	{
2342 	  if ((size_t) (pfile->buffer->cur - from) < sizeof "through" - 1)
2343 	    return false;
2344 	  if (memcmp (from + 1, all_upper ? "HROUGH" : "hrough",
2345 		      sizeof "hrough" - 1))
2346 	    return false;
2347 	  from += sizeof "through" - 1;
2348 	}
2349       else
2350 	from += sizeof "thru" - 1;
2351       while (*from == ' ' || *from == '\t' || *from == '.' || *from == '!')
2352 	from++;
2353       if (*from == '-')
2354 	{
2355 	  from++;
2356 	  if (*comment_start == '*')
2357 	    {
2358 	      do
2359 		{
2360 		  while (*from && *from != '*'
2361 			 && *from != '\n' && *from != '\r')
2362 		    from++;
2363 		  if (*from != '*' || from[1] == '/')
2364 		    break;
2365 		  from++;
2366 		}
2367 	      while (1);
2368 	    }
2369 	  else
2370 	    while (*from && *from != '\n' && *from != '\r')
2371 	      from++;
2372 	}
2373     }
2374   /* C block comment.  */
2375   if (*comment_start == '*')
2376     {
2377       if (*from != '*' || from[1] != '/')
2378 	return false;
2379     }
2380   /* C++ line comment.  */
2381   else if (*from != '\n')
2382     return false;
2383 
2384   return true;
2385 }
2386 
2387 /* Allocate COUNT tokens for RUN.  */
2388 void
2389 _cpp_init_tokenrun (tokenrun *run, unsigned int count)
2390 {
2391   run->base = XNEWVEC (cpp_token, count);
2392   run->limit = run->base + count;
2393   run->next = NULL;
2394 }
2395 
2396 /* Returns the next tokenrun, or creates one if there is none.  */
2397 static tokenrun *
2398 next_tokenrun (tokenrun *run)
2399 {
2400   if (run->next == NULL)
2401     {
2402       run->next = XNEW (tokenrun);
2403       run->next->prev = run;
2404       _cpp_init_tokenrun (run->next, 250);
2405     }
2406 
2407   return run->next;
2408 }
2409 
2410 /* Return the number of not yet processed token in a given
2411    context.  */
2412 int
2413 _cpp_remaining_tokens_num_in_context (cpp_context *context)
2414 {
2415   if (context->tokens_kind == TOKENS_KIND_DIRECT)
2416     return (LAST (context).token - FIRST (context).token);
2417   else if (context->tokens_kind == TOKENS_KIND_INDIRECT
2418 	   || context->tokens_kind == TOKENS_KIND_EXTENDED)
2419     return (LAST (context).ptoken - FIRST (context).ptoken);
2420   else
2421       abort ();
2422 }
2423 
2424 /* Returns the token present at index INDEX in a given context.  If
2425    INDEX is zero, the next token to be processed is returned.  */
2426 static const cpp_token*
2427 _cpp_token_from_context_at (cpp_context *context, int index)
2428 {
2429   if (context->tokens_kind == TOKENS_KIND_DIRECT)
2430     return &(FIRST (context).token[index]);
2431   else if (context->tokens_kind == TOKENS_KIND_INDIRECT
2432 	   || context->tokens_kind == TOKENS_KIND_EXTENDED)
2433     return FIRST (context).ptoken[index];
2434  else
2435    abort ();
2436 }
2437 
2438 /* Look ahead in the input stream.  */
2439 const cpp_token *
2440 cpp_peek_token (cpp_reader *pfile, int index)
2441 {
2442   cpp_context *context = pfile->context;
2443   const cpp_token *peektok;
2444   int count;
2445 
2446   /* First, scan through any pending cpp_context objects.  */
2447   while (context->prev)
2448     {
2449       ptrdiff_t sz = _cpp_remaining_tokens_num_in_context (context);
2450 
2451       if (index < (int) sz)
2452         return _cpp_token_from_context_at (context, index);
2453       index -= (int) sz;
2454       context = context->prev;
2455     }
2456 
2457   /* We will have to read some new tokens after all (and do so
2458      without invalidating preceding tokens).  */
2459   count = index;
2460   pfile->keep_tokens++;
2461 
2462   /* For peeked tokens temporarily disable line_change reporting,
2463      until the tokens are parsed for real.  */
2464   void (*line_change) (cpp_reader *, const cpp_token *, int)
2465     = pfile->cb.line_change;
2466   pfile->cb.line_change = NULL;
2467 
2468   do
2469     {
2470       peektok = _cpp_lex_token (pfile);
2471       if (peektok->type == CPP_EOF)
2472 	{
2473 	  index--;
2474 	  break;
2475 	}
2476     }
2477   while (index--);
2478 
2479   _cpp_backup_tokens_direct (pfile, count - index);
2480   pfile->keep_tokens--;
2481   pfile->cb.line_change = line_change;
2482 
2483   return peektok;
2484 }
2485 
2486 /* Allocate a single token that is invalidated at the same time as the
2487    rest of the tokens on the line.  Has its line and col set to the
2488    same as the last lexed token, so that diagnostics appear in the
2489    right place.  */
2490 cpp_token *
2491 _cpp_temp_token (cpp_reader *pfile)
2492 {
2493   cpp_token *old, *result;
2494   ptrdiff_t sz = pfile->cur_run->limit - pfile->cur_token;
2495   ptrdiff_t la = (ptrdiff_t) pfile->lookaheads;
2496 
2497   old = pfile->cur_token - 1;
2498   /* Any pre-existing lookaheads must not be clobbered.  */
2499   if (la)
2500     {
2501       if (sz <= la)
2502         {
2503           tokenrun *next = next_tokenrun (pfile->cur_run);
2504 
2505           if (sz < la)
2506             memmove (next->base + 1, next->base,
2507                      (la - sz) * sizeof (cpp_token));
2508 
2509           next->base[0] = pfile->cur_run->limit[-1];
2510         }
2511 
2512       if (sz > 1)
2513         memmove (pfile->cur_token + 1, pfile->cur_token,
2514                  MIN (la, sz - 1) * sizeof (cpp_token));
2515     }
2516 
2517   if (!sz && pfile->cur_token == pfile->cur_run->limit)
2518     {
2519       pfile->cur_run = next_tokenrun (pfile->cur_run);
2520       pfile->cur_token = pfile->cur_run->base;
2521     }
2522 
2523   result = pfile->cur_token++;
2524   result->src_loc = old->src_loc;
2525   return result;
2526 }
2527 
2528 /* Lex a token into RESULT (external interface).  Takes care of issues
2529    like directive handling, token lookahead, multiple include
2530    optimization and skipping.  */
2531 const cpp_token *
2532 _cpp_lex_token (cpp_reader *pfile)
2533 {
2534   cpp_token *result;
2535 
2536   for (;;)
2537     {
2538       if (pfile->cur_token == pfile->cur_run->limit)
2539 	{
2540 	  pfile->cur_run = next_tokenrun (pfile->cur_run);
2541 	  pfile->cur_token = pfile->cur_run->base;
2542 	}
2543       /* We assume that the current token is somewhere in the current
2544 	 run.  */
2545       if (pfile->cur_token < pfile->cur_run->base
2546 	  || pfile->cur_token >= pfile->cur_run->limit)
2547 	abort ();
2548 
2549       if (pfile->lookaheads)
2550 	{
2551 	  pfile->lookaheads--;
2552 	  result = pfile->cur_token++;
2553 	}
2554       else
2555 	result = _cpp_lex_direct (pfile);
2556 
2557       if (result->flags & BOL)
2558 	{
2559 	  /* Is this a directive.  If _cpp_handle_directive returns
2560 	     false, it is an assembler #.  */
2561 	  if (result->type == CPP_HASH
2562 	      /* 6.10.3 p 11: Directives in a list of macro arguments
2563 		 gives undefined behavior.  This implementation
2564 		 handles the directive as normal.  */
2565 	      && pfile->state.parsing_args != 1)
2566 	    {
2567 	      if (_cpp_handle_directive (pfile, result->flags & PREV_WHITE))
2568 		{
2569 		  if (pfile->directive_result.type == CPP_PADDING)
2570 		    continue;
2571 		  result = &pfile->directive_result;
2572 		}
2573 	    }
2574 	  else if (pfile->state.in_deferred_pragma)
2575 	    result = &pfile->directive_result;
2576 
2577 	  if (pfile->cb.line_change && !pfile->state.skipping)
2578 	    pfile->cb.line_change (pfile, result, pfile->state.parsing_args);
2579 	}
2580 
2581       /* We don't skip tokens in directives.  */
2582       if (pfile->state.in_directive || pfile->state.in_deferred_pragma)
2583 	break;
2584 
2585       /* Outside a directive, invalidate controlling macros.  At file
2586 	 EOF, _cpp_lex_direct takes care of popping the buffer, so we never
2587 	 get here and MI optimization works.  */
2588       pfile->mi_valid = false;
2589 
2590       if (!pfile->state.skipping || result->type == CPP_EOF)
2591 	break;
2592     }
2593 
2594   return result;
2595 }
2596 
2597 /* Returns true if a fresh line has been loaded.  */
2598 bool
2599 _cpp_get_fresh_line (cpp_reader *pfile)
2600 {
2601   int return_at_eof;
2602 
2603   /* We can't get a new line until we leave the current directive.  */
2604   if (pfile->state.in_directive)
2605     return false;
2606 
2607   for (;;)
2608     {
2609       cpp_buffer *buffer = pfile->buffer;
2610 
2611       if (!buffer->need_line)
2612 	return true;
2613 
2614       if (buffer->next_line < buffer->rlimit)
2615 	{
2616 	  _cpp_clean_line (pfile);
2617 	  return true;
2618 	}
2619 
2620       /* First, get out of parsing arguments state.  */
2621       if (pfile->state.parsing_args)
2622 	return false;
2623 
2624       /* End of buffer.  Non-empty files should end in a newline.  */
2625       if (buffer->buf != buffer->rlimit
2626 	  && buffer->next_line > buffer->rlimit
2627 	  && !buffer->from_stage3)
2628 	{
2629 	  /* Clip to buffer size.  */
2630 	  buffer->next_line = buffer->rlimit;
2631 	}
2632 
2633       return_at_eof = buffer->return_at_eof;
2634       _cpp_pop_buffer (pfile);
2635       if (pfile->buffer == NULL || return_at_eof)
2636 	return false;
2637     }
2638 }
2639 
2640 #define IF_NEXT_IS(CHAR, THEN_TYPE, ELSE_TYPE)		\
2641   do							\
2642     {							\
2643       result->type = ELSE_TYPE;				\
2644       if (*buffer->cur == CHAR)				\
2645 	buffer->cur++, result->type = THEN_TYPE;	\
2646     }							\
2647   while (0)
2648 
2649 /* Lex a token into pfile->cur_token, which is also incremented, to
2650    get diagnostics pointing to the correct location.
2651 
2652    Does not handle issues such as token lookahead, multiple-include
2653    optimization, directives, skipping etc.  This function is only
2654    suitable for use by _cpp_lex_token, and in special cases like
2655    lex_expansion_token which doesn't care for any of these issues.
2656 
2657    When meeting a newline, returns CPP_EOF if parsing a directive,
2658    otherwise returns to the start of the token buffer if permissible.
2659    Returns the location of the lexed token.  */
2660 cpp_token *
2661 _cpp_lex_direct (cpp_reader *pfile)
2662 {
2663   cppchar_t c;
2664   cpp_buffer *buffer;
2665   const unsigned char *comment_start;
2666   bool fallthrough_comment = false;
2667   cpp_token *result = pfile->cur_token++;
2668 
2669  fresh_line:
2670   result->flags = 0;
2671   buffer = pfile->buffer;
2672   if (buffer->need_line)
2673     {
2674       if (pfile->state.in_deferred_pragma)
2675 	{
2676 	  result->type = CPP_PRAGMA_EOL;
2677 	  pfile->state.in_deferred_pragma = false;
2678 	  if (!pfile->state.pragma_allow_expansion)
2679 	    pfile->state.prevent_expansion--;
2680 	  return result;
2681 	}
2682       if (!_cpp_get_fresh_line (pfile))
2683 	{
2684 	  result->type = CPP_EOF;
2685 	  if (!pfile->state.in_directive)
2686 	    {
2687 	      /* Tell the compiler the line number of the EOF token.  */
2688 	      result->src_loc = pfile->line_table->highest_line;
2689 	      result->flags = BOL;
2690 	    }
2691 	  return result;
2692 	}
2693       if (buffer != pfile->buffer)
2694 	fallthrough_comment = false;
2695       if (!pfile->keep_tokens)
2696 	{
2697 	  pfile->cur_run = &pfile->base_run;
2698 	  result = pfile->base_run.base;
2699 	  pfile->cur_token = result + 1;
2700 	}
2701       result->flags = BOL;
2702       if (pfile->state.parsing_args == 2)
2703 	result->flags |= PREV_WHITE;
2704     }
2705   buffer = pfile->buffer;
2706  update_tokens_line:
2707   result->src_loc = pfile->line_table->highest_line;
2708 
2709  skipped_white:
2710   if (buffer->cur >= buffer->notes[buffer->cur_note].pos
2711       && !pfile->overlaid_buffer)
2712     {
2713       _cpp_process_line_notes (pfile, false);
2714       result->src_loc = pfile->line_table->highest_line;
2715     }
2716   c = *buffer->cur++;
2717 
2718   if (pfile->forced_token_location_p)
2719     result->src_loc = *pfile->forced_token_location_p;
2720   else
2721     result->src_loc = linemap_position_for_column (pfile->line_table,
2722 					  CPP_BUF_COLUMN (buffer, buffer->cur));
2723 
2724   switch (c)
2725     {
2726     case ' ': case '\t': case '\f': case '\v': case '\0':
2727       result->flags |= PREV_WHITE;
2728       skip_whitespace (pfile, c);
2729       goto skipped_white;
2730 
2731     case '\n':
2732       if (buffer->cur < buffer->rlimit)
2733 	CPP_INCREMENT_LINE (pfile, 0);
2734       buffer->need_line = true;
2735       goto fresh_line;
2736 
2737     case '0': case '1': case '2': case '3': case '4':
2738     case '5': case '6': case '7': case '8': case '9':
2739       {
2740 	struct normalize_state nst = INITIAL_NORMALIZE_STATE;
2741 	result->type = CPP_NUMBER;
2742 	lex_number (pfile, &result->val.str, &nst);
2743 	warn_about_normalization (pfile, result, &nst);
2744 	break;
2745       }
2746 
2747     case 'L':
2748     case 'u':
2749     case 'U':
2750     case 'R':
2751       /* 'L', 'u', 'U', 'u8' or 'R' may introduce wide characters,
2752 	 wide strings or raw strings.  */
2753       if (c == 'L' || CPP_OPTION (pfile, rliterals)
2754 	  || (c != 'R' && CPP_OPTION (pfile, uliterals)))
2755 	{
2756 	  if ((*buffer->cur == '\'' && c != 'R')
2757 	      || *buffer->cur == '"'
2758 	      || (*buffer->cur == 'R'
2759 		  && c != 'R'
2760 		  && buffer->cur[1] == '"'
2761 		  && CPP_OPTION (pfile, rliterals))
2762 	      || (*buffer->cur == '8'
2763 		  && c == 'u'
2764 		  && ((buffer->cur[1] == '"' || (buffer->cur[1] == '\''
2765 				&& CPP_OPTION (pfile, utf8_char_literals)))
2766 		      || (buffer->cur[1] == 'R' && buffer->cur[2] == '"'
2767 			  && CPP_OPTION (pfile, rliterals)))))
2768 	    {
2769 	      lex_string (pfile, result, buffer->cur - 1);
2770 	      break;
2771 	    }
2772 	}
2773       /* Fall through.  */
2774 
2775     case '_':
2776     case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
2777     case 'g': case 'h': case 'i': case 'j': case 'k': case 'l':
2778     case 'm': case 'n': case 'o': case 'p': case 'q': case 'r':
2779     case 's': case 't':           case 'v': case 'w': case 'x':
2780     case 'y': case 'z':
2781     case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
2782     case 'G': case 'H': case 'I': case 'J': case 'K':
2783     case 'M': case 'N': case 'O': case 'P': case 'Q':
2784     case 'S': case 'T':           case 'V': case 'W': case 'X':
2785     case 'Y': case 'Z':
2786       result->type = CPP_NAME;
2787       {
2788 	struct normalize_state nst = INITIAL_NORMALIZE_STATE;
2789 	result->val.node.node = lex_identifier (pfile, buffer->cur - 1, false,
2790 						&nst,
2791 						&result->val.node.spelling);
2792 	warn_about_normalization (pfile, result, &nst);
2793       }
2794 
2795       /* Convert named operators to their proper types.  */
2796       if (result->val.node.node->flags & NODE_OPERATOR)
2797 	{
2798 	  result->flags |= NAMED_OP;
2799 	  result->type = (enum cpp_ttype) result->val.node.node->directive_index;
2800 	}
2801 
2802       /* Signal FALLTHROUGH comment followed by another token.  */
2803       if (fallthrough_comment)
2804 	result->flags |= PREV_FALLTHROUGH;
2805       break;
2806 
2807     case '\'':
2808     case '"':
2809       lex_string (pfile, result, buffer->cur - 1);
2810       break;
2811 
2812     case '/':
2813       /* A potential block or line comment.  */
2814       comment_start = buffer->cur;
2815       c = *buffer->cur;
2816 
2817       if (c == '*')
2818 	{
2819 	  if (_cpp_skip_block_comment (pfile))
2820 	    cpp_error (pfile, CPP_DL_ERROR, "unterminated comment");
2821 	}
2822       else if (c == '/' && ! CPP_OPTION (pfile, traditional))
2823 	{
2824 	  /* Don't warn for system headers.  */
2825 	  if (cpp_in_system_header (pfile))
2826 	    ;
2827 	  /* Warn about comments if pedantically GNUC89, and not
2828 	     in system headers.  */
2829 	  else if (CPP_OPTION (pfile, lang) == CLK_GNUC89
2830 		   && CPP_PEDANTIC (pfile)
2831 		   && ! buffer->warned_cplusplus_comments)
2832 	    {
2833 	      cpp_error (pfile, CPP_DL_PEDWARN,
2834 			 "C++ style comments are not allowed in ISO C90");
2835 	      cpp_error (pfile, CPP_DL_PEDWARN,
2836 			 "(this will be reported only once per input file)");
2837 	      buffer->warned_cplusplus_comments = 1;
2838 	    }
2839 	  /* Or if specifically desired via -Wc90-c99-compat.  */
2840 	  else if (CPP_OPTION (pfile, cpp_warn_c90_c99_compat) > 0
2841 		   && ! CPP_OPTION (pfile, cplusplus)
2842 		   && ! buffer->warned_cplusplus_comments)
2843 	    {
2844 	      cpp_error (pfile, CPP_DL_WARNING,
2845 			 "C++ style comments are incompatible with C90");
2846 	      cpp_error (pfile, CPP_DL_WARNING,
2847 			 "(this will be reported only once per input file)");
2848 	      buffer->warned_cplusplus_comments = 1;
2849 	    }
2850 	  /* In C89/C94, C++ style comments are forbidden.  */
2851 	  else if ((CPP_OPTION (pfile, lang) == CLK_STDC89
2852 		    || CPP_OPTION (pfile, lang) == CLK_STDC94))
2853 	    {
2854 	      /* But don't be confused about valid code such as
2855 	         - // immediately followed by *,
2856 		 - // in a preprocessing directive,
2857 		 - // in an #if 0 block.  */
2858 	      if (buffer->cur[1] == '*'
2859 		  || pfile->state.in_directive
2860 		  || pfile->state.skipping)
2861 		{
2862 		  result->type = CPP_DIV;
2863 		  break;
2864 		}
2865 	      else if (! buffer->warned_cplusplus_comments)
2866 		{
2867 		  cpp_error (pfile, CPP_DL_ERROR,
2868 			     "C++ style comments are not allowed in ISO C90");
2869 		  cpp_error (pfile, CPP_DL_ERROR,
2870 			     "(this will be reported only once per input "
2871 			     "file)");
2872 		  buffer->warned_cplusplus_comments = 1;
2873 		}
2874 	    }
2875 	  if (skip_line_comment (pfile) && CPP_OPTION (pfile, warn_comments))
2876 	    cpp_warning (pfile, CPP_W_COMMENTS, "multi-line comment");
2877 	}
2878       else if (c == '=')
2879 	{
2880 	  buffer->cur++;
2881 	  result->type = CPP_DIV_EQ;
2882 	  break;
2883 	}
2884       else
2885 	{
2886 	  result->type = CPP_DIV;
2887 	  break;
2888 	}
2889 
2890       if (fallthrough_comment_p (pfile, comment_start))
2891 	fallthrough_comment = true;
2892 
2893       if (!pfile->state.save_comments)
2894 	{
2895 	  result->flags |= PREV_WHITE;
2896 	  goto update_tokens_line;
2897 	}
2898 
2899       if (fallthrough_comment)
2900 	result->flags |= PREV_FALLTHROUGH;
2901 
2902       /* Save the comment as a token in its own right.  */
2903       save_comment (pfile, result, comment_start, c);
2904       break;
2905 
2906     case '<':
2907       if (pfile->state.angled_headers)
2908 	{
2909 	  lex_string (pfile, result, buffer->cur - 1);
2910 	  if (result->type != CPP_LESS)
2911 	    break;
2912 	}
2913 
2914       result->type = CPP_LESS;
2915       if (*buffer->cur == '=')
2916 	buffer->cur++, result->type = CPP_LESS_EQ;
2917       else if (*buffer->cur == '<')
2918 	{
2919 	  buffer->cur++;
2920 	  IF_NEXT_IS ('=', CPP_LSHIFT_EQ, CPP_LSHIFT);
2921 	}
2922       else if (CPP_OPTION (pfile, digraphs))
2923 	{
2924 	  if (*buffer->cur == ':')
2925 	    {
2926 	      /* C++11 [2.5/3 lex.pptoken], "Otherwise, if the next
2927 		 three characters are <:: and the subsequent character
2928 		 is neither : nor >, the < is treated as a preprocessor
2929 		 token by itself".  */
2930 	      if (CPP_OPTION (pfile, cplusplus)
2931 		  && CPP_OPTION (pfile, lang) != CLK_CXX98
2932 		  && CPP_OPTION (pfile, lang) != CLK_GNUCXX
2933 		  && buffer->cur[1] == ':'
2934 		  && buffer->cur[2] != ':' && buffer->cur[2] != '>')
2935 		break;
2936 
2937 	      buffer->cur++;
2938 	      result->flags |= DIGRAPH;
2939 	      result->type = CPP_OPEN_SQUARE;
2940 	    }
2941 	  else if (*buffer->cur == '%')
2942 	    {
2943 	      buffer->cur++;
2944 	      result->flags |= DIGRAPH;
2945 	      result->type = CPP_OPEN_BRACE;
2946 	    }
2947 	}
2948       break;
2949 
2950     case '>':
2951       result->type = CPP_GREATER;
2952       if (*buffer->cur == '=')
2953 	buffer->cur++, result->type = CPP_GREATER_EQ;
2954       else if (*buffer->cur == '>')
2955 	{
2956 	  buffer->cur++;
2957 	  IF_NEXT_IS ('=', CPP_RSHIFT_EQ, CPP_RSHIFT);
2958 	}
2959       break;
2960 
2961     case '%':
2962       result->type = CPP_MOD;
2963       if (*buffer->cur == '=')
2964 	buffer->cur++, result->type = CPP_MOD_EQ;
2965       else if (CPP_OPTION (pfile, digraphs))
2966 	{
2967 	  if (*buffer->cur == ':')
2968 	    {
2969 	      buffer->cur++;
2970 	      result->flags |= DIGRAPH;
2971 	      result->type = CPP_HASH;
2972 	      if (*buffer->cur == '%' && buffer->cur[1] == ':')
2973 		buffer->cur += 2, result->type = CPP_PASTE, result->val.token_no = 0;
2974 	    }
2975 	  else if (*buffer->cur == '>')
2976 	    {
2977 	      buffer->cur++;
2978 	      result->flags |= DIGRAPH;
2979 	      result->type = CPP_CLOSE_BRACE;
2980 	    }
2981 	}
2982       break;
2983 
2984     case '.':
2985       result->type = CPP_DOT;
2986       if (ISDIGIT (*buffer->cur))
2987 	{
2988 	  struct normalize_state nst = INITIAL_NORMALIZE_STATE;
2989 	  result->type = CPP_NUMBER;
2990 	  lex_number (pfile, &result->val.str, &nst);
2991 	  warn_about_normalization (pfile, result, &nst);
2992 	}
2993       else if (*buffer->cur == '.' && buffer->cur[1] == '.')
2994 	buffer->cur += 2, result->type = CPP_ELLIPSIS;
2995       else if (*buffer->cur == '*' && CPP_OPTION (pfile, cplusplus))
2996 	buffer->cur++, result->type = CPP_DOT_STAR;
2997       break;
2998 
2999     case '+':
3000       result->type = CPP_PLUS;
3001       if (*buffer->cur == '+')
3002 	buffer->cur++, result->type = CPP_PLUS_PLUS;
3003       else if (*buffer->cur == '=')
3004 	buffer->cur++, result->type = CPP_PLUS_EQ;
3005       break;
3006 
3007     case '-':
3008       result->type = CPP_MINUS;
3009       if (*buffer->cur == '>')
3010 	{
3011 	  buffer->cur++;
3012 	  result->type = CPP_DEREF;
3013 	  if (*buffer->cur == '*' && CPP_OPTION (pfile, cplusplus))
3014 	    buffer->cur++, result->type = CPP_DEREF_STAR;
3015 	}
3016       else if (*buffer->cur == '-')
3017 	buffer->cur++, result->type = CPP_MINUS_MINUS;
3018       else if (*buffer->cur == '=')
3019 	buffer->cur++, result->type = CPP_MINUS_EQ;
3020       break;
3021 
3022     case '&':
3023       result->type = CPP_AND;
3024       if (*buffer->cur == '&')
3025 	buffer->cur++, result->type = CPP_AND_AND;
3026       else if (*buffer->cur == '=')
3027 	buffer->cur++, result->type = CPP_AND_EQ;
3028       break;
3029 
3030     case '|':
3031       result->type = CPP_OR;
3032       if (*buffer->cur == '|')
3033 	buffer->cur++, result->type = CPP_OR_OR;
3034       else if (*buffer->cur == '=')
3035 	buffer->cur++, result->type = CPP_OR_EQ;
3036       break;
3037 
3038     case ':':
3039       result->type = CPP_COLON;
3040       if (*buffer->cur == ':' && CPP_OPTION (pfile, cplusplus))
3041 	buffer->cur++, result->type = CPP_SCOPE;
3042       else if (*buffer->cur == '>' && CPP_OPTION (pfile, digraphs))
3043 	{
3044 	  buffer->cur++;
3045 	  result->flags |= DIGRAPH;
3046 	  result->type = CPP_CLOSE_SQUARE;
3047 	}
3048       break;
3049 
3050     case '*': IF_NEXT_IS ('=', CPP_MULT_EQ, CPP_MULT); break;
3051     case '=': IF_NEXT_IS ('=', CPP_EQ_EQ, CPP_EQ); break;
3052     case '!': IF_NEXT_IS ('=', CPP_NOT_EQ, CPP_NOT); break;
3053     case '^': IF_NEXT_IS ('=', CPP_XOR_EQ, CPP_XOR); break;
3054     case '#': IF_NEXT_IS ('#', CPP_PASTE, CPP_HASH); result->val.token_no = 0; break;
3055 
3056     case '?': result->type = CPP_QUERY; break;
3057     case '~': result->type = CPP_COMPL; break;
3058     case ',': result->type = CPP_COMMA; break;
3059     case '(': result->type = CPP_OPEN_PAREN; break;
3060     case ')': result->type = CPP_CLOSE_PAREN; break;
3061     case '[': result->type = CPP_OPEN_SQUARE; break;
3062     case ']': result->type = CPP_CLOSE_SQUARE; break;
3063     case '{': result->type = CPP_OPEN_BRACE; break;
3064     case '}': result->type = CPP_CLOSE_BRACE; break;
3065     case ';': result->type = CPP_SEMICOLON; break;
3066 
3067       /* @ is a punctuator in Objective-C.  */
3068     case '@': result->type = CPP_ATSIGN; break;
3069 
3070     case '$':
3071     case '\\':
3072       {
3073 	const uchar *base = --buffer->cur;
3074 	struct normalize_state nst = INITIAL_NORMALIZE_STATE;
3075 
3076 	if (forms_identifier_p (pfile, true, &nst))
3077 	  {
3078 	    result->type = CPP_NAME;
3079 	    result->val.node.node = lex_identifier (pfile, base, true, &nst,
3080 						    &result->val.node.spelling);
3081 	    warn_about_normalization (pfile, result, &nst);
3082 	    break;
3083 	  }
3084 	buffer->cur++;
3085       }
3086       /* FALLTHRU */
3087 
3088     default:
3089       create_literal (pfile, result, buffer->cur - 1, 1, CPP_OTHER);
3090       break;
3091     }
3092 
3093   /* Potentially convert the location of the token to a range.  */
3094   if (result->src_loc >= RESERVED_LOCATION_COUNT
3095       && result->type != CPP_EOF)
3096     {
3097       /* Ensure that any line notes are processed, so that we have the
3098 	 correct physical line/column for the end-point of the token even
3099 	 when a logical line is split via one or more backslashes.  */
3100       if (buffer->cur >= buffer->notes[buffer->cur_note].pos
3101 	  && !pfile->overlaid_buffer)
3102 	_cpp_process_line_notes (pfile, false);
3103 
3104       source_range tok_range;
3105       tok_range.m_start = result->src_loc;
3106       tok_range.m_finish
3107 	= linemap_position_for_column (pfile->line_table,
3108 				       CPP_BUF_COLUMN (buffer, buffer->cur));
3109 
3110       result->src_loc = COMBINE_LOCATION_DATA (pfile->line_table,
3111 					       result->src_loc,
3112 					       tok_range, NULL);
3113     }
3114 
3115   return result;
3116 }
3117 
3118 /* An upper bound on the number of bytes needed to spell TOKEN.
3119    Does not include preceding whitespace.  */
3120 unsigned int
3121 cpp_token_len (const cpp_token *token)
3122 {
3123   unsigned int len;
3124 
3125   switch (TOKEN_SPELL (token))
3126     {
3127     default:		len = 6;				break;
3128     case SPELL_LITERAL:	len = token->val.str.len;		break;
3129     case SPELL_IDENT:	len = NODE_LEN (token->val.node.node) * 10;	break;
3130     }
3131 
3132   return len;
3133 }
3134 
3135 /* Parse UTF-8 out of NAMEP and place a \U escape in BUFFER.
3136    Return the number of bytes read out of NAME.  (There are always
3137    10 bytes written to BUFFER.)  */
3138 
3139 static size_t
3140 utf8_to_ucn (unsigned char *buffer, const unsigned char *name)
3141 {
3142   int j;
3143   int ucn_len = 0;
3144   int ucn_len_c;
3145   unsigned t;
3146   unsigned long utf32;
3147 
3148   /* Compute the length of the UTF-8 sequence.  */
3149   for (t = *name; t & 0x80; t <<= 1)
3150     ucn_len++;
3151 
3152   utf32 = *name & (0x7F >> ucn_len);
3153   for (ucn_len_c = 1; ucn_len_c < ucn_len; ucn_len_c++)
3154     {
3155       utf32 = (utf32 << 6) | (*++name & 0x3F);
3156 
3157       /* Ill-formed UTF-8.  */
3158       if ((*name & ~0x3F) != 0x80)
3159 	abort ();
3160     }
3161 
3162   *buffer++ = '\\';
3163   *buffer++ = 'U';
3164   for (j = 7; j >= 0; j--)
3165     *buffer++ = "0123456789abcdef"[(utf32 >> (4 * j)) & 0xF];
3166   return ucn_len;
3167 }
3168 
3169 /* Given a token TYPE corresponding to a digraph, return a pointer to
3170    the spelling of the digraph.  */
3171 static const unsigned char *
3172 cpp_digraph2name (enum cpp_ttype type)
3173 {
3174   return digraph_spellings[(int) type - (int) CPP_FIRST_DIGRAPH];
3175 }
3176 
3177 /* Write the spelling of an identifier IDENT, using UCNs, to BUFFER.
3178    The buffer must already contain the enough space to hold the
3179    token's spelling.  Returns a pointer to the character after the
3180    last character written.  */
3181 unsigned char *
3182 _cpp_spell_ident_ucns (unsigned char *buffer, cpp_hashnode *ident)
3183 {
3184   size_t i;
3185   const unsigned char *name = NODE_NAME (ident);
3186 
3187   for (i = 0; i < NODE_LEN (ident); i++)
3188     if (name[i] & ~0x7F)
3189       {
3190 	i += utf8_to_ucn (buffer, name + i) - 1;
3191 	buffer += 10;
3192       }
3193     else
3194       *buffer++ = name[i];
3195 
3196   return buffer;
3197 }
3198 
3199 /* Write the spelling of a token TOKEN to BUFFER.  The buffer must
3200    already contain the enough space to hold the token's spelling.
3201    Returns a pointer to the character after the last character written.
3202    FORSTRING is true if this is to be the spelling after translation
3203    phase 1 (with the original spelling of extended identifiers), false
3204    if extended identifiers should always be written using UCNs (there is
3205    no option for always writing them in the internal UTF-8 form).
3206    FIXME: Would be nice if we didn't need the PFILE argument.  */
3207 unsigned char *
3208 cpp_spell_token (cpp_reader *pfile, const cpp_token *token,
3209 		 unsigned char *buffer, bool forstring)
3210 {
3211   switch (TOKEN_SPELL (token))
3212     {
3213     case SPELL_OPERATOR:
3214       {
3215 	const unsigned char *spelling;
3216 	unsigned char c;
3217 
3218 	if (token->flags & DIGRAPH)
3219 	  spelling = cpp_digraph2name (token->type);
3220 	else if (token->flags & NAMED_OP)
3221 	  goto spell_ident;
3222 	else
3223 	  spelling = TOKEN_NAME (token);
3224 
3225 	while ((c = *spelling++) != '\0')
3226 	  *buffer++ = c;
3227       }
3228       break;
3229 
3230     spell_ident:
3231     case SPELL_IDENT:
3232       if (forstring)
3233 	{
3234 	  memcpy (buffer, NODE_NAME (token->val.node.spelling),
3235 		  NODE_LEN (token->val.node.spelling));
3236 	  buffer += NODE_LEN (token->val.node.spelling);
3237 	}
3238       else
3239 	buffer = _cpp_spell_ident_ucns (buffer, token->val.node.node);
3240       break;
3241 
3242     case SPELL_LITERAL:
3243       memcpy (buffer, token->val.str.text, token->val.str.len);
3244       buffer += token->val.str.len;
3245       break;
3246 
3247     case SPELL_NONE:
3248       cpp_error (pfile, CPP_DL_ICE,
3249 		 "unspellable token %s", TOKEN_NAME (token));
3250       break;
3251     }
3252 
3253   return buffer;
3254 }
3255 
3256 /* Returns TOKEN spelt as a null-terminated string.  The string is
3257    freed when the reader is destroyed.  Useful for diagnostics.  */
3258 unsigned char *
3259 cpp_token_as_text (cpp_reader *pfile, const cpp_token *token)
3260 {
3261   unsigned int len = cpp_token_len (token) + 1;
3262   unsigned char *start = _cpp_unaligned_alloc (pfile, len), *end;
3263 
3264   end = cpp_spell_token (pfile, token, start, false);
3265   end[0] = '\0';
3266 
3267   return start;
3268 }
3269 
3270 /* Returns a pointer to a string which spells the token defined by
3271    TYPE and FLAGS.  Used by C front ends, which really should move to
3272    using cpp_token_as_text.  */
3273 const char *
3274 cpp_type2name (enum cpp_ttype type, unsigned char flags)
3275 {
3276   if (flags & DIGRAPH)
3277     return (const char *) cpp_digraph2name (type);
3278   else if (flags & NAMED_OP)
3279     return cpp_named_operator2name (type);
3280 
3281   return (const char *) token_spellings[type].name;
3282 }
3283 
3284 /* Writes the spelling of token to FP, without any preceding space.
3285    Separated from cpp_spell_token for efficiency - to avoid stdio
3286    double-buffering.  */
3287 void
3288 cpp_output_token (const cpp_token *token, FILE *fp)
3289 {
3290   switch (TOKEN_SPELL (token))
3291     {
3292     case SPELL_OPERATOR:
3293       {
3294 	const unsigned char *spelling;
3295 	int c;
3296 
3297 	if (token->flags & DIGRAPH)
3298 	  spelling = cpp_digraph2name (token->type);
3299 	else if (token->flags & NAMED_OP)
3300 	  goto spell_ident;
3301 	else
3302 	  spelling = TOKEN_NAME (token);
3303 
3304 	c = *spelling;
3305 	do
3306 	  putc (c, fp);
3307 	while ((c = *++spelling) != '\0');
3308       }
3309       break;
3310 
3311     spell_ident:
3312     case SPELL_IDENT:
3313       {
3314 	size_t i;
3315 	const unsigned char * name = NODE_NAME (token->val.node.node);
3316 
3317 	for (i = 0; i < NODE_LEN (token->val.node.node); i++)
3318 	  if (name[i] & ~0x7F)
3319 	    {
3320 	      unsigned char buffer[10];
3321 	      i += utf8_to_ucn (buffer, name + i) - 1;
3322 	      fwrite (buffer, 1, 10, fp);
3323 	    }
3324 	  else
3325 	    fputc (NODE_NAME (token->val.node.node)[i], fp);
3326       }
3327       break;
3328 
3329     case SPELL_LITERAL:
3330       fwrite (token->val.str.text, 1, token->val.str.len, fp);
3331       break;
3332 
3333     case SPELL_NONE:
3334       /* An error, most probably.  */
3335       break;
3336     }
3337 }
3338 
3339 /* Compare two tokens.  */
3340 int
3341 _cpp_equiv_tokens (const cpp_token *a, const cpp_token *b)
3342 {
3343   if (a->type == b->type && a->flags == b->flags)
3344     switch (TOKEN_SPELL (a))
3345       {
3346       default:			/* Keep compiler happy.  */
3347       case SPELL_OPERATOR:
3348 	/* token_no is used to track where multiple consecutive ##
3349 	   tokens were originally located.  */
3350 	return (a->type != CPP_PASTE || a->val.token_no == b->val.token_no);
3351       case SPELL_NONE:
3352 	return (a->type != CPP_MACRO_ARG
3353 		|| (a->val.macro_arg.arg_no == b->val.macro_arg.arg_no
3354 		    && a->val.macro_arg.spelling == b->val.macro_arg.spelling));
3355       case SPELL_IDENT:
3356 	return (a->val.node.node == b->val.node.node
3357 		&& a->val.node.spelling == b->val.node.spelling);
3358       case SPELL_LITERAL:
3359 	return (a->val.str.len == b->val.str.len
3360 		&& !memcmp (a->val.str.text, b->val.str.text,
3361 			    a->val.str.len));
3362       }
3363 
3364   return 0;
3365 }
3366 
3367 /* Returns nonzero if a space should be inserted to avoid an
3368    accidental token paste for output.  For simplicity, it is
3369    conservative, and occasionally advises a space where one is not
3370    needed, e.g. "." and ".2".  */
3371 int
3372 cpp_avoid_paste (cpp_reader *pfile, const cpp_token *token1,
3373 		 const cpp_token *token2)
3374 {
3375   enum cpp_ttype a = token1->type, b = token2->type;
3376   cppchar_t c;
3377 
3378   if (token1->flags & NAMED_OP)
3379     a = CPP_NAME;
3380   if (token2->flags & NAMED_OP)
3381     b = CPP_NAME;
3382 
3383   c = EOF;
3384   if (token2->flags & DIGRAPH)
3385     c = digraph_spellings[(int) b - (int) CPP_FIRST_DIGRAPH][0];
3386   else if (token_spellings[b].category == SPELL_OPERATOR)
3387     c = token_spellings[b].name[0];
3388 
3389   /* Quickly get everything that can paste with an '='.  */
3390   if ((int) a <= (int) CPP_LAST_EQ && c == '=')
3391     return 1;
3392 
3393   switch (a)
3394     {
3395     case CPP_GREATER:	return c == '>';
3396     case CPP_LESS:	return c == '<' || c == '%' || c == ':';
3397     case CPP_PLUS:	return c == '+';
3398     case CPP_MINUS:	return c == '-' || c == '>';
3399     case CPP_DIV:	return c == '/' || c == '*'; /* Comments.  */
3400     case CPP_MOD:	return c == ':' || c == '>';
3401     case CPP_AND:	return c == '&';
3402     case CPP_OR:	return c == '|';
3403     case CPP_COLON:	return c == ':' || c == '>';
3404     case CPP_DEREF:	return c == '*';
3405     case CPP_DOT:	return c == '.' || c == '%' || b == CPP_NUMBER;
3406     case CPP_HASH:	return c == '#' || c == '%'; /* Digraph form.  */
3407     case CPP_NAME:	return ((b == CPP_NUMBER
3408 				 && name_p (pfile, &token2->val.str))
3409 				|| b == CPP_NAME
3410 				|| b == CPP_CHAR || b == CPP_STRING); /* L */
3411     case CPP_NUMBER:	return (b == CPP_NUMBER || b == CPP_NAME
3412 				|| c == '.' || c == '+' || c == '-');
3413 				      /* UCNs */
3414     case CPP_OTHER:	return ((token1->val.str.text[0] == '\\'
3415 				 && b == CPP_NAME)
3416 				|| (CPP_OPTION (pfile, objc)
3417 				    && token1->val.str.text[0] == '@'
3418 				    && (b == CPP_NAME || b == CPP_STRING)));
3419     case CPP_STRING:
3420     case CPP_WSTRING:
3421     case CPP_UTF8STRING:
3422     case CPP_STRING16:
3423     case CPP_STRING32:	return (CPP_OPTION (pfile, user_literals)
3424 				&& (b == CPP_NAME
3425 				    || (TOKEN_SPELL (token2) == SPELL_LITERAL
3426 					&& ISIDST (token2->val.str.text[0]))));
3427 
3428     default:		break;
3429     }
3430 
3431   return 0;
3432 }
3433 
3434 /* Output all the remaining tokens on the current line, and a newline
3435    character, to FP.  Leading whitespace is removed.  If there are
3436    macros, special token padding is not performed.  */
3437 void
3438 cpp_output_line (cpp_reader *pfile, FILE *fp)
3439 {
3440   const cpp_token *token;
3441 
3442   token = cpp_get_token (pfile);
3443   while (token->type != CPP_EOF)
3444     {
3445       cpp_output_token (token, fp);
3446       token = cpp_get_token (pfile);
3447       if (token->flags & PREV_WHITE)
3448 	putc (' ', fp);
3449     }
3450 
3451   putc ('\n', fp);
3452 }
3453 
3454 /* Return a string representation of all the remaining tokens on the
3455    current line.  The result is allocated using xmalloc and must be
3456    freed by the caller.  */
3457 unsigned char *
3458 cpp_output_line_to_string (cpp_reader *pfile, const unsigned char *dir_name)
3459 {
3460   const cpp_token *token;
3461   unsigned int out = dir_name ? ustrlen (dir_name) : 0;
3462   unsigned int alloced = 120 + out;
3463   unsigned char *result = (unsigned char *) xmalloc (alloced);
3464 
3465   /* If DIR_NAME is empty, there are no initial contents.  */
3466   if (dir_name)
3467     {
3468       sprintf ((char *) result, "#%s ", dir_name);
3469       out += 2;
3470     }
3471 
3472   token = cpp_get_token (pfile);
3473   while (token->type != CPP_EOF)
3474     {
3475       unsigned char *last;
3476       /* Include room for a possible space and the terminating nul.  */
3477       unsigned int len = cpp_token_len (token) + 2;
3478 
3479       if (out + len > alloced)
3480 	{
3481 	  alloced *= 2;
3482 	  if (out + len > alloced)
3483 	    alloced = out + len;
3484 	  result = (unsigned char *) xrealloc (result, alloced);
3485 	}
3486 
3487       last = cpp_spell_token (pfile, token, &result[out], 0);
3488       out = last - result;
3489 
3490       token = cpp_get_token (pfile);
3491       if (token->flags & PREV_WHITE)
3492 	result[out++] = ' ';
3493     }
3494 
3495   result[out] = '\0';
3496   return result;
3497 }
3498 
3499 /* Memory buffers.  Changing these three constants can have a dramatic
3500    effect on performance.  The values here are reasonable defaults,
3501    but might be tuned.  If you adjust them, be sure to test across a
3502    range of uses of cpplib, including heavy nested function-like macro
3503    expansion.  Also check the change in peak memory usage (NJAMD is a
3504    good tool for this).  */
3505 #define MIN_BUFF_SIZE 8000
3506 #define BUFF_SIZE_UPPER_BOUND(MIN_SIZE) (MIN_BUFF_SIZE + (MIN_SIZE) * 3 / 2)
3507 #define EXTENDED_BUFF_SIZE(BUFF, MIN_EXTRA) \
3508 	(MIN_EXTRA + ((BUFF)->limit - (BUFF)->cur) * 2)
3509 
3510 #if MIN_BUFF_SIZE > BUFF_SIZE_UPPER_BOUND (0)
3511   #error BUFF_SIZE_UPPER_BOUND must be at least as large as MIN_BUFF_SIZE!
3512 #endif
3513 
3514 /* Create a new allocation buffer.  Place the control block at the end
3515    of the buffer, so that buffer overflows will cause immediate chaos.  */
3516 static _cpp_buff *
3517 new_buff (size_t len)
3518 {
3519   _cpp_buff *result;
3520   unsigned char *base;
3521 
3522   if (len < MIN_BUFF_SIZE)
3523     len = MIN_BUFF_SIZE;
3524   len = CPP_ALIGN (len);
3525 
3526 #ifdef ENABLE_VALGRIND_ANNOTATIONS
3527   /* Valgrind warns about uses of interior pointers, so put _cpp_buff
3528      struct first.  */
3529   size_t slen = CPP_ALIGN2 (sizeof (_cpp_buff), 2 * DEFAULT_ALIGNMENT);
3530   base = XNEWVEC (unsigned char, len + slen);
3531   result = (_cpp_buff *) base;
3532   base += slen;
3533 #else
3534   base = XNEWVEC (unsigned char, len + sizeof (_cpp_buff));
3535   result = (_cpp_buff *) (base + len);
3536 #endif
3537   result->base = base;
3538   result->cur = base;
3539   result->limit = base + len;
3540   result->next = NULL;
3541   return result;
3542 }
3543 
3544 /* Place a chain of unwanted allocation buffers on the free list.  */
3545 void
3546 _cpp_release_buff (cpp_reader *pfile, _cpp_buff *buff)
3547 {
3548   _cpp_buff *end = buff;
3549 
3550   while (end->next)
3551     end = end->next;
3552   end->next = pfile->free_buffs;
3553   pfile->free_buffs = buff;
3554 }
3555 
3556 /* Return a free buffer of size at least MIN_SIZE.  */
3557 _cpp_buff *
3558 _cpp_get_buff (cpp_reader *pfile, size_t min_size)
3559 {
3560   _cpp_buff *result, **p;
3561 
3562   for (p = &pfile->free_buffs;; p = &(*p)->next)
3563     {
3564       size_t size;
3565 
3566       if (*p == NULL)
3567 	return new_buff (min_size);
3568       result = *p;
3569       size = result->limit - result->base;
3570       /* Return a buffer that's big enough, but don't waste one that's
3571          way too big.  */
3572       if (size >= min_size && size <= BUFF_SIZE_UPPER_BOUND (min_size))
3573 	break;
3574     }
3575 
3576   *p = result->next;
3577   result->next = NULL;
3578   result->cur = result->base;
3579   return result;
3580 }
3581 
3582 /* Creates a new buffer with enough space to hold the uncommitted
3583    remaining bytes of BUFF, and at least MIN_EXTRA more bytes.  Copies
3584    the excess bytes to the new buffer.  Chains the new buffer after
3585    BUFF, and returns the new buffer.  */
3586 _cpp_buff *
3587 _cpp_append_extend_buff (cpp_reader *pfile, _cpp_buff *buff, size_t min_extra)
3588 {
3589   size_t size = EXTENDED_BUFF_SIZE (buff, min_extra);
3590   _cpp_buff *new_buff = _cpp_get_buff (pfile, size);
3591 
3592   buff->next = new_buff;
3593   memcpy (new_buff->base, buff->cur, BUFF_ROOM (buff));
3594   return new_buff;
3595 }
3596 
3597 /* Creates a new buffer with enough space to hold the uncommitted
3598    remaining bytes of the buffer pointed to by BUFF, and at least
3599    MIN_EXTRA more bytes.  Copies the excess bytes to the new buffer.
3600    Chains the new buffer before the buffer pointed to by BUFF, and
3601    updates the pointer to point to the new buffer.  */
3602 void
3603 _cpp_extend_buff (cpp_reader *pfile, _cpp_buff **pbuff, size_t min_extra)
3604 {
3605   _cpp_buff *new_buff, *old_buff = *pbuff;
3606   size_t size = EXTENDED_BUFF_SIZE (old_buff, min_extra);
3607 
3608   new_buff = _cpp_get_buff (pfile, size);
3609   memcpy (new_buff->base, old_buff->cur, BUFF_ROOM (old_buff));
3610   new_buff->next = old_buff;
3611   *pbuff = new_buff;
3612 }
3613 
3614 /* Free a chain of buffers starting at BUFF.  */
3615 void
3616 _cpp_free_buff (_cpp_buff *buff)
3617 {
3618   _cpp_buff *next;
3619 
3620   for (; buff; buff = next)
3621     {
3622       next = buff->next;
3623 #ifdef ENABLE_VALGRIND_ANNOTATIONS
3624       free (buff);
3625 #else
3626       free (buff->base);
3627 #endif
3628     }
3629 }
3630 
3631 /* Allocate permanent, unaligned storage of length LEN.  */
3632 unsigned char *
3633 _cpp_unaligned_alloc (cpp_reader *pfile, size_t len)
3634 {
3635   _cpp_buff *buff = pfile->u_buff;
3636   unsigned char *result = buff->cur;
3637 
3638   if (len > (size_t) (buff->limit - result))
3639     {
3640       buff = _cpp_get_buff (pfile, len);
3641       buff->next = pfile->u_buff;
3642       pfile->u_buff = buff;
3643       result = buff->cur;
3644     }
3645 
3646   buff->cur = result + len;
3647   return result;
3648 }
3649 
3650 /* Allocate permanent, unaligned storage of length LEN from a_buff.
3651    That buffer is used for growing allocations when saving macro
3652    replacement lists in a #define, and when parsing an answer to an
3653    assertion in #assert, #unassert or #if (and therefore possibly
3654    whilst expanding macros).  It therefore must not be used by any
3655    code that they might call: specifically the lexer and the guts of
3656    the macro expander.
3657 
3658    All existing other uses clearly fit this restriction: storing
3659    registered pragmas during initialization.  */
3660 unsigned char *
3661 _cpp_aligned_alloc (cpp_reader *pfile, size_t len)
3662 {
3663   _cpp_buff *buff = pfile->a_buff;
3664   unsigned char *result = buff->cur;
3665 
3666   if (len > (size_t) (buff->limit - result))
3667     {
3668       buff = _cpp_get_buff (pfile, len);
3669       buff->next = pfile->a_buff;
3670       pfile->a_buff = buff;
3671       result = buff->cur;
3672     }
3673 
3674   buff->cur = result + len;
3675   return result;
3676 }
3677 
3678 /* Say which field of TOK is in use.  */
3679 
3680 enum cpp_token_fld_kind
3681 cpp_token_val_index (const cpp_token *tok)
3682 {
3683   switch (TOKEN_SPELL (tok))
3684     {
3685     case SPELL_IDENT:
3686       return CPP_TOKEN_FLD_NODE;
3687     case SPELL_LITERAL:
3688       return CPP_TOKEN_FLD_STR;
3689     case SPELL_OPERATOR:
3690       if (tok->type == CPP_PASTE)
3691 	return CPP_TOKEN_FLD_TOKEN_NO;
3692       else
3693 	return CPP_TOKEN_FLD_NONE;
3694     case SPELL_NONE:
3695       if (tok->type == CPP_MACRO_ARG)
3696 	return CPP_TOKEN_FLD_ARG_NO;
3697       else if (tok->type == CPP_PADDING)
3698 	return CPP_TOKEN_FLD_SOURCE;
3699       else if (tok->type == CPP_PRAGMA)
3700 	return CPP_TOKEN_FLD_PRAGMA;
3701       /* fall through */
3702     default:
3703       return CPP_TOKEN_FLD_NONE;
3704     }
3705 }
3706 
3707 /* All tokens lexed in R after calling this function will be forced to have
3708    their source_location the same as the location referenced by P, until
3709    cpp_stop_forcing_token_locations is called for R.  */
3710 
3711 void
3712 cpp_force_token_locations (cpp_reader *r, source_location *p)
3713 {
3714   r->forced_token_location_p = p;
3715 }
3716 
3717 /* Go back to assigning locations naturally for lexed tokens.  */
3718 
3719 void
3720 cpp_stop_forcing_token_locations (cpp_reader *r)
3721 {
3722   r->forced_token_location_p = NULL;
3723 }
3724