]>
Commit | Line | Data |
---|---|---|
2b066ec1 | 1 | // elaboration functions |
aa30ccd3 | 2 | // Copyright (C) 2005-2007 Red Hat Inc. |
69c68955 FCE |
3 | // |
4 | // This file is part of systemtap, and is free software. You can | |
5 | // redistribute it and/or modify it under the terms of the GNU General | |
6 | // Public License (GPL); either version 2, or (at your option) any | |
7 | // later version. | |
2b066ec1 FCE |
8 | |
9 | #include "config.h" | |
10 | #include "elaborate.h" | |
11 | #include "parse.h" | |
20c6c071 | 12 | #include "tapsets.h" |
dc38c0ae | 13 | #include "session.h" |
72dbc915 | 14 | #include "util.h" |
2b066ec1 FCE |
15 | |
16 | extern "C" { | |
17 | #include <sys/utsname.h> | |
a477f3f1 | 18 | #include <fnmatch.h> |
2b066ec1 FCE |
19 | } |
20 | ||
2b066ec1 | 21 | #include <algorithm> |
67c0a579 GH |
22 | #include <fstream> |
23 | #include <map> | |
29e64872 | 24 | #include <cassert> |
67c0a579 GH |
25 | #include <set> |
26 | #include <vector> | |
1b07c728 FCE |
27 | #include <algorithm> |
28 | #include <iterator> | |
29 | ||
2b066ec1 | 30 | |
2b066ec1 FCE |
31 | using namespace std; |
32 | ||
33 | ||
34 | // ------------------------------------------------------------------------ | |
35 | ||
36 | ||
37 | derived_probe::derived_probe (probe *p): | |
38 | base (p) | |
39 | { | |
77de5e9e GH |
40 | if (p) |
41 | { | |
42 | this->locations = p->locations; | |
43 | this->tok = p->tok; | |
37ebca01 | 44 | this->privileged = p->privileged; |
77de5e9e GH |
45 | this->body = deep_copy_visitor::deep_copy(p->body); |
46 | } | |
2b066ec1 FCE |
47 | } |
48 | ||
49 | ||
50 | derived_probe::derived_probe (probe *p, probe_point *l): | |
51 | base (p) | |
52 | { | |
30a279be FCE |
53 | if (l) |
54 | this->locations.push_back (l); | |
35d4ab18 | 55 | |
77de5e9e GH |
56 | if (p) |
57 | { | |
58 | this->tok = p->tok; | |
37ebca01 | 59 | this->privileged = p->privileged; |
77de5e9e GH |
60 | this->body = deep_copy_visitor::deep_copy(p->body); |
61 | } | |
2b066ec1 FCE |
62 | } |
63 | ||
98afd80e | 64 | |
b8da0ad1 FCE |
65 | void |
66 | derived_probe::printsig (ostream& o) const | |
67 | { | |
68 | probe::printsig (o); | |
69 | printsig_nested (o); | |
70 | } | |
71 | ||
72 | void | |
73 | derived_probe::printsig_nested (ostream& o) const | |
74 | { | |
75 | // We'd like to enclose the probe derivation chain in a /* */ | |
76 | // comment delimiter. But just printing /* base->printsig() */ is | |
77 | // not enough, since base might itself be a derived_probe. So we, | |
78 | // er, "cleverly" encode our nesting state as a formatting flag for | |
79 | // the ostream. | |
80 | ios::fmtflags f = o.flags (ios::internal); | |
81 | if (f & ios::internal) | |
82 | { | |
83 | // already nested | |
84 | o << " <- "; | |
85 | base->printsig (o); | |
86 | } | |
87 | else | |
88 | { | |
89 | // outermost nesting | |
90 | o << " /* <- "; | |
91 | base->printsig (o); | |
92 | o << " */"; | |
93 | } | |
94 | // restore flags | |
95 | (void) o.flags (f); | |
96 | } | |
97 | ||
98 | ||
b20febf3 | 99 | probe_point* |
b8da0ad1 | 100 | derived_probe::sole_location () const |
dc38c0ae | 101 | { |
b20febf3 FCE |
102 | if (locations.size() == 0) |
103 | throw semantic_error ("derived_probe with no locations", this->tok); | |
104 | else if (locations.size() > 1) | |
105 | throw semantic_error ("derived_probe with too many locations", this->tok); | |
106 | else | |
107 | return locations[0]; | |
dc38c0ae DS |
108 | } |
109 | ||
110 | ||
47dd066d | 111 | |
20c6c071 | 112 | // ------------------------------------------------------------------------ |
98afd80e FCE |
113 | // Members of derived_probe_builder |
114 | ||
115 | bool | |
116 | derived_probe_builder::get_param (std::map<std::string, literal*> const & params, | |
117 | const std::string& key, | |
118 | std::string& value) | |
119 | { | |
120 | map<string, literal *>::const_iterator i = params.find (key); | |
121 | if (i == params.end()) | |
122 | return false; | |
123 | literal_string * ls = dynamic_cast<literal_string *>(i->second); | |
124 | if (!ls) | |
125 | return false; | |
126 | value = ls->value; | |
127 | return true; | |
128 | } | |
129 | ||
20c6c071 | 130 | |
98afd80e FCE |
131 | bool |
132 | derived_probe_builder::get_param (std::map<std::string, literal*> const & params, | |
133 | const std::string& key, | |
134 | int64_t& value) | |
135 | { | |
136 | map<string, literal *>::const_iterator i = params.find (key); | |
137 | if (i == params.end()) | |
138 | return false; | |
139 | if (i->second == NULL) | |
140 | return false; | |
141 | literal_number * ln = dynamic_cast<literal_number *>(i->second); | |
142 | if (!ln) | |
143 | return false; | |
144 | value = ln->value; | |
145 | return true; | |
146 | } | |
147 | ||
148 | ||
149 | ||
150 | // ------------------------------------------------------------------------ | |
20c6c071 GH |
151 | // Members of match_key. |
152 | ||
153 | match_key::match_key(string const & n) | |
154 | : name(n), | |
155 | have_parameter(false), | |
156 | parameter_type(tok_junk) | |
157 | { | |
158 | } | |
159 | ||
160 | match_key::match_key(probe_point::component const & c) | |
161 | : name(c.functor), | |
162 | have_parameter(c.arg != NULL), | |
163 | parameter_type(c.arg ? c.arg->tok->type : tok_junk) | |
164 | { | |
165 | } | |
166 | ||
167 | match_key & | |
168 | match_key::with_number() | |
169 | { | |
170 | have_parameter = true; | |
171 | parameter_type = tok_number; | |
172 | return *this; | |
173 | } | |
174 | ||
175 | match_key & | |
176 | match_key::with_string() | |
177 | { | |
178 | have_parameter = true; | |
179 | parameter_type = tok_string; | |
180 | return *this; | |
181 | } | |
182 | ||
183 | string | |
184 | match_key::str() const | |
185 | { | |
186 | if (have_parameter) | |
187 | switch (parameter_type) | |
188 | { | |
189 | case tok_string: return name + "(string)"; | |
190 | case tok_number: return name + "(number)"; | |
191 | default: return name + "(...)"; | |
192 | } | |
193 | return name; | |
194 | } | |
195 | ||
196 | bool | |
197 | match_key::operator<(match_key const & other) const | |
198 | { | |
199 | return ((name < other.name) | |
200 | ||
b4ceace2 | 201 | || (name == other.name |
20c6c071 GH |
202 | && have_parameter < other.have_parameter) |
203 | ||
b4ceace2 | 204 | || (name == other.name |
20c6c071 GH |
205 | && have_parameter == other.have_parameter |
206 | && parameter_type < other.parameter_type)); | |
207 | } | |
2b066ec1 | 208 | |
a477f3f1 DS |
209 | static bool |
210 | isglob(string const & str) | |
211 | { | |
212 | return(str.find('*') != str.npos); | |
213 | } | |
214 | ||
215 | bool | |
216 | match_key::globmatch(match_key const & other) const | |
217 | { | |
218 | const char *other_str = other.name.c_str(); | |
219 | const char *name_str = name.c_str(); | |
220 | ||
221 | return ((fnmatch(name_str, other_str, FNM_NOESCAPE) == 0) | |
222 | && have_parameter == other.have_parameter | |
223 | && parameter_type == other.parameter_type); | |
224 | } | |
225 | ||
2b066ec1 | 226 | // ------------------------------------------------------------------------ |
20c6c071 GH |
227 | // Members of match_node |
228 | // ------------------------------------------------------------------------ | |
229 | ||
230 | match_node::match_node() | |
231 | : end(NULL) | |
232 | {} | |
233 | ||
234 | match_node * | |
235 | match_node::bind(match_key const & k) | |
236 | { | |
b4ceace2 FCE |
237 | if (k.name == "*") |
238 | throw semantic_error("invalid use of wildcard probe point component"); | |
239 | ||
20c6c071 GH |
240 | map<match_key, match_node *>::const_iterator i = sub.find(k); |
241 | if (i != sub.end()) | |
242 | return i->second; | |
243 | match_node * n = new match_node(); | |
244 | sub.insert(make_pair(k, n)); | |
245 | return n; | |
246 | } | |
247 | ||
248 | void | |
249 | match_node::bind(derived_probe_builder * e) | |
250 | { | |
251 | if (end) | |
59bafbe8 | 252 | throw semantic_error("duplicate probe point pattern"); |
20c6c071 GH |
253 | end = e; |
254 | } | |
255 | ||
256 | match_node * | |
257 | match_node::bind(string const & k) | |
258 | { | |
259 | return bind(match_key(k)); | |
260 | } | |
261 | ||
262 | match_node * | |
263 | match_node::bind_str(string const & k) | |
264 | { | |
265 | return bind(match_key(k).with_string()); | |
266 | } | |
267 | ||
268 | match_node * | |
269 | match_node::bind_num(string const & k) | |
270 | { | |
271 | return bind(match_key(k).with_number()); | |
272 | } | |
273 | ||
b4ceace2 FCE |
274 | |
275 | void | |
276 | match_node::find_and_build (systemtap_session& s, | |
277 | probe* p, probe_point *loc, unsigned pos, | |
278 | vector<derived_probe *>& results) | |
20c6c071 | 279 | { |
b4ceace2 FCE |
280 | assert (pos <= loc->components.size()); |
281 | if (pos == loc->components.size()) // matched all probe point components so far | |
20c6c071 | 282 | { |
b4ceace2 FCE |
283 | derived_probe_builder *b = end; // may be 0 if only nested names are bound |
284 | ||
285 | if (! b) | |
286 | { | |
287 | string alternatives; | |
288 | for (sub_map_iterator_t i = sub.begin(); i != sub.end(); i++) | |
289 | alternatives += string(" ") + i->first.str(); | |
290 | ||
291 | throw semantic_error (string("probe point truncated at position ") + | |
292 | lex_cast<string> (pos) + | |
293 | " (follow:" + alternatives + ")"); | |
294 | } | |
295 | ||
296 | map<string, literal *> param_map; | |
297 | for (unsigned i=0; i<pos; i++) | |
298 | param_map[loc->components[i]->functor] = loc->components[i]->arg; | |
299 | // maybe 0 | |
300 | ||
301 | b->build (s, p, loc, param_map, results); | |
20c6c071 | 302 | } |
a477f3f1 | 303 | else if (isglob(loc->components[pos]->functor)) // wildcard? |
20c6c071 | 304 | { |
a477f3f1 DS |
305 | match_key match (* loc->components[pos]); |
306 | ||
828c3ed5 DS |
307 | // Call find_and_build for each possible match. Ignore errors - |
308 | // unless we don't find any match. | |
309 | unsigned int num_results = results.size(); | |
b4ceace2 FCE |
310 | for (sub_map_iterator_t i = sub.begin(); i != sub.end(); i++) |
311 | { | |
a477f3f1 | 312 | const match_key& subkey = i->first; |
828c3ed5 | 313 | match_node* subnode = i->second; |
a477f3f1 DS |
314 | |
315 | if (match.globmatch(subkey)) | |
828c3ed5 | 316 | { |
a477f3f1 DS |
317 | // recurse |
318 | try | |
319 | { | |
320 | subnode->find_and_build (s, p, loc, pos+1, results); | |
321 | } | |
322 | catch (const semantic_error& e) | |
323 | { | |
324 | // Ignore semantic_errors while expanding wildcards. | |
325 | // If we get done and nothing was expanded, the code | |
326 | // following the loop will complain. | |
327 | } | |
828c3ed5 | 328 | } |
a477f3f1 | 329 | } |
cedd10f4 | 330 | if (! loc->optional && num_results == results.size()) |
828c3ed5 DS |
331 | { |
332 | // We didn't find any wildcard matches (since the size of | |
333 | // the result vector didn't change). Throw an error. | |
334 | string alternatives; | |
335 | for (sub_map_iterator_t i = sub.begin(); i != sub.end(); i++) | |
336 | alternatives += string(" ") + i->first.str(); | |
337 | ||
338 | throw semantic_error(string("probe point mismatch at position ") + | |
339 | lex_cast<string> (pos) + | |
340 | " (alternatives:" + alternatives + ")"); | |
341 | } | |
b4ceace2 FCE |
342 | } |
343 | else | |
20c6c071 | 344 | { |
b4ceace2 FCE |
345 | match_key match (* loc->components[pos]); |
346 | sub_map_iterator_t i = sub.find (match); | |
347 | if (i == sub.end()) // no match | |
348 | { | |
349 | string alternatives; | |
350 | for (sub_map_iterator_t i = sub.begin(); i != sub.end(); i++) | |
351 | alternatives += string(" ") + i->first.str(); | |
352 | ||
353 | throw semantic_error (string("probe point mismatch at position ") + | |
354 | lex_cast<string> (pos) + | |
355 | " (alternatives:" + alternatives + ")"); | |
356 | } | |
357 | ||
358 | match_node* subnode = i->second; | |
359 | // recurse | |
360 | subnode->find_and_build (s, p, loc, pos+1, results); | |
20c6c071 GH |
361 | } |
362 | } | |
363 | ||
b4ceace2 | 364 | |
aa30ccd3 FCE |
365 | void |
366 | match_node::build_no_more (systemtap_session& s) | |
367 | { | |
368 | for (sub_map_iterator_t i = sub.begin(); i != sub.end(); i++) | |
369 | i->second->build_no_more (s); | |
370 | if (end) end->build_no_more (s); | |
371 | } | |
372 | ||
373 | ||
20c6c071 GH |
374 | // ------------------------------------------------------------------------ |
375 | // Alias probes | |
376 | // ------------------------------------------------------------------------ | |
377 | ||
c1d5f3f6 FCE |
378 | struct alias_derived_probe: public derived_probe |
379 | { | |
380 | alias_derived_probe (probe* base): derived_probe (base) {} | |
381 | ||
b20febf3 FCE |
382 | void upchuck () { throw semantic_error ("inappropriate", this->tok); } |
383 | ||
384 | // Alias probes are immediately expanded to other derived_probe | |
385 | // types, and are not themselves emitted or listed in | |
386 | // systemtap_session.probes | |
dc38c0ae | 387 | |
78f6bba6 | 388 | void join_group (systemtap_session&) { upchuck (); } |
c1d5f3f6 FCE |
389 | }; |
390 | ||
391 | ||
20c6c071 GH |
392 | struct |
393 | alias_expansion_builder | |
394 | : public derived_probe_builder | |
395 | { | |
396 | probe_alias * alias; | |
397 | ||
398 | alias_expansion_builder(probe_alias * a) | |
399 | : alias(a) | |
400 | {} | |
401 | ||
5227f1ea GH |
402 | virtual void build(systemtap_session & sess, |
403 | probe * use, | |
20c6c071 | 404 | probe_point * location, |
78f6bba6 | 405 | std::map<std::string, literal *> const &, |
20c6c071 GH |
406 | vector<derived_probe *> & finished_results) |
407 | { | |
408 | // We're going to build a new probe and wrap it up in an | |
c8959a29 | 409 | // alias_expansion_probe so that the expansion loop recognizes it as |
20c6c071 GH |
410 | // such and re-expands its expansion. |
411 | ||
c1d5f3f6 | 412 | probe * n = new alias_derived_probe (use); |
20c6c071 GH |
413 | n->body = new block(); |
414 | ||
415 | // The new probe gets the location list of the alias, | |
416 | n->locations = alias->locations; | |
417 | ||
418 | // the token location of the use, | |
419 | n->tok = use->tok; | |
5e309481 | 420 | n->body->tok = use->tok; |
20c6c071 | 421 | |
5227f1ea GH |
422 | // and statements representing the concatenation of the alias' |
423 | // body with the use's. | |
424 | // | |
425 | // NB: locals are *not* copied forward, from either alias or | |
426 | // use. The expansion should have its locals re-inferred since | |
427 | // there's concatenated code here and we only want one vardecl per | |
428 | // resulting variable. | |
429 | ||
cedd10f4 FCE |
430 | if (alias->epilogue_style) |
431 | { | |
432 | for (unsigned i = 0; i < use->body->statements.size(); ++i) | |
433 | n->body->statements.push_back | |
434 | (deep_copy_visitor::deep_copy(use->body->statements[i])); | |
b4ceace2 | 435 | |
cedd10f4 FCE |
436 | for (unsigned i = 0; i < alias->body->statements.size(); ++i) |
437 | n->body->statements.push_back | |
438 | (deep_copy_visitor::deep_copy(alias->body->statements[i])); | |
439 | } | |
440 | else | |
441 | { | |
442 | for (unsigned i = 0; i < alias->body->statements.size(); ++i) | |
443 | n->body->statements.push_back | |
444 | (deep_copy_visitor::deep_copy(alias->body->statements[i])); | |
445 | ||
446 | for (unsigned i = 0; i < use->body->statements.size(); ++i) | |
447 | n->body->statements.push_back | |
448 | (deep_copy_visitor::deep_copy(use->body->statements[i])); | |
449 | } | |
450 | ||
451 | derive_probes (sess, n, finished_results, location->optional); | |
20c6c071 GH |
452 | } |
453 | }; | |
454 | ||
455 | ||
456 | // ------------------------------------------------------------------------ | |
457 | // Pattern matching | |
458 | // ------------------------------------------------------------------------ | |
459 | ||
460 | ||
461 | // Register all the aliases we've seen in library files, and the user | |
462 | // file, as patterns. | |
463 | ||
464 | void | |
465 | systemtap_session::register_library_aliases() | |
466 | { | |
467 | vector<stapfile*> files(library_files); | |
468 | files.push_back(user_file); | |
469 | ||
470 | for (unsigned f = 0; f < files.size(); ++f) | |
471 | { | |
472 | stapfile * file = files[f]; | |
473 | for (unsigned a = 0; a < file->aliases.size(); ++a) | |
474 | { | |
475 | probe_alias * alias = file->aliases[a]; | |
59bafbe8 FCE |
476 | try |
477 | { | |
478 | for (unsigned n = 0; n < alias->alias_names.size(); ++n) | |
479 | { | |
480 | probe_point * name = alias->alias_names[n]; | |
481 | match_node * n = pattern_root; | |
482 | for (unsigned c = 0; c < name->components.size(); ++c) | |
483 | { | |
484 | probe_point::component * comp = name->components[c]; | |
485 | // XXX: alias parameters | |
486 | if (comp->arg) | |
487 | throw semantic_error("alias component " | |
488 | + comp->functor | |
489 | + " contains illegal parameter"); | |
490 | n = n->bind(comp->functor); | |
491 | } | |
492 | n->bind(new alias_expansion_builder(alias)); | |
493 | } | |
494 | } | |
495 | catch (const semantic_error& e) | |
496 | { | |
7e41d3dc FCE |
497 | semantic_error* er = new semantic_error (e); // copy it |
498 | stringstream msg; | |
499 | msg << e.msg2; | |
500 | msg << " while registering probe alias "; | |
501 | alias->printsig(msg); | |
502 | er->msg2 = msg.str(); | |
503 | print_error (* er); | |
504 | delete er; | |
59bafbe8 | 505 | } |
20c6c071 GH |
506 | } |
507 | } | |
508 | } | |
509 | ||
510 | ||
5227f1ea GH |
511 | static unsigned max_recursion = 100; |
512 | ||
513 | struct | |
514 | recursion_guard | |
515 | { | |
516 | unsigned & i; | |
517 | recursion_guard(unsigned & i) : i(i) | |
518 | { | |
519 | if (i > max_recursion) | |
520 | throw semantic_error("recursion limit reached"); | |
521 | ++i; | |
522 | } | |
523 | ~recursion_guard() | |
524 | { | |
525 | --i; | |
526 | } | |
527 | }; | |
528 | ||
20c6c071 GH |
529 | // The match-and-expand loop. |
530 | void | |
b4ceace2 FCE |
531 | derive_probes (systemtap_session& s, |
532 | probe *p, vector<derived_probe*>& dps, | |
cedd10f4 | 533 | bool optional) |
20c6c071 GH |
534 | { |
535 | for (unsigned i = 0; i < p->locations.size(); ++i) | |
536 | { | |
537 | probe_point *loc = p->locations[i]; | |
a971b891 FCE |
538 | |
539 | // Pass down optional flag from e.g. alias reference to each | |
540 | // probe_point instance. We do this by temporarily overriding | |
541 | // the probe_point optional flag. We could instead deep-copy | |
542 | // and set a flag on the copy permanently. | |
543 | ||
544 | bool old_loc_opt = loc->optional; | |
545 | loc->optional = loc->optional || optional; | |
546 | ||
fe3d01fa FCE |
547 | try |
548 | { | |
fe3d01fa | 549 | unsigned num_atbegin = dps.size(); |
b4ceace2 | 550 | s.pattern_root->find_and_build (s, p, loc, 0, dps); |
fe3d01fa | 551 | unsigned num_atend = dps.size(); |
b4ceace2 | 552 | |
a971b891 | 553 | if (! loc->optional && // something required, but |
cedd10f4 | 554 | num_atbegin == num_atend) // nothing new derived! |
fe3d01fa FCE |
555 | throw semantic_error ("no match for probe point"); |
556 | } | |
557 | catch (const semantic_error& e) | |
558 | { | |
b4ceace2 | 559 | // XXX: prefer not to print_error at every nest/unroll level |
7e41d3dc FCE |
560 | |
561 | semantic_error* er = new semantic_error (e); // copy it | |
562 | stringstream msg; | |
563 | msg << e.msg2; | |
564 | msg << " while resolving probe point " << *loc; | |
565 | er->msg2 = msg.str(); | |
566 | s.print_error (* er); | |
567 | delete er; | |
fe3d01fa | 568 | } |
a971b891 FCE |
569 | |
570 | loc->optional = old_loc_opt; | |
20c6c071 GH |
571 | } |
572 | } | |
573 | ||
b4ceace2 FCE |
574 | |
575 | ||
20c6c071 | 576 | // ------------------------------------------------------------------------ |
67c0a579 | 577 | // |
d02548c0 | 578 | // Indexable usage checks |
67c0a579 GH |
579 | // |
580 | ||
d02548c0 | 581 | struct symbol_fetcher |
07c17d67 | 582 | : public throwing_visitor |
67c0a579 | 583 | { |
d02548c0 | 584 | symbol *&sym; |
67c0a579 | 585 | |
d02548c0 GH |
586 | symbol_fetcher (symbol *&sym) |
587 | : sym(sym) | |
67c0a579 GH |
588 | {} |
589 | ||
d02548c0 GH |
590 | void visit_symbol (symbol* e) |
591 | { | |
592 | sym = e; | |
593 | } | |
594 | ||
595 | void visit_arrayindex (arrayindex* e) | |
596 | { | |
597 | e->base->visit_indexable (this); | |
598 | } | |
599 | ||
600 | void throwone (const token* t) | |
601 | { | |
602 | throw semantic_error ("Expecting symbol or array index expression", t); | |
603 | } | |
604 | }; | |
605 | ||
07c17d67 | 606 | symbol * |
d02548c0 GH |
607 | get_symbol_within_expression (expression *e) |
608 | { | |
609 | symbol *sym = NULL; | |
610 | symbol_fetcher fetcher(sym); | |
611 | e->visit (&fetcher); | |
612 | if (!sym) | |
613 | throw semantic_error("Unable to find symbol in expression", e->tok); | |
614 | return sym; | |
615 | } | |
616 | ||
617 | static symbol * | |
618 | get_symbol_within_indexable (indexable *ix) | |
619 | { | |
620 | symbol *array = NULL; | |
621 | hist_op *hist = NULL; | |
622 | classify_indexable(ix, array, hist); | |
623 | if (array) | |
624 | return array; | |
625 | else | |
626 | return get_symbol_within_expression (hist->stat); | |
627 | } | |
628 | ||
629 | struct mutated_var_collector | |
07c17d67 | 630 | : public traversing_visitor |
d02548c0 GH |
631 | { |
632 | set<vardecl *> * mutated_vars; | |
633 | ||
634 | mutated_var_collector (set<vardecl *> * mm) | |
635 | : mutated_vars (mm) | |
636 | {} | |
637 | ||
638 | void visit_assignment(assignment* e) | |
639 | { | |
640 | if (e->type == pe_stats && e->op == "<<<") | |
641 | { | |
642 | vardecl *vd = get_symbol_within_expression (e->left)->referent; | |
643 | if (vd) | |
644 | mutated_vars->insert (vd); | |
645 | } | |
1bbeef03 | 646 | traversing_visitor::visit_assignment(e); |
d02548c0 GH |
647 | } |
648 | ||
67c0a579 GH |
649 | void visit_arrayindex (arrayindex *e) |
650 | { | |
d02548c0 GH |
651 | if (is_active_lvalue (e)) |
652 | { | |
653 | symbol *sym; | |
654 | if (e->base->is_symbol (sym)) | |
655 | mutated_vars->insert (sym->referent); | |
656 | else | |
657 | throw semantic_error("Assignment to read-only histogram bucket", e->tok); | |
658 | } | |
1bbeef03 | 659 | traversing_visitor::visit_arrayindex (e); |
67c0a579 GH |
660 | } |
661 | }; | |
662 | ||
663 | ||
d02548c0 | 664 | struct no_var_mutation_during_iteration_check |
07c17d67 | 665 | : public traversing_visitor |
67c0a579 GH |
666 | { |
667 | systemtap_session & session; | |
d02548c0 GH |
668 | map<functiondecl *,set<vardecl *> *> & function_mutates_vars; |
669 | vector<vardecl *> vars_being_iterated; | |
67c0a579 | 670 | |
d02548c0 | 671 | no_var_mutation_during_iteration_check |
67c0a579 | 672 | (systemtap_session & sess, |
d02548c0 GH |
673 | map<functiondecl *,set<vardecl *> *> & fmv) |
674 | : session(sess), function_mutates_vars (fmv) | |
67c0a579 GH |
675 | {} |
676 | ||
677 | void visit_arrayindex (arrayindex *e) | |
678 | { | |
d7f3e0c5 | 679 | if (is_active_lvalue(e)) |
67c0a579 | 680 | { |
d02548c0 GH |
681 | vardecl *vd = get_symbol_within_indexable (e->base)->referent; |
682 | if (vd) | |
67c0a579 | 683 | { |
d02548c0 | 684 | for (unsigned i = 0; i < vars_being_iterated.size(); ++i) |
67c0a579 | 685 | { |
d02548c0 GH |
686 | vardecl *v = vars_being_iterated[i]; |
687 | if (v == vd) | |
688 | { | |
689 | string err = ("variable '" + v->name + | |
690 | "' modified during 'foreach' iteration"); | |
691 | session.print_error (semantic_error (err, e->tok)); | |
692 | } | |
67c0a579 GH |
693 | } |
694 | } | |
695 | } | |
1bbeef03 | 696 | traversing_visitor::visit_arrayindex (e); |
67c0a579 GH |
697 | } |
698 | ||
699 | void visit_functioncall (functioncall* e) | |
700 | { | |
701 | map<functiondecl *,set<vardecl *> *>::const_iterator i | |
d02548c0 | 702 | = function_mutates_vars.find (e->referent); |
67c0a579 | 703 | |
d02548c0 | 704 | if (i != function_mutates_vars.end()) |
67c0a579 | 705 | { |
d02548c0 | 706 | for (unsigned j = 0; j < vars_being_iterated.size(); ++j) |
67c0a579 | 707 | { |
d02548c0 | 708 | vardecl *m = vars_being_iterated[j]; |
67c0a579 GH |
709 | if (i->second->find (m) != i->second->end()) |
710 | { | |
d02548c0 | 711 | string err = ("function call modifies var '" + m->name + |
67c0a579 GH |
712 | "' during 'foreach' iteration"); |
713 | session.print_error (semantic_error (err, e->tok)); | |
714 | } | |
715 | } | |
716 | } | |
717 | ||
1bbeef03 | 718 | traversing_visitor::visit_functioncall (e); |
67c0a579 GH |
719 | } |
720 | ||
721 | void visit_foreach_loop(foreach_loop* s) | |
722 | { | |
d02548c0 GH |
723 | vardecl *vd = get_symbol_within_indexable (s->base)->referent; |
724 | ||
725 | if (vd) | |
726 | vars_being_iterated.push_back (vd); | |
727 | ||
1bbeef03 | 728 | traversing_visitor::visit_foreach_loop (s); |
d02548c0 GH |
729 | |
730 | if (vd) | |
731 | vars_being_iterated.pop_back(); | |
67c0a579 GH |
732 | } |
733 | }; | |
20c6c071 | 734 | |
2b066ec1 | 735 | |
67c0a579 | 736 | static int |
d02548c0 | 737 | semantic_pass_vars (systemtap_session & sess) |
67c0a579 GH |
738 | { |
739 | ||
d02548c0 GH |
740 | map<functiondecl *, set<vardecl *> *> fmv; |
741 | no_var_mutation_during_iteration_check chk(sess, fmv); | |
67c0a579 GH |
742 | |
743 | for (unsigned i = 0; i < sess.functions.size(); ++i) | |
744 | { | |
745 | functiondecl * fn = sess.functions[i]; | |
746 | if (fn->body) | |
747 | { | |
748 | set<vardecl *> * m = new set<vardecl *>(); | |
d02548c0 | 749 | mutated_var_collector mc (m); |
67c0a579 | 750 | fn->body->visit (&mc); |
d02548c0 | 751 | fmv[fn] = m; |
67c0a579 GH |
752 | } |
753 | } | |
754 | ||
755 | for (unsigned i = 0; i < sess.functions.size(); ++i) | |
756 | { | |
757 | if (sess.functions[i]->body) | |
758 | sess.functions[i]->body->visit (&chk); | |
759 | } | |
760 | ||
761 | for (unsigned i = 0; i < sess.probes.size(); ++i) | |
762 | { | |
763 | if (sess.probes[i]->body) | |
764 | sess.probes[i]->body->visit (&chk); | |
765 | } | |
766 | ||
7e41d3dc | 767 | return sess.num_errors(); |
67c0a579 GH |
768 | } |
769 | ||
770 | // ------------------------------------------------------------------------ | |
771 | ||
07c17d67 GH |
772 | struct stat_decl_collector |
773 | : public traversing_visitor | |
774 | { | |
775 | systemtap_session & session; | |
776 | ||
777 | stat_decl_collector(systemtap_session & sess) | |
778 | : session(sess) | |
779 | {} | |
780 | ||
781 | void visit_stat_op (stat_op* e) | |
782 | { | |
783 | symbol *sym = get_symbol_within_expression (e->stat); | |
784 | if (session.stat_decls.find(sym->name) == session.stat_decls.end()) | |
785 | session.stat_decls[sym->name] = statistic_decl(); | |
786 | } | |
787 | ||
788 | void visit_assignment (assignment* e) | |
789 | { | |
790 | if (e->op == "<<<") | |
791 | { | |
792 | symbol *sym = get_symbol_within_expression (e->left); | |
793 | if (session.stat_decls.find(sym->name) == session.stat_decls.end()) | |
794 | session.stat_decls[sym->name] = statistic_decl(); | |
795 | } | |
796 | else | |
797 | traversing_visitor::visit_assignment(e); | |
798 | } | |
799 | ||
800 | void visit_hist_op (hist_op* e) | |
801 | { | |
802 | symbol *sym = get_symbol_within_expression (e->stat); | |
803 | statistic_decl new_stat; | |
804 | ||
805 | if (e->htype == hist_linear) | |
806 | { | |
807 | new_stat.type = statistic_decl::linear; | |
808 | assert (e->params.size() == 3); | |
809 | new_stat.linear_low = e->params[0]; | |
810 | new_stat.linear_high = e->params[1]; | |
811 | new_stat.linear_step = e->params[2]; | |
812 | } | |
813 | else | |
814 | { | |
815 | assert (e->htype == hist_log); | |
816 | new_stat.type = statistic_decl::logarithmic; | |
817 | assert (e->params.size() == 1); | |
818 | new_stat.logarithmic_buckets = e->params[0]; | |
819 | } | |
820 | ||
821 | map<string, statistic_decl>::iterator i = session.stat_decls.find(sym->name); | |
822 | if (i == session.stat_decls.end()) | |
823 | session.stat_decls[sym->name] = new_stat; | |
824 | else | |
825 | { | |
826 | statistic_decl & old_stat = i->second; | |
827 | if (!(old_stat == new_stat)) | |
828 | { | |
829 | if (old_stat.type == statistic_decl::none) | |
830 | i->second = new_stat; | |
831 | else | |
832 | { | |
833 | // FIXME: Support multiple co-declared histogram types | |
834 | semantic_error se("multiple histogram types declared on '" + sym->name + "'", | |
835 | e->tok); | |
836 | session.print_error (se); | |
837 | } | |
838 | } | |
839 | } | |
840 | } | |
841 | ||
842 | }; | |
843 | ||
844 | static int | |
845 | semantic_pass_stats (systemtap_session & sess) | |
846 | { | |
847 | stat_decl_collector sdc(sess); | |
848 | ||
849 | for (unsigned i = 0; i < sess.functions.size(); ++i) | |
850 | sess.functions[i]->body->visit (&sdc); | |
851 | ||
852 | for (unsigned i = 0; i < sess.probes.size(); ++i) | |
853 | sess.probes[i]->body->visit (&sdc); | |
854 | ||
855 | for (unsigned i = 0; i < sess.globals.size(); ++i) | |
856 | { | |
857 | vardecl *v = sess.globals[i]; | |
858 | if (v->type == pe_stats) | |
859 | { | |
860 | ||
861 | if (sess.stat_decls.find(v->name) == sess.stat_decls.end()) | |
862 | { | |
863 | semantic_error se("unable to infer statistic parameters for global '" + v->name + "'"); | |
864 | sess.print_error (se); | |
865 | } | |
866 | } | |
867 | } | |
868 | ||
7e41d3dc | 869 | return sess.num_errors(); |
07c17d67 GH |
870 | } |
871 | ||
872 | // ------------------------------------------------------------------------ | |
873 | ||
2b066ec1 FCE |
874 | |
875 | static int semantic_pass_symbols (systemtap_session&); | |
c214bd6a DS |
876 | static int semantic_pass_optimize1 (systemtap_session&); |
877 | static int semantic_pass_optimize2 (systemtap_session&); | |
2b066ec1 | 878 | static int semantic_pass_types (systemtap_session&); |
d02548c0 | 879 | static int semantic_pass_vars (systemtap_session&); |
07c17d67 | 880 | static int semantic_pass_stats (systemtap_session&); |
2b066ec1 FCE |
881 | |
882 | ||
883 | ||
884 | // Link up symbols to their declarations. Set the session's | |
885 | // files/probes/functions/globals vectors from the transitively | |
886 | // reached set of stapfiles in s.library_files, starting from | |
20c6c071 | 887 | // s.user_file. Perform automatic tapset inclusion and probe |
2b066ec1 FCE |
888 | // alias expansion. |
889 | static int | |
890 | semantic_pass_symbols (systemtap_session& s) | |
891 | { | |
892 | symresolution_info sym (s); | |
893 | ||
894 | // NB: s.files can grow during this iteration, so size() can | |
895 | // return gradually increasing numbers. | |
896 | s.files.push_back (s.user_file); | |
897 | for (unsigned i = 0; i < s.files.size(); i++) | |
898 | { | |
899 | stapfile* dome = s.files[i]; | |
900 | ||
901 | // Pass 1: add globals and functions to systemtap-session master list, | |
902 | // so the find_* functions find them | |
903 | ||
904 | for (unsigned i=0; i<dome->globals.size(); i++) | |
905 | s.globals.push_back (dome->globals[i]); | |
906 | ||
907 | for (unsigned i=0; i<dome->functions.size(); i++) | |
908 | s.functions.push_back (dome->functions[i]); | |
909 | ||
54dfabe9 FCE |
910 | for (unsigned i=0; i<dome->embeds.size(); i++) |
911 | s.embeds.push_back (dome->embeds[i]); | |
912 | ||
2b066ec1 FCE |
913 | // Pass 2: process functions |
914 | ||
915 | for (unsigned i=0; i<dome->functions.size(); i++) | |
916 | { | |
917 | functiondecl* fd = dome->functions[i]; | |
918 | ||
919 | try | |
920 | { | |
8a43522c GH |
921 | sym.current_function = fd; |
922 | sym.current_probe = 0; | |
923 | fd->body->visit (& sym); | |
2b066ec1 FCE |
924 | } |
925 | catch (const semantic_error& e) | |
926 | { | |
927 | s.print_error (e); | |
928 | } | |
929 | } | |
930 | ||
5227f1ea GH |
931 | // Pass 3: derive probes and resolve any further symbols in the |
932 | // derived results. | |
2b066ec1 FCE |
933 | |
934 | for (unsigned i=0; i<dome->probes.size(); i++) | |
935 | { | |
936 | probe* p = dome->probes [i]; | |
937 | vector<derived_probe*> dps; | |
938 | ||
a971b891 FCE |
939 | // much magic happens here: probe alias expansion, wildcard |
940 | // matching, low-level derived_probe construction. | |
b4ceace2 | 941 | derive_probes (s, p, dps); |
2b066ec1 FCE |
942 | |
943 | for (unsigned j=0; j<dps.size(); j++) | |
944 | { | |
945 | derived_probe* dp = dps[j]; | |
b20febf3 FCE |
946 | s.probes.push_back (dp); |
947 | dp->join_group (s); | |
2b066ec1 FCE |
948 | |
949 | try | |
950 | { | |
951 | sym.current_function = 0; | |
5227f1ea | 952 | sym.current_probe = dp; |
2b066ec1 FCE |
953 | dp->body->visit (& sym); |
954 | } | |
955 | catch (const semantic_error& e) | |
956 | { | |
957 | s.print_error (e); | |
958 | } | |
959 | } | |
960 | } | |
961 | } | |
aa30ccd3 FCE |
962 | |
963 | // Inform all derived_probe builders that we're done with | |
964 | // all resolution, so it's time to release caches. | |
965 | s.pattern_root->build_no_more (s); | |
fe3d01fa | 966 | |
7e41d3dc | 967 | return s.num_errors(); // all those print_error calls |
2b066ec1 FCE |
968 | } |
969 | ||
970 | ||
971 | ||
972 | int | |
973 | semantic_pass (systemtap_session& s) | |
974 | { | |
59bafbe8 | 975 | int rc = 0; |
20c6c071 | 976 | |
59bafbe8 FCE |
977 | try |
978 | { | |
979 | s.register_library_aliases(); | |
980 | register_standard_tapsets(s); | |
981 | ||
982 | rc = semantic_pass_symbols (s); | |
c214bd6a | 983 | if (rc == 0 && ! s.unoptimized) rc = semantic_pass_optimize1 (s); |
59bafbe8 | 984 | if (rc == 0) rc = semantic_pass_types (s); |
c214bd6a | 985 | if (rc == 0 && ! s.unoptimized) rc = semantic_pass_optimize2 (s); |
d02548c0 | 986 | if (rc == 0) rc = semantic_pass_vars (s); |
07c17d67 | 987 | if (rc == 0) rc = semantic_pass_stats (s); |
59bafbe8 FCE |
988 | } |
989 | catch (const semantic_error& e) | |
990 | { | |
991 | s.print_error (e); | |
992 | } | |
993 | ||
2b066ec1 FCE |
994 | return rc; |
995 | } | |
996 | ||
997 | ||
998 | // ------------------------------------------------------------------------ | |
999 | ||
1000 | ||
1001 | systemtap_session::systemtap_session (): | |
b20febf3 | 1002 | // NB: pointer members must be manually initialized! |
20c6c071 | 1003 | pattern_root(new match_node), |
b20febf3 FCE |
1004 | user_file (0), |
1005 | be_derived_probes(0), | |
1006 | dwarf_derived_probes(0), | |
1007 | timer_derived_probes(0), | |
1008 | profile_derived_probes(0), | |
1009 | mark_derived_probes(0), | |
1010 | hrtimer_derived_probes(0), | |
1011 | perfmon_derived_probes(0), | |
84048984 | 1012 | op (0), up (0) |
2b066ec1 FCE |
1013 | { |
1014 | } | |
1015 | ||
1016 | ||
1017 | void | |
1018 | systemtap_session::print_error (const semantic_error& e) | |
1019 | { | |
7e41d3dc FCE |
1020 | string message_str; |
1021 | stringstream message; | |
1022 | ||
1023 | message << "semantic error: " << e.what (); | |
d8067b24 | 1024 | if (e.tok1 || e.tok2) |
7e41d3dc FCE |
1025 | message << ": "; |
1026 | if (e.tok1) message << *e.tok1; | |
1027 | message << e.msg2; | |
1028 | if (e.tok2) message << *e.tok2; | |
1029 | message << endl; | |
1030 | message_str = message.str(); | |
1031 | ||
1032 | // Duplicate elimination | |
1033 | if (seen_errors.find (message_str) == seen_errors.end()) | |
1034 | { | |
1035 | seen_errors.insert (message_str); | |
1036 | cerr << message_str; | |
1037 | } | |
1038 | ||
1039 | if (e.chain) | |
1040 | print_error (* e.chain); | |
2b066ec1 FCE |
1041 | } |
1042 | ||
1043 | ||
1044 | // ------------------------------------------------------------------------ | |
1045 | // semantic processing: symbol resolution | |
1046 | ||
1047 | ||
1048 | symresolution_info::symresolution_info (systemtap_session& s): | |
5227f1ea | 1049 | session (s), current_function (0), current_probe (0) |
2b066ec1 FCE |
1050 | { |
1051 | } | |
1052 | ||
1053 | ||
1054 | void | |
1055 | symresolution_info::visit_block (block* e) | |
1056 | { | |
1057 | for (unsigned i=0; i<e->statements.size(); i++) | |
1058 | { | |
1059 | try | |
1060 | { | |
1061 | e->statements[i]->visit (this); | |
1062 | } | |
1063 | catch (const semantic_error& e) | |
1064 | { | |
1065 | session.print_error (e); | |
1066 | } | |
1067 | } | |
1068 | } | |
1069 | ||
1070 | ||
69c68955 FCE |
1071 | void |
1072 | symresolution_info::visit_foreach_loop (foreach_loop* e) | |
1073 | { | |
1074 | for (unsigned i=0; i<e->indexes.size(); i++) | |
1075 | e->indexes[i]->visit (this); | |
1076 | ||
d02548c0 GH |
1077 | symbol *array = NULL; |
1078 | hist_op *hist = NULL; | |
1079 | classify_indexable (e->base, array, hist); | |
69c68955 | 1080 | |
d02548c0 GH |
1081 | if (array) |
1082 | { | |
1083 | if (!array->referent) | |
1084 | { | |
1085 | vardecl* d = find_var (array->name, e->indexes.size ()); | |
1086 | if (d) | |
1087 | array->referent = d; | |
1088 | else | |
2a99f48f | 1089 | { |
7e41d3dc FCE |
1090 | stringstream msg; |
1091 | msg << "unresolved arity-" << e->indexes.size() | |
1092 | << " global array " << array->name; | |
1093 | throw semantic_error (msg.str(), e->tok); | |
2a99f48f | 1094 | } |
d02548c0 GH |
1095 | } |
1096 | } | |
1097 | else | |
1098 | { | |
1099 | assert (hist); | |
1100 | hist->visit (this); | |
1101 | } | |
69c68955 | 1102 | |
27f21e8c DS |
1103 | if (e->limit) |
1104 | e->limit->visit (this); | |
1105 | ||
69c68955 FCE |
1106 | e->block->visit (this); |
1107 | } | |
1108 | ||
d02548c0 | 1109 | |
d98d459c GH |
1110 | struct |
1111 | delete_statement_symresolution_info: | |
1112 | public traversing_visitor | |
1113 | { | |
1114 | symresolution_info *parent; | |
1115 | ||
1116 | delete_statement_symresolution_info (symresolution_info *p): | |
1117 | parent(p) | |
1118 | {} | |
1119 | ||
1120 | void visit_arrayindex (arrayindex* e) | |
1121 | { | |
1122 | parent->visit_arrayindex (e); | |
1123 | } | |
1124 | void visit_functioncall (functioncall* e) | |
1125 | { | |
1126 | parent->visit_functioncall (e); | |
1127 | } | |
1128 | ||
1129 | void visit_symbol (symbol* e) | |
1130 | { | |
1131 | if (e->referent) | |
1132 | return; | |
1133 | ||
1134 | vardecl* d = parent->find_var (e->name, -1); | |
1135 | if (d) | |
1136 | e->referent = d; | |
1137 | else | |
1138 | throw semantic_error ("unresolved array in delete statement", e->tok); | |
1139 | } | |
1140 | }; | |
1141 | ||
1142 | void | |
1143 | symresolution_info::visit_delete_statement (delete_statement* s) | |
1144 | { | |
1145 | delete_statement_symresolution_info di (this); | |
1146 | s->value->visit (&di); | |
1147 | } | |
1148 | ||
69c68955 | 1149 | |
2b066ec1 FCE |
1150 | void |
1151 | symresolution_info::visit_symbol (symbol* e) | |
1152 | { | |
1153 | if (e->referent) | |
1154 | return; | |
1155 | ||
313b2f74 | 1156 | vardecl* d = find_var (e->name, 0); |
2b066ec1 FCE |
1157 | if (d) |
1158 | e->referent = d; | |
1159 | else | |
1160 | { | |
1161 | // new local | |
1162 | vardecl* v = new vardecl; | |
1163 | v->name = e->name; | |
1164 | v->tok = e->tok; | |
1165 | if (current_function) | |
1166 | current_function->locals.push_back (v); | |
1167 | else if (current_probe) | |
1168 | current_probe->locals.push_back (v); | |
1169 | else | |
ce10591c FCE |
1170 | // must not happen |
1171 | throw semantic_error ("no current probe/function", e->tok); | |
2b066ec1 FCE |
1172 | e->referent = v; |
1173 | } | |
1174 | } | |
1175 | ||
1176 | ||
1177 | void | |
1178 | symresolution_info::visit_arrayindex (arrayindex* e) | |
1179 | { | |
1180 | for (unsigned i=0; i<e->indexes.size(); i++) | |
1181 | e->indexes[i]->visit (this); | |
1182 | ||
d02548c0 GH |
1183 | symbol *array = NULL; |
1184 | hist_op *hist = NULL; | |
1185 | classify_indexable(e->base, array, hist); | |
2b066ec1 | 1186 | |
d02548c0 | 1187 | if (array) |
313b2f74 | 1188 | { |
d02548c0 GH |
1189 | if (array->referent) |
1190 | return; | |
1191 | ||
1192 | vardecl* d = find_var (array->name, e->indexes.size ()); | |
1193 | if (d) | |
1194 | array->referent = d; | |
313b2f74 | 1195 | else |
d02548c0 GH |
1196 | { |
1197 | // new local | |
1198 | vardecl* v = new vardecl; | |
1199 | v->set_arity(e->indexes.size()); | |
1200 | v->name = array->name; | |
1201 | v->tok = array->tok; | |
1202 | if (current_function) | |
1203 | current_function->locals.push_back (v); | |
1204 | else if (current_probe) | |
1205 | current_probe->locals.push_back (v); | |
1206 | else | |
1207 | // must not happen | |
1208 | throw semantic_error ("no current probe/function", e->tok); | |
1209 | array->referent = v; | |
1210 | } | |
1211 | } | |
1212 | else | |
1213 | { | |
1214 | assert (hist); | |
1215 | hist->visit (this); | |
313b2f74 | 1216 | } |
2b066ec1 FCE |
1217 | } |
1218 | ||
1219 | ||
1220 | void | |
1221 | symresolution_info::visit_functioncall (functioncall* e) | |
1222 | { | |
1223 | for (unsigned i=0; i<e->args.size(); i++) | |
1224 | e->args[i]->visit (this); | |
1225 | ||
1226 | if (e->referent) | |
1227 | return; | |
1228 | ||
1229 | functiondecl* d = find_function (e->function, e->args.size ()); | |
1230 | if (d) | |
1231 | e->referent = d; | |
1232 | else | |
2a99f48f | 1233 | { |
7e41d3dc FCE |
1234 | stringstream msg; |
1235 | msg << "unresolved arity-" << e->args.size() | |
1236 | << " function"; | |
1237 | throw semantic_error (msg.str(), e->tok); | |
2a99f48f | 1238 | } |
2b066ec1 FCE |
1239 | } |
1240 | ||
1241 | ||
1242 | vardecl* | |
d98d459c | 1243 | symresolution_info::find_var (const string& name, int arity) |
2b066ec1 | 1244 | { |
313b2f74 | 1245 | |
2b066ec1 FCE |
1246 | // search locals |
1247 | vector<vardecl*>& locals = (current_function ? | |
1248 | current_function->locals : | |
5227f1ea | 1249 | current_probe->locals); |
6deb8173 | 1250 | |
313b2f74 | 1251 | |
2b066ec1 | 1252 | for (unsigned i=0; i<locals.size(); i++) |
313b2f74 GH |
1253 | if (locals[i]->name == name |
1254 | && locals[i]->compatible_arity(arity)) | |
8846477c | 1255 | { |
313b2f74 | 1256 | locals[i]->set_arity (arity); |
8846477c FCE |
1257 | return locals[i]; |
1258 | } | |
2b066ec1 | 1259 | |
313b2f74 GH |
1260 | // search function formal parameters (for scalars) |
1261 | if (arity == 0 && current_function) | |
2b066ec1 FCE |
1262 | for (unsigned i=0; i<current_function->formal_args.size(); i++) |
1263 | if (current_function->formal_args[i]->name == name) | |
8846477c FCE |
1264 | { |
1265 | // NB: no need to check arity here: formal args always scalar | |
1266 | current_function->formal_args[i]->set_arity (0); | |
1267 | return current_function->formal_args[i]; | |
1268 | } | |
2b066ec1 | 1269 | |
313b2f74 | 1270 | // search processed globals |
2b066ec1 | 1271 | for (unsigned i=0; i<session.globals.size(); i++) |
313b2f74 | 1272 | if (session.globals[i]->name == name |
d98d459c | 1273 | && session.globals[i]->compatible_arity(arity)) |
8846477c | 1274 | { |
313b2f74 | 1275 | session.globals[i]->set_arity (arity); |
8846477c FCE |
1276 | return session.globals[i]; |
1277 | } | |
d98d459c | 1278 | |
2b066ec1 FCE |
1279 | // search library globals |
1280 | for (unsigned i=0; i<session.library_files.size(); i++) | |
1281 | { | |
1282 | stapfile* f = session.library_files[i]; | |
1283 | for (unsigned j=0; j<f->globals.size(); j++) | |
84e5ea0f FCE |
1284 | { |
1285 | vardecl* g = f->globals[j]; | |
1286 | if (g->name == name && g->compatible_arity (arity)) | |
1287 | { | |
d98d459c | 1288 | g->set_arity (arity); |
84e5ea0f FCE |
1289 | |
1290 | // put library into the queue if not already there | |
1291 | if (find (session.files.begin(), session.files.end(), f) | |
1292 | == session.files.end()) | |
1293 | session.files.push_back (f); | |
1294 | ||
1295 | return g; | |
1296 | } | |
1297 | } | |
2b066ec1 FCE |
1298 | } |
1299 | ||
2b066ec1 | 1300 | return 0; |
2b066ec1 FCE |
1301 | } |
1302 | ||
1303 | ||
2b066ec1 FCE |
1304 | functiondecl* |
1305 | symresolution_info::find_function (const string& name, unsigned arity) | |
1306 | { | |
1307 | for (unsigned j = 0; j < session.functions.size(); j++) | |
1308 | { | |
1309 | functiondecl* fd = session.functions[j]; | |
1310 | if (fd->name == name && | |
1311 | fd->formal_args.size() == arity) | |
1312 | return fd; | |
1313 | } | |
1314 | ||
1315 | // search library globals | |
1316 | for (unsigned i=0; i<session.library_files.size(); i++) | |
1317 | { | |
1318 | stapfile* f = session.library_files[i]; | |
1319 | for (unsigned j=0; j<f->functions.size(); j++) | |
1320 | if (f->functions[j]->name == name && | |
1321 | f->functions[j]->formal_args.size() == arity) | |
1322 | { | |
1323 | // put library into the queue if not already there | |
1324 | if (0) // session.verbose_resolution | |
1325 | cerr << " function " << name << " " | |
1326 | << "is defined from " << f->name << endl; | |
1327 | ||
1328 | if (find (session.files.begin(), session.files.end(), f) | |
1329 | == session.files.end()) | |
1330 | session.files.push_back (f); | |
1331 | // else .. print different message? | |
1332 | ||
1333 | return f->functions[j]; | |
1334 | } | |
1335 | } | |
1336 | ||
1337 | return 0; | |
2b066ec1 FCE |
1338 | } |
1339 | ||
1340 | ||
cbfbbf69 FCE |
1341 | |
1342 | // ------------------------------------------------------------------------ | |
1343 | // optimization | |
1344 | ||
1345 | ||
1346 | // Do away with functiondecls that are never (transitively) called | |
1347 | // from probes. | |
1348 | void semantic_pass_opt1 (systemtap_session& s, bool& relaxed_p) | |
1349 | { | |
1350 | functioncall_traversing_visitor ftv; | |
1351 | for (unsigned i=0; i<s.probes.size(); i++) | |
1352 | s.probes[i]->body->visit (& ftv); | |
1353 | for (unsigned i=0; i<s.functions.size(); /* see below */) | |
1354 | { | |
1355 | if (ftv.traversed.find(s.functions[i]) == ftv.traversed.end()) | |
1356 | { | |
b0ee93c4 | 1357 | if (s.verbose>2) |
cbfbbf69 FCE |
1358 | clog << "Eliding unused function " << s.functions[i]->name |
1359 | << endl; | |
1360 | s.functions.erase (s.functions.begin() + i); | |
1361 | relaxed_p = false; | |
1362 | // NB: don't increment i | |
1363 | } | |
1364 | else | |
1365 | i++; | |
1366 | } | |
1367 | } | |
1368 | ||
1369 | ||
1370 | // ------------------------------------------------------------------------ | |
1371 | ||
1372 | // Do away with local & global variables that are never | |
1373 | // written nor read. | |
1374 | void semantic_pass_opt2 (systemtap_session& s, bool& relaxed_p) | |
1375 | { | |
1376 | varuse_collecting_visitor vut; | |
1377 | ||
1378 | for (unsigned i=0; i<s.probes.size(); i++) | |
1379 | s.probes[i]->body->visit (& vut); | |
1380 | // NB: Since varuse_collecting_visitor also traverses down | |
1381 | // actually called functions, we don't need to explicitly | |
1382 | // iterate over them. Uncalled ones should have been pruned | |
1383 | // in _opt1 above. | |
1384 | // | |
1385 | // for (unsigned i=0; i<s.functions.size(); i++) | |
1386 | // s.functions[i]->body->visit (& vut); | |
1387 | ||
1388 | // Now in vut.read/written, we have a mixture of all locals, globals | |
1389 | ||
1390 | for (unsigned i=0; i<s.probes.size(); i++) | |
1391 | for (unsigned j=0; j<s.probes[i]->locals.size(); /* see below */) | |
1392 | { | |
1393 | vardecl* l = s.probes[i]->locals[j]; | |
1394 | if (vut.read.find (l) == vut.read.end() && | |
1395 | vut.written.find (l) == vut.written.end()) | |
1396 | { | |
b0ee93c4 | 1397 | if (s.verbose>2) |
cbfbbf69 | 1398 | clog << "Eliding unused local variable " |
1b07c728 | 1399 | << l->name << " in " << s.probes[i]->name << endl; |
cbfbbf69 FCE |
1400 | s.probes[i]->locals.erase(s.probes[i]->locals.begin() + j); |
1401 | relaxed_p = false; | |
1402 | // don't increment j | |
1403 | } | |
1404 | else | |
1405 | j++; | |
1406 | } | |
1407 | for (unsigned i=0; i<s.functions.size(); i++) | |
1408 | for (unsigned j=0; j<s.functions[i]->locals.size(); /* see below */) | |
1409 | { | |
1410 | vardecl* l = s.functions[i]->locals[j]; | |
1411 | if (vut.read.find (l) == vut.read.end() && | |
1412 | vut.written.find (l) == vut.written.end()) | |
1413 | { | |
b0ee93c4 | 1414 | if (s.verbose>2) |
cbfbbf69 FCE |
1415 | clog << "Eliding unused local variable " |
1416 | << l->name << " in function " << s.functions[i]->name | |
1417 | << endl; | |
1418 | s.functions[i]->locals.erase(s.functions[i]->locals.begin() + j); | |
1419 | relaxed_p = false; | |
1420 | // don't increment j | |
1421 | } | |
1422 | else | |
1423 | j++; | |
1424 | } | |
1425 | for (unsigned i=0; i<s.globals.size(); /* see below */) | |
1426 | { | |
1427 | vardecl* l = s.globals[i]; | |
1428 | if (vut.read.find (l) == vut.read.end() && | |
1429 | vut.written.find (l) == vut.written.end()) | |
1430 | { | |
b0ee93c4 | 1431 | if (s.verbose>2) |
cbfbbf69 FCE |
1432 | clog << "Eliding unused global variable " |
1433 | << l->name << endl; | |
1434 | s.globals.erase(s.globals.begin() + i); | |
1435 | relaxed_p = false; | |
1436 | // don't increment i | |
1437 | } | |
1438 | else | |
1439 | i++; | |
1440 | } | |
1441 | } | |
1442 | ||
1443 | ||
1444 | // ------------------------------------------------------------------------ | |
1445 | ||
1446 | struct dead_assignment_remover: public traversing_visitor | |
1447 | { | |
1448 | systemtap_session& session; | |
1449 | bool& relaxed_p; | |
1450 | const varuse_collecting_visitor& vut; | |
1451 | expression** current_expr; | |
1452 | ||
1453 | dead_assignment_remover(systemtap_session& s, bool& r, | |
1454 | const varuse_collecting_visitor& v): | |
1455 | session(s), relaxed_p(r), vut(v), current_expr(0) {} | |
1456 | ||
1457 | void visit_expr_statement (expr_statement* s); | |
72948d62 FCE |
1458 | // XXX: other places where an assignment may be nested should be |
1459 | // handled too (e.g., loop/if conditionals, array indexes, function | |
1460 | // parameters). Until then, they result in visit_assignment() being | |
1461 | // called with null current_expr. | |
cbfbbf69 FCE |
1462 | |
1463 | void visit_assignment (assignment* e); | |
1464 | }; | |
1465 | ||
1466 | ||
1467 | void | |
1468 | dead_assignment_remover::visit_expr_statement (expr_statement* s) | |
1469 | { | |
1470 | expression** last_expr = current_expr; | |
1471 | current_expr = & s->value; | |
1472 | s->value->visit (this); | |
3066c15c | 1473 | s->tok = s->value->tok; // in case it was replaced |
cbfbbf69 FCE |
1474 | current_expr = last_expr; |
1475 | } | |
1476 | ||
1477 | ||
1478 | void | |
1479 | dead_assignment_remover::visit_assignment (assignment* e) | |
1480 | { | |
1481 | symbol* left = get_symbol_within_expression (e->left); | |
1482 | vardecl* leftvar = left->referent; | |
72948d62 FCE |
1483 | if (current_expr && // see XXX above: this case represents a missed |
1484 | // optimization opportunity | |
1485 | *current_expr == e) // we're not nested any deeper than expected | |
cbfbbf69 FCE |
1486 | { |
1487 | // clog << "Checking assignment to " << leftvar->name << " at " << *e->tok << endl; | |
1488 | if (vut.read.find(leftvar) == vut.read.end()) // var never read? | |
1489 | { | |
b0ee93c4 | 1490 | if (session.verbose>2) |
cbfbbf69 FCE |
1491 | clog << "Eliding assignment to " << leftvar->name |
1492 | << " at " << *e->tok << endl; | |
1493 | *current_expr = e->right; // goodbye assignment* | |
1494 | relaxed_p = false; | |
1495 | } | |
1496 | } | |
1497 | } | |
1498 | ||
1499 | ||
1500 | // Let's remove assignments to variables that are never read. We | |
1501 | // rewrite "(foo = expr)" as "(expr)". This makes foo a candidate to | |
1502 | // be optimized away as an unused variable, and expr a candidate to be | |
1503 | // removed as a side-effect-free statement expression. Wahoo! | |
1504 | void semantic_pass_opt3 (systemtap_session& s, bool& relaxed_p) | |
1505 | { | |
1506 | // Recompute the varuse data, which will probably match the opt2 | |
1507 | // copy of the computation, except for those totally unused | |
1508 | // variables that opt2 removed. | |
1509 | varuse_collecting_visitor vut; | |
1510 | for (unsigned i=0; i<s.probes.size(); i++) | |
1511 | s.probes[i]->body->visit (& vut); // includes reachable functions too | |
1512 | ||
1513 | dead_assignment_remover dar (s, relaxed_p, vut); | |
1514 | // This instance may be reused for multiple probe/function body trims. | |
1515 | ||
1516 | for (unsigned i=0; i<s.probes.size(); i++) | |
1517 | s.probes[i]->body->visit (& dar); | |
1518 | for (unsigned i=0; i<s.functions.size(); i++) | |
1519 | s.functions[i]->body->visit (& dar); | |
1520 | // The rewrite operation is performed within the visitor. | |
1521 | } | |
1522 | ||
1523 | ||
1524 | // ------------------------------------------------------------------------ | |
1525 | ||
1526 | struct dead_stmtexpr_remover: public traversing_visitor | |
1527 | { | |
1528 | systemtap_session& session; | |
1529 | bool& relaxed_p; | |
1530 | statement** current_stmt; // pointer to current stmt* being iterated | |
1b07c728 | 1531 | set<vardecl*> focal_vars; // vars considered subject to side-effects |
cbfbbf69 FCE |
1532 | |
1533 | dead_stmtexpr_remover(systemtap_session& s, bool& r): | |
1534 | session(s), relaxed_p(r), current_stmt(0) {} | |
1535 | ||
1536 | void visit_block (block *s); | |
1537 | // XXX: and other places where stmt_expr's might be nested | |
1538 | ||
1539 | void visit_expr_statement (expr_statement *s); | |
1540 | }; | |
1541 | ||
1542 | ||
1543 | void | |
1544 | dead_stmtexpr_remover::visit_block (block *s) | |
1545 | { | |
1546 | for (unsigned i=0; i<s->statements.size(); i++) | |
1547 | { | |
1548 | statement** last_stmt = current_stmt; | |
1549 | current_stmt = & s->statements[i]; | |
1550 | s->statements[i]->visit (this); | |
1551 | current_stmt = last_stmt; | |
1552 | } | |
1553 | } | |
1554 | ||
1555 | ||
1556 | void | |
1557 | dead_stmtexpr_remover::visit_expr_statement (expr_statement *s) | |
1558 | { | |
1559 | // Run a varuse query against the operand expression. If it has no | |
1560 | // side-effects, replace the entire statement expression by a null | |
1561 | // statement. This replacement is done by overwriting the | |
1562 | // current_stmt pointer. | |
1563 | // | |
1564 | // Unlike many other visitors, we do *not* traverse this outermost | |
1565 | // one into the expression subtrees. There is no need - no | |
1566 | // expr_statement nodes will be found there. (Function bodies | |
1567 | // need to be visited explicitly by our caller.) | |
1568 | // | |
1569 | // NB. While we don't share nodes in the parse tree, let's not | |
1570 | // deallocate *s anyway, just in case... | |
1571 | ||
1572 | varuse_collecting_visitor vut; | |
1573 | s->value->visit (& vut); | |
1b07c728 FCE |
1574 | |
1575 | // Note that side-effect-freeness is not simply this test: | |
1576 | // | |
1577 | // (vut.written.empty() && !vut.embedded_seen) | |
1578 | // | |
1579 | // That's because the vut.written list may consist of local | |
1580 | // variables of called functions. Visible side-effects occur if | |
1581 | // *our* locals, or any *globals* are written-to. | |
1582 | ||
1583 | ||
1584 | set<vardecl*> intersection; | |
1585 | insert_iterator<set<vardecl*> > int_it (intersection, intersection.begin()); | |
1586 | set_intersection (vut.written.begin(), vut.written.end(), | |
1587 | focal_vars.begin(), focal_vars.end(), | |
1588 | int_it); | |
1589 | ||
1590 | if (intersection.empty() && ! vut.embedded_seen) | |
cbfbbf69 | 1591 | { |
b0ee93c4 | 1592 | if (session.verbose>2) |
cbfbbf69 FCE |
1593 | clog << "Eliding side-effect-free expression " |
1594 | << *s->tok << endl; | |
1595 | ||
1596 | null_statement* ns = new null_statement; | |
1597 | ns->tok = s->tok; | |
1598 | * current_stmt = ns; | |
1599 | // XXX: A null_statement carries more weight in the translator's | |
1600 | // output than a nonexistent statement. It might be nice to | |
1601 | // work a little harder and completely eliminate all traces of | |
1602 | // an elided statement. | |
1603 | ||
1604 | relaxed_p = false; | |
1605 | } | |
1b07c728 FCE |
1606 | else if(session.verbose>3) |
1607 | { | |
1608 | clog << "keeping expression " << *s->tok | |
1609 | << " because it writes: "; | |
1610 | for (std::set<vardecl*>::iterator k = intersection.begin(); | |
1611 | k != intersection.end(); k++) | |
1612 | clog << (*k)->name << " "; | |
1613 | clog << "and/or embedded: " << vut.embedded_seen << endl; | |
1614 | } | |
cbfbbf69 FCE |
1615 | } |
1616 | ||
1617 | ||
1618 | void semantic_pass_opt4 (systemtap_session& s, bool& relaxed_p) | |
1619 | { | |
1620 | // Finally, let's remove some statement-expressions that have no | |
1621 | // side-effect. These should be exactly those whose private varuse | |
1622 | // visitors come back with an empty "written" and "embedded" lists. | |
1623 | ||
1624 | dead_stmtexpr_remover duv (s, relaxed_p); | |
1625 | // This instance may be reused for multiple probe/function body trims. | |
1626 | ||
1627 | for (unsigned i=0; i<s.probes.size(); i++) | |
1b07c728 FCE |
1628 | { |
1629 | duv.focal_vars.clear (); | |
1630 | duv.focal_vars.insert (s.globals.begin(), | |
1631 | s.globals.end()); | |
1632 | duv.focal_vars.insert (s.probes[i]->locals.begin(), | |
1633 | s.probes[i]->locals.end()); | |
1634 | s.probes[i]->body->visit (& duv); | |
1635 | } | |
cbfbbf69 | 1636 | for (unsigned i=0; i<s.functions.size(); i++) |
1b07c728 FCE |
1637 | { |
1638 | duv.focal_vars.clear (); | |
1639 | duv.focal_vars.insert (s.functions[i]->locals.begin(), | |
1640 | s.functions[i]->locals.end()); | |
1641 | duv.focal_vars.insert (s.functions[i]->formal_args.begin(), | |
1642 | s.functions[i]->formal_args.end()); | |
1643 | duv.focal_vars.insert (s.globals.begin(), | |
1644 | s.globals.end()); | |
1645 | s.functions[i]->body->visit (& duv); | |
1646 | } | |
cbfbbf69 FCE |
1647 | } |
1648 | ||
88bbd60d DS |
1649 | struct duplicate_function_remover: public functioncall_traversing_visitor |
1650 | { | |
1651 | systemtap_session& s; | |
88bbd60d DS |
1652 | map<functiondecl*, functiondecl*>& duplicate_function_map; |
1653 | ||
c214bd6a | 1654 | duplicate_function_remover(systemtap_session& sess, |
88bbd60d | 1655 | map<functiondecl*, functiondecl*>&dfm): |
c214bd6a | 1656 | s(sess), duplicate_function_map(dfm) {}; |
88bbd60d DS |
1657 | |
1658 | void visit_functioncall (functioncall* e); | |
1659 | }; | |
1660 | ||
1661 | void | |
1662 | duplicate_function_remover::visit_functioncall (functioncall *e) | |
1663 | { | |
1664 | functioncall_traversing_visitor::visit_functioncall (e); | |
1665 | ||
1666 | // If the current function call reference points to a function that | |
1667 | // is a duplicate, replace it. | |
1668 | if (duplicate_function_map.count(e->referent) != 0) | |
1669 | { | |
1670 | if (s.verbose>2) | |
1671 | clog << "Changing " << e->referent->name | |
1672 | << " reference to " | |
1673 | << duplicate_function_map[e->referent]->name | |
1674 | << " reference\n"; | |
1675 | e->tok = duplicate_function_map[e->referent]->tok; | |
1676 | e->function = duplicate_function_map[e->referent]->name; | |
1677 | e->referent = duplicate_function_map[e->referent]; | |
88bbd60d DS |
1678 | } |
1679 | } | |
1680 | ||
1681 | static string | |
1682 | get_functionsig (functiondecl* f) | |
1683 | { | |
1684 | ostringstream s; | |
1685 | ||
1686 | // Get the "name:args body" of the function in s. We have to | |
1687 | // include the args since the function 'x1(a, b)' is different than | |
1688 | // the function 'x2(b, a)' even if the bodies of the two functions | |
1689 | // are exactly the same. | |
1690 | f->printsig(s); | |
1691 | f->body->print(s); | |
1692 | ||
1693 | // printsig puts f->name + ':' on the front. Remove this | |
1694 | // (otherwise, functions would never compare equal). | |
1695 | string str = s.str().erase(0, f->name.size() + 1); | |
1696 | ||
1697 | // Return the function signature. | |
1698 | return str; | |
1699 | } | |
1700 | ||
1701 | void semantic_pass_opt5 (systemtap_session& s, bool& relaxed_p) | |
1702 | { | |
1703 | // Walk through all the functions, looking for duplicates. | |
1704 | map<string, functiondecl*> functionsig_map; | |
1705 | map<functiondecl*, functiondecl*> duplicate_function_map; | |
c214bd6a | 1706 | for (unsigned i=0; i < s.functions.size(); /* see below */) |
88bbd60d DS |
1707 | { |
1708 | string functionsig = get_functionsig(s.functions[i]); | |
1709 | ||
1710 | if (functionsig_map.count(functionsig) == 0) | |
c214bd6a DS |
1711 | { |
1712 | // This function is unique. Remember it. | |
1713 | functionsig_map[functionsig] = s.functions[i]; | |
1714 | i++; | |
1715 | } | |
88bbd60d | 1716 | else |
c214bd6a DS |
1717 | { |
1718 | // This function is a duplicate. | |
1719 | duplicate_function_map[s.functions[i]] | |
88bbd60d | 1720 | = functionsig_map[functionsig]; |
c214bd6a DS |
1721 | |
1722 | // Remove the duplicate function (since we don't need it | |
1723 | // anymore). | |
1724 | s.functions.erase (s.functions.begin() + i); | |
1725 | ||
1726 | relaxed_p = false; | |
1727 | // NB: don't increment i | |
1728 | } | |
88bbd60d DS |
1729 | } |
1730 | ||
1731 | // If we have duplicate functions, traverse down the tree, replacing | |
1732 | // the appropriate function calls. | |
1733 | // duplicate_function_remover::visit_functioncall() handles the | |
c214bd6a | 1734 | // details of replacing the function calls. |
88bbd60d DS |
1735 | if (duplicate_function_map.size() != 0) |
1736 | { | |
c214bd6a | 1737 | duplicate_function_remover dfr (s, duplicate_function_map); |
88bbd60d DS |
1738 | |
1739 | for (unsigned i=0; i < s.probes.size(); i++) | |
1740 | s.probes[i]->body->visit(&dfr); | |
1741 | } | |
1742 | } | |
1743 | ||
cbfbbf69 FCE |
1744 | |
1745 | static int | |
c214bd6a | 1746 | semantic_pass_optimize1 (systemtap_session& s) |
cbfbbf69 FCE |
1747 | { |
1748 | // In this pass, we attempt to rewrite probe/function bodies to | |
1749 | // eliminate some blatantly unnecessary code. This is run before | |
1750 | // type inference, but after symbol resolution and derived_probe | |
1751 | // creation. We run an outer "relaxation" loop that repeats the | |
1752 | // optimizations until none of them find anything to remove. | |
1753 | ||
1754 | int rc = 0; | |
1755 | ||
1756 | bool relaxed_p = false; | |
1757 | while (! relaxed_p) | |
1758 | { | |
1759 | relaxed_p = true; // until proven otherwise | |
1760 | ||
1761 | semantic_pass_opt1 (s, relaxed_p); | |
1762 | semantic_pass_opt2 (s, relaxed_p); | |
1763 | semantic_pass_opt3 (s, relaxed_p); | |
1764 | semantic_pass_opt4 (s, relaxed_p); | |
c214bd6a DS |
1765 | } |
1766 | ||
1767 | if (s.probes.size() == 0) | |
1768 | { | |
1769 | cerr << "semantic error: no probes found." << endl; | |
1770 | rc = 1; | |
1771 | } | |
1772 | ||
1773 | return rc; | |
1774 | } | |
1775 | ||
1776 | ||
1777 | static int | |
1778 | semantic_pass_optimize2 (systemtap_session& s) | |
1779 | { | |
1780 | // This is run after type inference. We run an outer "relaxation" | |
1781 | // loop that repeats the optimizations until none of them find | |
1782 | // anything to remove. | |
1783 | ||
1784 | int rc = 0; | |
1785 | ||
1786 | bool relaxed_p = false; | |
1787 | while (! relaxed_p) | |
1788 | { | |
1789 | relaxed_p = true; // until proven otherwise | |
1790 | ||
88bbd60d | 1791 | semantic_pass_opt5 (s, relaxed_p); |
cbfbbf69 FCE |
1792 | } |
1793 | ||
1794 | if (s.probes.size() == 0) | |
1795 | { | |
1796 | cerr << "semantic error: no probes found." << endl; | |
1797 | rc = 1; | |
1798 | } | |
1799 | ||
1800 | return rc; | |
1801 | } | |
1802 | ||
1803 | ||
1804 | ||
2b066ec1 FCE |
1805 | // ------------------------------------------------------------------------ |
1806 | // type resolution | |
1807 | ||
1808 | ||
1809 | static int | |
1810 | semantic_pass_types (systemtap_session& s) | |
1811 | { | |
1812 | int rc = 0; | |
1813 | ||
1814 | // next pass: type inference | |
1815 | unsigned iterations = 0; | |
1816 | typeresolution_info ti (s); | |
1817 | ||
1818 | ti.assert_resolvability = false; | |
1819 | // XXX: maybe convert to exception-based error signalling | |
1820 | while (1) | |
1821 | { | |
1822 | iterations ++; | |
1823 | // cerr << "Type resolution, iteration " << iterations << endl; | |
1824 | ti.num_newly_resolved = 0; | |
1825 | ti.num_still_unresolved = 0; | |
1826 | ||
1827 | for (unsigned j=0; j<s.functions.size(); j++) | |
1828 | { | |
1829 | functiondecl* fn = s.functions[j]; | |
bdef2583 | 1830 | ti.current_probe = 0; |
8a43522c GH |
1831 | ti.current_function = fn; |
1832 | ti.t = pe_unknown; | |
1833 | fn->body->visit (& ti); | |
1834 | // NB: we don't have to assert a known type for | |
1835 | // functions here, to permit a "void" function. | |
1836 | // The translator phase will omit the "retvalue". | |
1837 | // | |
1838 | // if (fn->type == pe_unknown) | |
1839 | // ti.unresolved (fn->tok); | |
2b066ec1 FCE |
1840 | } |
1841 | ||
1842 | for (unsigned j=0; j<s.probes.size(); j++) | |
1843 | { | |
1844 | derived_probe* pn = s.probes[j]; | |
1845 | ti.current_function = 0; | |
bdef2583 | 1846 | ti.current_probe = pn; |
2b066ec1 FCE |
1847 | ti.t = pe_unknown; |
1848 | pn->body->visit (& ti); | |
1849 | } | |
1850 | ||
1851 | for (unsigned j=0; j<s.globals.size(); j++) | |
1852 | { | |
1853 | vardecl* gd = s.globals[j]; | |
1854 | if (gd->type == pe_unknown) | |
1855 | ti.unresolved (gd->tok); | |
1856 | } | |
1857 | ||
1858 | if (ti.num_newly_resolved == 0) // converged | |
78f6bba6 FCE |
1859 | { |
1860 | if (ti.num_still_unresolved == 0) | |
1861 | break; // successfully | |
1862 | else if (! ti.assert_resolvability) | |
1863 | ti.assert_resolvability = true; // last pass, with error msgs | |
1864 | else | |
1865 | { // unsuccessful conclusion | |
1866 | rc ++; | |
1867 | break; | |
1868 | } | |
1869 | } | |
2b066ec1 FCE |
1870 | } |
1871 | ||
7e41d3dc | 1872 | return rc + s.num_errors(); |
2b066ec1 FCE |
1873 | } |
1874 | ||
1875 | ||
bdef2583 FCE |
1876 | |
1877 | typeresolution_info::typeresolution_info (systemtap_session& s): | |
1878 | session(s), current_function(0), current_probe(0) | |
1879 | { | |
1880 | } | |
1881 | ||
1882 | ||
2b066ec1 FCE |
1883 | void |
1884 | typeresolution_info::visit_literal_number (literal_number* e) | |
1885 | { | |
1886 | assert (e->type == pe_long); | |
1887 | if ((t == e->type) || (t == pe_unknown)) | |
1888 | return; | |
1889 | ||
1890 | mismatch (e->tok, e->type, t); | |
1891 | } | |
1892 | ||
1893 | ||
1894 | void | |
1895 | typeresolution_info::visit_literal_string (literal_string* e) | |
1896 | { | |
1897 | assert (e->type == pe_string); | |
1898 | if ((t == e->type) || (t == pe_unknown)) | |
1899 | return; | |
1900 | ||
1901 | mismatch (e->tok, e->type, t); | |
1902 | } | |
1903 | ||
1904 | ||
1905 | void | |
1906 | typeresolution_info::visit_logical_or_expr (logical_or_expr *e) | |
1907 | { | |
1908 | visit_binary_expression (e); | |
1909 | } | |
1910 | ||
1911 | ||
1912 | void | |
1913 | typeresolution_info::visit_logical_and_expr (logical_and_expr *e) | |
1914 | { | |
1915 | visit_binary_expression (e); | |
1916 | } | |
1917 | ||
1918 | ||
1919 | void | |
1920 | typeresolution_info::visit_comparison (comparison *e) | |
1921 | { | |
d5d7c2cc | 1922 | // NB: result of any comparison is an integer! |
553d27a5 FCE |
1923 | if (t == pe_stats || t == pe_string) |
1924 | invalid (e->tok, t); | |
1925 | ||
1926 | t = (e->right->type != pe_unknown) ? e->right->type : pe_unknown; | |
1927 | e->left->visit (this); | |
1928 | t = (e->left->type != pe_unknown) ? e->left->type : pe_unknown; | |
1929 | e->right->visit (this); | |
1930 | ||
1931 | if (e->left->type != pe_unknown && | |
1932 | e->right->type != pe_unknown && | |
1933 | e->left->type != e->right->type) | |
1934 | mismatch (e->tok, e->left->type, e->right->type); | |
1935 | ||
1936 | if (e->type == pe_unknown) | |
1937 | { | |
1938 | e->type = pe_long; | |
1939 | resolved (e->tok, e->type); | |
1940 | } | |
2b066ec1 FCE |
1941 | } |
1942 | ||
1943 | ||
1944 | void | |
1945 | typeresolution_info::visit_concatenation (concatenation *e) | |
1946 | { | |
553d27a5 FCE |
1947 | if (t != pe_unknown && t != pe_string) |
1948 | invalid (e->tok, t); | |
1949 | ||
1950 | t = pe_string; | |
1951 | e->left->visit (this); | |
1952 | t = pe_string; | |
1953 | e->right->visit (this); | |
1954 | ||
1955 | if (e->type == pe_unknown) | |
1956 | { | |
1957 | e->type = pe_string; | |
1958 | resolved (e->tok, e->type); | |
1959 | } | |
2b066ec1 FCE |
1960 | } |
1961 | ||
1962 | ||
2b066ec1 FCE |
1963 | void |
1964 | typeresolution_info::visit_assignment (assignment *e) | |
1965 | { | |
553d27a5 FCE |
1966 | if (t == pe_stats) |
1967 | invalid (e->tok, t); | |
2b066ec1 | 1968 | |
2b066ec1 FCE |
1969 | if (e->op == "<<<") // stats aggregation |
1970 | { | |
553d27a5 FCE |
1971 | if (t == pe_string) |
1972 | invalid (e->tok, t); | |
1973 | ||
2b066ec1 FCE |
1974 | t = pe_stats; |
1975 | e->left->visit (this); | |
1976 | t = pe_long; | |
1977 | e->right->visit (this); | |
57b73400 GH |
1978 | if (e->type == pe_unknown || |
1979 | e->type == pe_stats) | |
2b066ec1 FCE |
1980 | { |
1981 | e->type = pe_long; | |
1982 | resolved (e->tok, e->type); | |
1983 | } | |
1984 | } | |
57b73400 GH |
1985 | |
1986 | else if (e->left->type == pe_stats) | |
1987 | invalid (e->left->tok, e->left->type); | |
1988 | ||
1989 | else if (e->right->type == pe_stats) | |
1990 | invalid (e->right->tok, e->right->type); | |
1991 | ||
553d27a5 | 1992 | else if (e->op == "+=" || // numeric only |
d5d7c2cc FCE |
1993 | e->op == "-=" || |
1994 | e->op == "*=" || | |
1995 | e->op == "/=" || | |
1996 | e->op == "%=" || | |
1997 | e->op == "&=" || | |
1998 | e->op == "^=" || | |
1999 | e->op == "|=" || | |
2000 | e->op == "<<=" || | |
2001 | e->op == ">>=" || | |
553d27a5 | 2002 | false) |
2b066ec1 | 2003 | { |
553d27a5 | 2004 | visit_binary_expression (e); |
2b066ec1 | 2005 | } |
d5d7c2cc FCE |
2006 | else if (e->op == ".=" || // string only |
2007 | false) | |
2008 | { | |
2009 | if (t == pe_long || t == pe_stats) | |
2010 | invalid (e->tok, t); | |
2011 | ||
2012 | t = pe_string; | |
2013 | e->left->visit (this); | |
2014 | t = pe_string; | |
2015 | e->right->visit (this); | |
2016 | if (e->type == pe_unknown) | |
2017 | { | |
2018 | e->type = pe_string; | |
2019 | resolved (e->tok, e->type); | |
2020 | } | |
2021 | } | |
2022 | else if (e->op == "=") // overloaded = for string & numeric operands | |
2b066ec1 | 2023 | { |
553d27a5 | 2024 | // logic similar to ternary_expression |
2b066ec1 | 2025 | exp_type sub_type = t; |
553d27a5 FCE |
2026 | |
2027 | // Infer types across the l/r values | |
2028 | if (sub_type == pe_unknown && e->type != pe_unknown) | |
2b066ec1 | 2029 | sub_type = e->type; |
553d27a5 FCE |
2030 | |
2031 | t = (sub_type != pe_unknown) ? sub_type : | |
2032 | (e->right->type != pe_unknown) ? e->right->type : | |
2033 | pe_unknown; | |
2b066ec1 | 2034 | e->left->visit (this); |
553d27a5 FCE |
2035 | t = (sub_type != pe_unknown) ? sub_type : |
2036 | (e->left->type != pe_unknown) ? e->left->type : | |
2037 | pe_unknown; | |
2b066ec1 | 2038 | e->right->visit (this); |
553d27a5 FCE |
2039 | |
2040 | if ((sub_type != pe_unknown) && (e->type == pe_unknown)) | |
2b066ec1 FCE |
2041 | { |
2042 | e->type = sub_type; | |
2043 | resolved (e->tok, e->type); | |
2044 | } | |
553d27a5 | 2045 | if ((sub_type == pe_unknown) && (e->left->type != pe_unknown)) |
2b066ec1 FCE |
2046 | { |
2047 | e->type = e->left->type; | |
2048 | resolved (e->tok, e->type); | |
2049 | } | |
553d27a5 FCE |
2050 | |
2051 | if (e->left->type != pe_unknown && | |
2052 | e->right->type != pe_unknown && | |
2053 | e->left->type != e->right->type) | |
2054 | mismatch (e->tok, e->left->type, e->right->type); | |
d02548c0 | 2055 | |
553d27a5 | 2056 | } |
d5d7c2cc FCE |
2057 | else |
2058 | throw semantic_error ("unsupported assignment operator " + e->op); | |
553d27a5 FCE |
2059 | } |
2060 | ||
2061 | ||
2062 | void | |
2063 | typeresolution_info::visit_binary_expression (binary_expression* e) | |
2064 | { | |
2065 | if (t == pe_stats || t == pe_string) | |
2066 | invalid (e->tok, t); | |
2067 | ||
2068 | t = pe_long; | |
2069 | e->left->visit (this); | |
2070 | t = pe_long; | |
2071 | e->right->visit (this); | |
2072 | ||
2073 | if (e->left->type != pe_unknown && | |
2074 | e->right->type != pe_unknown && | |
2075 | e->left->type != e->right->type) | |
2076 | mismatch (e->tok, e->left->type, e->right->type); | |
2077 | ||
2078 | if (e->type == pe_unknown) | |
2079 | { | |
2080 | e->type = pe_long; | |
2081 | resolved (e->tok, e->type); | |
2b066ec1 FCE |
2082 | } |
2083 | } | |
2084 | ||
2085 | ||
2086 | void | |
2087 | typeresolution_info::visit_pre_crement (pre_crement *e) | |
2088 | { | |
2089 | visit_unary_expression (e); | |
2090 | } | |
2091 | ||
2092 | ||
2093 | void | |
2094 | typeresolution_info::visit_post_crement (post_crement *e) | |
2095 | { | |
2096 | visit_unary_expression (e); | |
2097 | } | |
2098 | ||
2099 | ||
2100 | void | |
2101 | typeresolution_info::visit_unary_expression (unary_expression* e) | |
2102 | { | |
553d27a5 FCE |
2103 | if (t == pe_stats || t == pe_string) |
2104 | invalid (e->tok, t); | |
2105 | ||
2b066ec1 FCE |
2106 | t = pe_long; |
2107 | e->operand->visit (this); | |
2108 | ||
553d27a5 | 2109 | if (e->type == pe_unknown) |
2b066ec1 FCE |
2110 | { |
2111 | e->type = pe_long; | |
2112 | resolved (e->tok, e->type); | |
2113 | } | |
2114 | } | |
2115 | ||
2116 | ||
2b066ec1 FCE |
2117 | void |
2118 | typeresolution_info::visit_ternary_expression (ternary_expression* e) | |
2119 | { | |
2120 | exp_type sub_type = t; | |
2121 | ||
2122 | t = pe_long; | |
2123 | e->cond->visit (this); | |
2124 | ||
553d27a5 | 2125 | // Infer types across the true/false arms of the ternary expression. |
2b066ec1 FCE |
2126 | |
2127 | if (sub_type == pe_unknown && e->type != pe_unknown) | |
2128 | sub_type = e->type; | |
2129 | t = sub_type; | |
2130 | e->truevalue->visit (this); | |
2131 | t = sub_type; | |
2132 | e->falsevalue->visit (this); | |
2133 | ||
2134 | if ((sub_type == pe_unknown) && (e->type != pe_unknown)) | |
2135 | ; // already resolved | |
2136 | else if ((sub_type != pe_unknown) && (e->type == pe_unknown)) | |
2137 | { | |
2138 | e->type = sub_type; | |
2139 | resolved (e->tok, e->type); | |
2140 | } | |
2141 | else if ((sub_type == pe_unknown) && (e->truevalue->type != pe_unknown)) | |
2142 | { | |
2143 | e->type = e->truevalue->type; | |
2144 | resolved (e->tok, e->type); | |
2145 | } | |
2146 | else if ((sub_type == pe_unknown) && (e->falsevalue->type != pe_unknown)) | |
2147 | { | |
2148 | e->type = e->falsevalue->type; | |
2149 | resolved (e->tok, e->type); | |
2150 | } | |
2151 | else if (e->type != sub_type) | |
2152 | mismatch (e->tok, sub_type, e->type); | |
2153 | } | |
2154 | ||
2155 | ||
2156 | template <class Referrer, class Referent> | |
2157 | void resolve_2types (Referrer* referrer, Referent* referent, | |
31966088 | 2158 | typeresolution_info* r, exp_type t, bool accept_unknown = false) |
2b066ec1 FCE |
2159 | { |
2160 | exp_type& re_type = referrer->type; | |
2161 | const token* re_tok = referrer->tok; | |
2162 | exp_type& te_type = referent->type; | |
2163 | const token* te_tok = referent->tok; | |
2164 | ||
2165 | if (t != pe_unknown && re_type == t && re_type == te_type) | |
2166 | ; // do nothing: all three e->types in agreement | |
2167 | else if (t == pe_unknown && re_type != pe_unknown && re_type == te_type) | |
2168 | ; // do nothing: two known e->types in agreement | |
2169 | else if (re_type != pe_unknown && te_type != pe_unknown && re_type != te_type) | |
2170 | r->mismatch (re_tok, re_type, te_type); | |
2171 | else if (re_type != pe_unknown && t != pe_unknown && re_type != t) | |
2172 | r->mismatch (re_tok, re_type, t); | |
2173 | else if (te_type != pe_unknown && t != pe_unknown && te_type != t) | |
2174 | r->mismatch (te_tok, te_type, t); | |
2175 | else if (re_type == pe_unknown && t != pe_unknown) | |
2176 | { | |
2177 | // propagate from upstream | |
2178 | re_type = t; | |
2179 | r->resolved (re_tok, re_type); | |
2180 | // catch re_type/te_type mismatch later | |
2181 | } | |
2182 | else if (re_type == pe_unknown && te_type != pe_unknown) | |
2183 | { | |
2184 | // propagate from referent | |
2185 | re_type = te_type; | |
2186 | r->resolved (re_tok, re_type); | |
2187 | // catch re_type/t mismatch later | |
2188 | } | |
2189 | else if (re_type != pe_unknown && te_type == pe_unknown) | |
2190 | { | |
2191 | // propagate to referent | |
2192 | te_type = re_type; | |
2193 | r->resolved (te_tok, te_type); | |
2194 | // catch re_type/t mismatch later | |
2195 | } | |
31966088 | 2196 | else if (! accept_unknown) |
2b066ec1 FCE |
2197 | r->unresolved (re_tok); |
2198 | } | |
2199 | ||
2200 | ||
2201 | void | |
2202 | typeresolution_info::visit_symbol (symbol* e) | |
2203 | { | |
2204 | assert (e->referent != 0); | |
d02548c0 | 2205 | resolve_2types (e, e->referent, this, t); |
2b066ec1 FCE |
2206 | } |
2207 | ||
2208 | ||
d7f3e0c5 GH |
2209 | void |
2210 | typeresolution_info::visit_target_symbol (target_symbol* e) | |
2211 | { | |
cbfbbf69 FCE |
2212 | // This occurs only if a target symbol was not resolved over in |
2213 | // tapset.cxx land, that error was properly suppressed, and the | |
2214 | // later unused-expression-elimination pass didn't get rid of it | |
2215 | // either. So we have a target symbol that is believed to be of | |
2216 | // genuine use, yet unresolved by the provider. | |
9b48ce88 FCE |
2217 | if (e->saved_conversion_error) |
2218 | throw (* (e->saved_conversion_error)); | |
2219 | else | |
2220 | throw semantic_error("unresolved target-symbol expression", e->tok); | |
d7f3e0c5 GH |
2221 | } |
2222 | ||
2223 | ||
2b066ec1 FCE |
2224 | void |
2225 | typeresolution_info::visit_arrayindex (arrayindex* e) | |
2226 | { | |
2b066ec1 | 2227 | |
d02548c0 GH |
2228 | symbol *array = NULL; |
2229 | hist_op *hist = NULL; | |
2230 | classify_indexable(e->base, array, hist); | |
2231 | ||
2232 | // Every hist_op has type [int]:int, that is to say, every hist_op | |
2233 | // is a pseudo-one-dimensional integer array type indexed by | |
2234 | // integers (bucket numbers). | |
2235 | ||
2236 | if (hist) | |
2237 | { | |
2238 | if (e->indexes.size() != 1) | |
2239 | unresolved (e->tok); | |
2240 | t = pe_long; | |
2241 | e->indexes[0]->visit (this); | |
2242 | if (e->indexes[0]->type != pe_long) | |
2243 | unresolved (e->tok); | |
a4636912 | 2244 | hist->visit (this); |
1bbeef03 GH |
2245 | if (e->type != pe_long) |
2246 | { | |
2247 | e->type = pe_long; | |
2248 | resolved (e->tok, pe_long); | |
2249 | } | |
d02548c0 GH |
2250 | return; |
2251 | } | |
2252 | ||
2253 | // Now we are left with "normal" map inference and index checking. | |
2254 | ||
2255 | assert (array); | |
2256 | assert (array->referent != 0); | |
2257 | resolve_2types (e, array->referent, this, t); | |
2b066ec1 FCE |
2258 | |
2259 | // now resolve the array indexes | |
69c68955 FCE |
2260 | |
2261 | // if (e->referent->index_types.size() == 0) | |
2262 | // // redesignate referent as array | |
2263 | // e->referent->set_arity (e->indexes.size ()); | |
2b066ec1 | 2264 | |
d02548c0 | 2265 | if (e->indexes.size() != array->referent->index_types.size()) |
2b066ec1 FCE |
2266 | unresolved (e->tok); // symbol resolution should prevent this |
2267 | else for (unsigned i=0; i<e->indexes.size(); i++) | |
2268 | { | |
2269 | expression* ee = e->indexes[i]; | |
d02548c0 | 2270 | exp_type& ft = array->referent->index_types [i]; |
2b066ec1 FCE |
2271 | t = ft; |
2272 | ee->visit (this); | |
2273 | exp_type at = ee->type; | |
2274 | ||
2275 | if ((at == pe_string || at == pe_long) && ft == pe_unknown) | |
2276 | { | |
2277 | // propagate to formal type | |
2278 | ft = at; | |
d02548c0 | 2279 | resolved (array->referent->tok, ft); |
2b066ec1 FCE |
2280 | // uses array decl as there is no token for "formal type" |
2281 | } | |
2282 | if (at == pe_stats) | |
2283 | invalid (ee->tok, at); | |
2284 | if (ft == pe_stats) | |
2285 | invalid (ee->tok, ft); | |
2286 | if (at != pe_unknown && ft != pe_unknown && ft != at) | |
2287 | mismatch (e->tok, at, ft); | |
2288 | if (at == pe_unknown) | |
313b2f74 | 2289 | unresolved (ee->tok); |
2b066ec1 FCE |
2290 | } |
2291 | } | |
2292 | ||
2293 | ||
2294 | void | |
2295 | typeresolution_info::visit_functioncall (functioncall* e) | |
2296 | { | |
2297 | assert (e->referent != 0); | |
2298 | ||
31966088 | 2299 | resolve_2types (e, e->referent, this, t, true); // accept unknown type |
2b066ec1 FCE |
2300 | |
2301 | if (e->type == pe_stats) | |
2302 | invalid (e->tok, e->type); | |
2303 | ||
2b066ec1 FCE |
2304 | // now resolve the function parameters |
2305 | if (e->args.size() != e->referent->formal_args.size()) | |
2306 | unresolved (e->tok); // symbol resolution should prevent this | |
2307 | else for (unsigned i=0; i<e->args.size(); i++) | |
2308 | { | |
2309 | expression* ee = e->args[i]; | |
2310 | exp_type& ft = e->referent->formal_args[i]->type; | |
2311 | const token* fe_tok = e->referent->formal_args[i]->tok; | |
2312 | t = ft; | |
2313 | ee->visit (this); | |
2314 | exp_type at = ee->type; | |
2315 | ||
2316 | if (((at == pe_string) || (at == pe_long)) && ft == pe_unknown) | |
2317 | { | |
2318 | // propagate to formal arg | |
2319 | ft = at; | |
2320 | resolved (e->referent->formal_args[i]->tok, ft); | |
2321 | } | |
2322 | if (at == pe_stats) | |
2323 | invalid (e->tok, at); | |
2324 | if (ft == pe_stats) | |
2325 | invalid (fe_tok, ft); | |
2326 | if (at != pe_unknown && ft != pe_unknown && ft != at) | |
2327 | mismatch (e->tok, at, ft); | |
2328 | if (at == pe_unknown) | |
2329 | unresolved (e->tok); | |
2330 | } | |
2331 | } | |
2332 | ||
2333 | ||
2334 | void | |
2335 | typeresolution_info::visit_block (block* e) | |
2336 | { | |
2337 | for (unsigned i=0; i<e->statements.size(); i++) | |
2338 | { | |
2339 | try | |
2340 | { | |
2341 | t = pe_unknown; | |
2342 | e->statements[i]->visit (this); | |
2343 | } | |
2344 | catch (const semantic_error& e) | |
2345 | { | |
2346 | session.print_error (e); | |
2347 | } | |
2348 | } | |
2349 | } | |
2350 | ||
2351 | ||
54dfabe9 | 2352 | void |
78f6bba6 | 2353 | typeresolution_info::visit_embeddedcode (embeddedcode*) |
54dfabe9 FCE |
2354 | { |
2355 | } | |
2356 | ||
2357 | ||
2b066ec1 FCE |
2358 | void |
2359 | typeresolution_info::visit_if_statement (if_statement* e) | |
2360 | { | |
2361 | t = pe_long; | |
2362 | e->condition->visit (this); | |
2363 | ||
2364 | t = pe_unknown; | |
2365 | e->thenblock->visit (this); | |
2366 | ||
2367 | if (e->elseblock) | |
2368 | { | |
2369 | t = pe_unknown; | |
2370 | e->elseblock->visit (this); | |
2371 | } | |
2372 | } | |
2373 | ||
2374 | ||
2375 | void | |
2376 | typeresolution_info::visit_for_loop (for_loop* e) | |
2377 | { | |
2378 | t = pe_unknown; | |
cbfbbf69 | 2379 | if (e->init) e->init->visit (this); |
2b066ec1 FCE |
2380 | t = pe_long; |
2381 | e->cond->visit (this); | |
2382 | t = pe_unknown; | |
cbfbbf69 | 2383 | if (e->incr) e->incr->visit (this); |
2b066ec1 FCE |
2384 | t = pe_unknown; |
2385 | e->block->visit (this); | |
2386 | } | |
2387 | ||
2388 | ||
69c68955 FCE |
2389 | void |
2390 | typeresolution_info::visit_foreach_loop (foreach_loop* e) | |
2391 | { | |
2392 | // See also visit_arrayindex. | |
2393 | // This is different in that, being a statement, we can't assign | |
2394 | // a type to the outer array, only propagate to/from the indexes | |
2395 | ||
2396 | // if (e->referent->index_types.size() == 0) | |
2397 | // // redesignate referent as array | |
2398 | // e->referent->set_arity (e->indexes.size ()); | |
2399 | ||
d02548c0 GH |
2400 | symbol *array = NULL; |
2401 | hist_op *hist = NULL; | |
2402 | classify_indexable(e->base, array, hist); | |
69c68955 | 2403 | |
d02548c0 GH |
2404 | if (hist) |
2405 | { | |
2406 | if (e->indexes.size() != 1) | |
2407 | unresolved (e->tok); | |
2408 | t = pe_long; | |
2409 | e->indexes[0]->visit (this); | |
2410 | if (e->indexes[0]->type != pe_long) | |
2411 | unresolved (e->tok); | |
a4636912 | 2412 | hist->visit (this); |
d02548c0 GH |
2413 | } |
2414 | else | |
2415 | { | |
2416 | assert (array); | |
2417 | if (e->indexes.size() != array->referent->index_types.size()) | |
2418 | unresolved (e->tok); // symbol resolution should prevent this | |
2419 | else for (unsigned i=0; i<e->indexes.size(); i++) | |
2420 | { | |
2421 | expression* ee = e->indexes[i]; | |
2422 | exp_type& ft = array->referent->index_types [i]; | |
2423 | t = ft; | |
2424 | ee->visit (this); | |
2425 | exp_type at = ee->type; | |
2426 | ||
2427 | if ((at == pe_string || at == pe_long) && ft == pe_unknown) | |
2428 | { | |
2429 | // propagate to formal type | |
2430 | ft = at; | |
2431 | resolved (array->referent->tok, ft); | |
2432 | // uses array decl as there is no token for "formal type" | |
2433 | } | |
2434 | if (at == pe_stats) | |
2435 | invalid (ee->tok, at); | |
2436 | if (ft == pe_stats) | |
2437 | invalid (ee->tok, ft); | |
2438 | if (at != pe_unknown && ft != pe_unknown && ft != at) | |
2439 | mismatch (e->tok, at, ft); | |
2440 | if (at == pe_unknown) | |
2441 | unresolved (ee->tok); | |
2442 | } | |
69c68955 FCE |
2443 | } |
2444 | ||
27f21e8c DS |
2445 | if (e->limit) |
2446 | { | |
2447 | t = pe_long; | |
2448 | e->limit->visit (this); | |
2449 | } | |
2450 | ||
69c68955 FCE |
2451 | t = pe_unknown; |
2452 | e->block->visit (this); | |
2453 | } | |
2454 | ||
2455 | ||
2b066ec1 | 2456 | void |
78f6bba6 | 2457 | typeresolution_info::visit_null_statement (null_statement*) |
2b066ec1 FCE |
2458 | { |
2459 | } | |
2460 | ||
2461 | ||
2462 | void | |
2463 | typeresolution_info::visit_expr_statement (expr_statement* e) | |
2464 | { | |
2465 | t = pe_unknown; | |
2466 | e->value->visit (this); | |
2467 | } | |
2468 | ||
2469 | ||
d98d459c GH |
2470 | struct delete_statement_typeresolution_info: |
2471 | public throwing_visitor | |
2472 | { | |
2473 | typeresolution_info *parent; | |
2474 | delete_statement_typeresolution_info (typeresolution_info *p): | |
2475 | throwing_visitor ("invalid operand of delete expression"), | |
2476 | parent (p) | |
2477 | {} | |
2478 | ||
2479 | void visit_arrayindex (arrayindex* e) | |
2480 | { | |
2481 | parent->visit_arrayindex (e); | |
2482 | } | |
2483 | ||
2484 | void visit_symbol (symbol* e) | |
2485 | { | |
2486 | exp_type ignored = pe_unknown; | |
2487 | assert (e->referent != 0); | |
2488 | resolve_2types (e, e->referent, parent, ignored); | |
2489 | } | |
2490 | }; | |
2491 | ||
2492 | ||
2b066ec1 FCE |
2493 | void |
2494 | typeresolution_info::visit_delete_statement (delete_statement* e) | |
2495 | { | |
d98d459c GH |
2496 | delete_statement_typeresolution_info di (this); |
2497 | t = pe_unknown; | |
2498 | e->value->visit (&di); | |
2b066ec1 FCE |
2499 | } |
2500 | ||
2501 | ||
f3c26ea5 | 2502 | void |
78f6bba6 | 2503 | typeresolution_info::visit_next_statement (next_statement*) |
f3c26ea5 FCE |
2504 | { |
2505 | } | |
2506 | ||
2507 | ||
2508 | void | |
78f6bba6 | 2509 | typeresolution_info::visit_break_statement (break_statement*) |
f3c26ea5 FCE |
2510 | { |
2511 | } | |
2512 | ||
2513 | ||
2514 | void | |
78f6bba6 | 2515 | typeresolution_info::visit_continue_statement (continue_statement*) |
f3c26ea5 FCE |
2516 | { |
2517 | } | |
2518 | ||
2519 | ||
2b066ec1 FCE |
2520 | void |
2521 | typeresolution_info::visit_array_in (array_in* e) | |
2522 | { | |
ce10591c FCE |
2523 | // all unary operators only work on numerics |
2524 | exp_type t1 = t; | |
2525 | t = pe_unknown; // array value can be anything | |
2526 | e->operand->visit (this); | |
2527 | ||
2528 | if (t1 == pe_unknown && e->type != pe_unknown) | |
2529 | ; // already resolved | |
2530 | else if (t1 == pe_string || t1 == pe_stats) | |
2531 | mismatch (e->tok, t1, pe_long); | |
2532 | else if (e->type == pe_unknown) | |
2533 | { | |
2534 | e->type = pe_long; | |
2535 | resolved (e->tok, e->type); | |
2536 | } | |
2b066ec1 FCE |
2537 | } |
2538 | ||
2539 | ||
2540 | void | |
2541 | typeresolution_info::visit_return_statement (return_statement* e) | |
2542 | { | |
2543 | // This is like symbol, where the referent is | |
2544 | // the return value of the function. | |
2545 | ||
22f46231 | 2546 | // translation pass will print error |
2b066ec1 | 2547 | if (current_function == 0) |
22f46231 | 2548 | return; |
2b066ec1 FCE |
2549 | |
2550 | exp_type& e_type = current_function->type; | |
2551 | t = current_function->type; | |
2552 | e->value->visit (this); | |
2553 | ||
2554 | if (e_type != pe_unknown && e->value->type != pe_unknown | |
2555 | && e_type != e->value->type) | |
2556 | mismatch (current_function->tok, e_type, e->value->type); | |
2557 | if (e_type == pe_unknown && | |
2558 | (e->value->type == pe_long || e->value->type == pe_string)) | |
2559 | { | |
2560 | // propagate non-statistics from value | |
2561 | e_type = e->value->type; | |
2562 | resolved (current_function->tok, e->value->type); | |
2563 | } | |
2564 | if (e->value->type == pe_stats) | |
2565 | invalid (e->value->tok, e->value->type); | |
2566 | } | |
2567 | ||
d02548c0 GH |
2568 | void |
2569 | typeresolution_info::visit_print_format (print_format* e) | |
2570 | { | |
2571 | size_t unresolved_args = 0; | |
2572 | ||
1bbeef03 GH |
2573 | if (e->hist) |
2574 | { | |
2575 | e->hist->visit(this); | |
2576 | } | |
2577 | ||
2578 | else if (e->print_with_format) | |
d02548c0 GH |
2579 | { |
2580 | // If there's a format string, we can do both inference *and* | |
2581 | // checking. | |
2582 | ||
2583 | // First we extract the subsequence of formatting components | |
2584 | // which are conversions (not just literal string components) | |
2585 | ||
2586 | std::vector<print_format::format_component> components; | |
2587 | for (size_t i = 0; i < e->components.size(); ++i) | |
2588 | { | |
2589 | if (e->components[i].type == print_format::conv_unspecified) | |
2590 | throw semantic_error ("Unspecified conversion in print operator format string", | |
2591 | e->tok); | |
dc0b623a MH |
2592 | else if (e->components[i].type == print_format::conv_literal |
2593 | || e->components[i].type == print_format::conv_size) | |
d02548c0 GH |
2594 | continue; |
2595 | components.push_back(e->components[i]); | |
2596 | } | |
2597 | ||
2598 | // Then we check that the number of conversions and the number | |
2599 | // of args agree. | |
2600 | ||
2601 | if (components.size() != e->args.size()) | |
2602 | throw semantic_error ("Wrong number of args to formatted print operator", | |
2603 | e->tok); | |
2604 | ||
2605 | // Then we check that the types of the conversions match the types | |
2606 | // of the args. | |
2607 | for (size_t i = 0; i < components.size(); ++i) | |
2608 | { | |
2609 | exp_type wanted = pe_unknown; | |
2610 | ||
2611 | switch (components[i].type) | |
2612 | { | |
d02548c0 GH |
2613 | case print_format::conv_unspecified: |
2614 | case print_format::conv_literal: | |
dc0b623a | 2615 | case print_format::conv_size: |
d02548c0 GH |
2616 | assert (false); |
2617 | break; | |
2618 | ||
2619 | case print_format::conv_signed_decimal: | |
2620 | case print_format::conv_unsigned_decimal: | |
2621 | case print_format::conv_unsigned_octal: | |
58cf0567 | 2622 | case print_format::conv_unsigned_ptr: |
d02548c0 GH |
2623 | case print_format::conv_unsigned_uppercase_hex: |
2624 | case print_format::conv_unsigned_lowercase_hex: | |
dc0b623a | 2625 | case print_format::conv_binary: |
d02548c0 GH |
2626 | wanted = pe_long; |
2627 | break; | |
2628 | ||
2629 | case print_format::conv_string: | |
2630 | wanted = pe_string; | |
2631 | break; | |
2632 | } | |
2633 | ||
2634 | assert (wanted != pe_unknown); | |
2635 | ||
2636 | t = wanted; | |
2637 | e->args[i]->visit (this); | |
2638 | ||
2639 | if (e->args[i]->type == pe_unknown) | |
2640 | { | |
2641 | e->args[i]->type = wanted; | |
2642 | resolved (e->args[i]->tok, wanted); | |
2643 | } | |
2644 | else if (e->args[i]->type != wanted) | |
2645 | { | |
2646 | mismatch (e->args[i]->tok, e->args[i]->type, wanted); | |
2647 | } | |
2648 | } | |
2649 | } | |
2650 | else | |
2651 | { | |
2652 | // Without a format string, the best we can do is require that | |
2653 | // each argument resolve to a concrete type. | |
2654 | for (size_t i = 0; i < e->args.size(); ++i) | |
2655 | { | |
2656 | t = pe_unknown; | |
2657 | e->args[i]->visit (this); | |
2658 | if (e->args[i]->type == pe_unknown) | |
2659 | { | |
2660 | unresolved (e->args[i]->tok); | |
2661 | ++unresolved_args; | |
2662 | } | |
2663 | } | |
2664 | } | |
2665 | ||
2666 | if (unresolved_args == 0) | |
2667 | { | |
2668 | if (e->type == pe_unknown) | |
2669 | { | |
2670 | if (e->print_to_stream) | |
2671 | e->type = pe_long; | |
2672 | else | |
2673 | e->type = pe_string; | |
2674 | resolved (e->tok, e->type); | |
2675 | } | |
2676 | } | |
2677 | else | |
2678 | { | |
2679 | e->type = pe_unknown; | |
2680 | unresolved (e->tok); | |
2681 | } | |
2682 | } | |
2683 | ||
2684 | ||
2685 | void | |
2686 | typeresolution_info::visit_stat_op (stat_op* e) | |
2687 | { | |
2688 | t = pe_stats; | |
2689 | e->stat->visit (this); | |
2690 | if (e->type == pe_unknown) | |
2691 | { | |
2692 | e->type = pe_long; | |
2693 | resolved (e->tok, e->type); | |
2694 | } | |
07c17d67 | 2695 | else if (e->type != pe_long) |
d02548c0 GH |
2696 | mismatch (e->tok, e->type, pe_long); |
2697 | } | |
2698 | ||
2699 | void | |
2700 | typeresolution_info::visit_hist_op (hist_op* e) | |
2701 | { | |
2702 | t = pe_stats; | |
2703 | e->stat->visit (this); | |
2704 | } | |
2705 | ||
2b066ec1 FCE |
2706 | |
2707 | void | |
2708 | typeresolution_info::unresolved (const token* tok) | |
2709 | { | |
2710 | num_still_unresolved ++; | |
2711 | ||
2712 | if (assert_resolvability) | |
2713 | { | |
bdef2583 FCE |
2714 | string nm = (current_function ? current_function->name : |
2715 | current_probe ? current_probe->name : | |
2716 | "?"); | |
2717 | cerr << "semantic error: " + nm + " with unresolved type for "; | |
2b066ec1 FCE |
2718 | if (tok) |
2719 | cerr << *tok; | |
2720 | else | |
2721 | cerr << "a token"; | |
2722 | cerr << endl; | |
2723 | } | |
2724 | } | |
2725 | ||
2726 | ||
2727 | void | |
2728 | typeresolution_info::invalid (const token* tok, exp_type pe) | |
2729 | { | |
2730 | num_still_unresolved ++; | |
2731 | ||
2732 | if (assert_resolvability) | |
2733 | { | |
bdef2583 FCE |
2734 | string nm = (current_function ? current_function->name : |
2735 | current_probe ? current_probe->name : | |
2736 | "?"); | |
022b623f DS |
2737 | if (tok && tok->type == tok_operator) |
2738 | { | |
2739 | cerr << "semantic error: " + nm + " uses invalid " << *tok; | |
2740 | } | |
2b066ec1 | 2741 | else |
022b623f DS |
2742 | { |
2743 | cerr << "semantic error: " + nm + " with invalid type " << pe << " for "; | |
2744 | if (tok) | |
2745 | cerr << *tok; | |
2746 | else | |
2747 | cerr << "a token"; | |
2748 | } | |
2b066ec1 FCE |
2749 | cerr << endl; |
2750 | } | |
2751 | } | |
2752 | ||
2753 | ||
2754 | void | |
2755 | typeresolution_info::mismatch (const token* tok, exp_type t1, exp_type t2) | |
2756 | { | |
2757 | num_still_unresolved ++; | |
2758 | ||
2759 | if (assert_resolvability) | |
2760 | { | |
bdef2583 FCE |
2761 | string nm = (current_function ? current_function->name : |
2762 | current_probe ? current_probe->name : | |
2763 | "?"); | |
2764 | cerr << "semantic error: " + nm + " with type mismatch for "; | |
2b066ec1 FCE |
2765 | if (tok) |
2766 | cerr << *tok; | |
2767 | else | |
2768 | cerr << "a token"; | |
2769 | cerr << ": " << t1 << " vs. " << t2 << endl; | |
2770 | } | |
2771 | } | |
2772 | ||
2773 | ||
2774 | void | |
78f6bba6 | 2775 | typeresolution_info::resolved (const token*, exp_type) |
2b066ec1 FCE |
2776 | { |
2777 | num_newly_resolved ++; | |
2778 | // cerr << "resolved " << *e->tok << " type " << t << endl; | |
2779 | } | |
2780 |