1 //: Everything this project/binary supports.
  2 //: This should give you a sense for what to look forward to in later layers.
  3 
  4 :(before "End Commandline Parsing")
  5 if (argc <= 1 || is_equal(argv[1], "--help")) {
  6   //: this is the functionality later layers will provide
  7   // currently no automated tests for commandline arg parsing
  8   cerr << get(Help, "usage");
  9   return 0;
 10 }
 11 
 12 //: Support for option parsing.
 13 //: Options always begin with '--' and are always the first arguments. An
 14 //: option will never follow a non-option.
 15 char** arg = &argv[1];
 16 while (argc > 1 && starts_with(*arg, "--")) {
 17   if (false)
 18     ;  // no-op branch just so any further additions can consistently always start with 'else'
 19   // End Commandline Options(*arg)
 20   else
 21     cerr << "skipping unknown option " << *arg << '\n';
 22   --argc;  ++argv;  ++arg;
 23 }
 24 
 25 if (is_equal(argv[1], "help")) {
 26   if (argc == 2) {
 27     cerr << "help on what?\n";
 28     help_contents();
 29     return 0;
 30   }
 31   string key(argv[2]);
 32   // End Help Special-cases(key)
 33   if (contains_key(Help, key)) {
 34     cerr << get(Help, key);
 35     return 0;
 36   }
 37   else {
 38     cerr << "No help found for '" << key << "'\n";
 39     help_contents();
 40     cerr << "Please check your command for typos.\n";
 41     return 1;
 42   }
 43 }
 44 
 45 :(code)
 46 void help_contents() {
 47   cerr << "Available top-level topics:\n";
 48   cerr << "  usage\n";
 49   // End Help Contents
 50 }
 51 
 52 :(before "End Globals")
 53 map<string, string> Help;
 54 :(before "End Includes")
 55 #include <map>
 56 using std::map;
 57 :(before "End One-time Setup")
 58 init_help();
 59 :(code)
 60 void init_help() {
 61   put(Help, "usage",
 62     "bootstrap: the bootstrap translator for SubX.\n"
 63     "This program also wraps some miscellaneous useful functionality:\n"
 64     "  - an x86 emulator: `bootstrap run`\n"
 65     "  - online help: `bootstrap help`\n"
 66     "\n"
 67     "== Ways to invoke bootstrap\n"
 68     "- See this message:\n"
 69     "    bootstrap --help\n"
 70     "- Convert a textual SubX program into a standard ELF binary that you can\n"
 71     "  run on your computer:\n"
 72     "    bootstrap translate input1.subx input2.subx ... -o <output ELF binary>\n"
 73     "- Run a SubX binary using SubX itself (for better error messages):\n"
 74     "    bootstrap run <ELF binary>\n"
 75     "- Run all bootstrap's unit tests:\n"
 76     "    bootstrap test\n"
 77     "- Run a single unit test:\n"
 78     "    bootstrap test <test name>\n"
 79     "     e.g. bootstrap test test_copy_imm32_to_EAX\n"
 80     "\n"
 81     "== Debugging aids\n"
 82     "- Add '--trace' to any of these commands to save a trace to disk at the end.\n"
 83     "  This can run out of memory for long-running commands.\n"
 84     "- Add '--debug' to add information to traces. 'bootstrap --debug translate'\n"
 85     "  will save metadata to disk that 'bootstrap --trace run' uses to make traces\n"
 86     "  more informative.\n"
 87     "- Add '--dump --trace' to emit a trace incrementally to stderr.\n"
 88     "  This approach will work even for long-running programs.\n"
 89     "  (Though the combination of flags is counter-intuitive and can probably\n"
 90     "  be improved.)\n"
 91     "\n"
 92     "Options starting with '--' must always come before any other arguments.\n"
 93     "\n"
 94     "To start learning how to write SubX programs, see Readme.md (particularly\n"
 95     "the section on the x86 instruction set) and then run:\n"
 96     "  bootstrap help\n"
 97   );
 98   // End Help Texts
 99 }
100 
101 :(code)
102 bool is_equal(char* s, const char* lit) {
103   return strncmp(s, lit, strlen(lit)) == 0;
104 }
105 
106 bool starts_with(const string& s, const string& pat) {
107   string::const_iterator a=s.begin(), b=pat.begin();
108   for (/*nada*/;  a!=s.end() && b!=pat.end();  ++a, ++b)
109     if (*a != *b) return false;
110   return b == pat.end();
111 }
112 
113 //: I'll throw some style conventions here for want of a better place for them.
114 //: As a rule I hate style guides. Do what you want, that's my motto. But since
115 //: we're dealing with C/C++, the one big thing we want to avoid is undefined
116 //: behavior. If a compiler ever encounters undefined behavior it can make
117 //: your program do anything it wants.
118 //:
119 //: For reference, my checklist of undefined behaviors to watch out for:
120 //:   out-of-bounds access
121 //:   uninitialized variables
122 //:   use after free
123 //:   dereferencing invalid pointers: null, a new of size 0, others
124 //:
125 //:   casting a large number to a type too small to hold it
126 //:
127 //:   integer overflow
128 //:   division by zero and other undefined expressions
129 //:   left-shift by negative count
130 //:   shifting values by more than or equal to the number of bits they contain
131 //:   bitwise operations on signed numbers
132 //:
133 //:   Converting pointers to types of different alignment requirements
134 //:     T* -> void* -> T*: defined
135 //:     T* -> U* -> T*: defined if non-function pointers and alignment requirements are same
136 //:     function pointers may be cast to other function pointers
137 //:
138 //:       Casting a numeric value into a value that can't be represented by the target type (either directly or via static_cast)
139 //:
140 //: To guard against these, some conventions:
141 //:
142 //: 0. Initialize all primitive variables in functions and constructors.
143 //:
144 //: 1. Minimize use of pointers and pointer arithmetic. Avoid 'new' and
145 //: 'delete' as far as possible. Rely on STL to perform memory management to
146 //: avoid use-after-free issues (and memory leaks).
147 //:
148 //: 2. Avoid naked arrays to avoid out-of-bounds access. Never use operator[]
149 //: except with map. Use at() with STL vectors and so on.
150 //:
151 //: 3. Valgrind all the things.
152 //:
153 //: 4. Avoid unsigned numbers. Not strictly an undefined-behavior issue, but
154 //: the extra range doesn't matter, and it's one less confusing category of
155 //: interaction gotchas to worry about.
156 //:
157 //: Corollary: don't use the size() method on containers, since it returns an
158 //: unsigned and that'll cause warnings about mixing signed and unsigned,
159 //: yadda-yadda. Instead use this macro below to perform an unsafe cast to
160 //: signed. We'll just give up immediately if a container's ever too large.
161 //: Basically, Mu is not concerned about this being a little slower than it
162 //: could be. (https://gist.github.com/rygorous/e0f055bfb74e3d5f0af20690759de5a7)
163 //:
164 //: Addendum to corollary: We're going to uniformly use int everywhere, to
165 //: indicate that we're oblivious to number size, and since Clang on 32-bit
166 //: platforms doesn't yet support multiplication over 64-bit integers, and
167 //: since multiplying two integers seems like a more common situation to end
168 //: up in than integer overflow.
169 :(before "End Includes")
170 #define SIZE(X) (assert((X).size() < (1LL<<(sizeof(int)*8-2))), static_cast<int>((X).size()))
171 
172 //: 5. Integer overflow is guarded against at runtime using the -ftrapv flag
173 //: to the compiler, supported by Clang (GCC version only works sometimes:
174 //: http://stackoverflow.com/questions/20851061/how-to-make-gcc-ftrapv-work).
175 :(before "atexit(reset)")
176 initialize_signal_handlers();  // not always necessary, but doesn't hurt
177 //? cerr << INT_MAX+1 << '\n';  // test overflow
178 //? assert(false);  // test SIGABRT
179 :(code)
180 // based on https://spin.atomicobject.com/2013/01/13/exceptions-stack-traces-c
181 void initialize_signal_handlers() {
182   struct sigaction action;
183   bzero(&action, sizeof(action));
184   action.sa_sigaction = dump_and_exit;
185   sigemptyset(&action.sa_mask);
186   sigaction(SIGABRT, &action, NULL);  // assert() failure or integer overflow on linux (with -ftrapv)
187   sigaction(SIGILL,  &action, NULL);  // integer overflow on OS X (with -ftrapv)
188 }
189 void dump_and_exit(int sig, siginfo_t* /*unused*/, void* /*unused*/) {
190   switch (sig) {
191     case SIGABRT:
192       #ifndef __APPLE__
193         cerr << "SIGABRT: might be an integer overflow if it wasn't an assert() failure\n";
194         _Exit(1);
195       #endif
196       break;
197     case SIGILL:
198       #ifdef __APPLE__
199         cerr << "SIGILL: most likely caused by integer overflow\n";
200         _Exit(1);
201       #endif
202       break;
203     default:
204       break;
205   }
206 }
207 :(before "End Includes")
208 #include <signal.h>
209 
210 //: 6. Map's operator[] being non-const is fucking evil.
211 :(before "Globals")  // can't generate prototypes for these
212 // from http://stackoverflow.com/questions/152643/idiomatic-c-for-reading-from-a-const-map
213 template<typename T> typename T::mapped_type& get(T& map, typename T::key_type const& key) {
214   typename T::iterator iter(map.find(key));
215   if (iter == map.end()) {
216     cerr << "get couldn't find key '" << key << "'\n";
217     assert(iter != map.end());
218   }
219   return iter->second;
220 }
221 template<typename T> typename T::mapped_type const& get(const T& map, typename T::key_type const& key) {
222   typename T::const_iterator iter(map.find(key));
223   if (iter == map.end()) {
224     cerr << "get couldn't find key '" << key << "'\n";
225     assert(iter != map.end());
226   }
227   return iter->second;
228 }
229 template<typename T> typename T::mapped_type const& put(T& map, typename T::key_type const& key, typename T::mapped_type const& value) {
230   map[key] = value;
231   return map[key];
232 }
233 template<typename T> bool contains_key(T& map, typename T::key_type const& key) {
234   return map.find(key) != map.end();
235 }
236 template<typename T> typename T::mapped_type& get_or_insert(T& map, typename T::key_type const& key) {
237   return map[key];
238 }
239 template<typename T> typename T::mapped_type const& put_new(T& map, typename T::key_type const& key, typename T::mapped_type const& value) {
240   assert(map.find(key) == map.end());
241   map[key] = value;
242   return map[key];
243 }
244 //: The contract: any container that relies on get_or_insert should never call
245 //: contains_key.
246 
247 //: 7. istreams are a royal pain in the arse. You have to be careful about
248 //: what subclass you try to putback into. You have to watch out for the pesky
249 //: failbit and badbit. Just avoid eof() and use this helper instead.
250 :(code)
251 bool has_data(istream& in) {
252   return in && !in.eof();
253 }
254 
255 :(before "End Includes")
256 #include <assert.h>
257 
258 #include <iostream>
259 using std::istream;
260 using std::ostream;
261 using std::iostream;
262 using std::cin;
263 using std::cout;
264 using std::cerr;
265 #include <iomanip>
266 
267 #include <string.h>
268 #include <string>
269 using std::string;
270 
271 #include <algorithm>
272 using std::min;
273 using std::max;