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 emit additional debug information during translation.\n"
 85     "  'bootstrap --debug translate' will save metadata to disk that\n"
 86     "  'bootstrap --trace run' uses to make traces more informative.\n"
 87     "\n"
 88     "Options starting with '--' must always come before any other arguments.\n"
 89     "\n"
 90     "To start learning how to write SubX programs, see Readme.md (particularly\n"
 91     "the section on the x86 instruction set) and then run:\n"
 92     "  bootstrap help\n"
 93   );
 94   // End Help Texts
 95 }
 96 
 97 :(code)
 98 bool is_equal(char* s, const char* lit) {
 99   return strncmp(s, lit, strlen(lit)) == 0;
100 }
101 
102 bool starts_with(const string& s, const string& pat) {
103   string::const_iterator a=s.begin(), b=pat.begin();
104   for (/*nada*/;  a!=s.end() && b!=pat.end();  ++a, ++b)
105     if (*a != *b) return false;
106   return b == pat.end();
107 }
108 
109 //: I'll throw some style conventions here for want of a better place for them.
110 //: As a rule I hate style guides. Do what you want, that's my motto. But since
111 //: we're dealing with C/C++, the one big thing we want to avoid is undefined
112 //: behavior. If a compiler ever encounters undefined behavior it can make
113 //: your program do anything it wants.
114 //:
115 //: For reference, my checklist of undefined behaviors to watch out for:
116 //:   out-of-bounds access
117 //:   uninitialized variables
118 //:   use after free
119 //:   dereferencing invalid pointers: null, a new of size 0, others
120 //:
121 //:   casting a large number to a type too small to hold it
122 //:
123 //:   integer overflow
124 //:   division by zero and other undefined expressions
125 //:   left-shift by negative count
126 //:   shifting values by more than or equal to the number of bits they contain
127 //:   bitwise operations on signed numbers
128 //:
129 //:   Converting pointers to types of different alignment requirements
130 //:     T* -> void* -> T*: defined
131 //:     T* -> U* -> T*: defined if non-function pointers and alignment requirements are same
132 //:     function pointers may be cast to other function pointers
133 //:
134 //:       Casting a numeric value into a value that can't be represented by the target type (either directly or via static_cast)
135 //:
136 //: To guard against these, some conventions:
137 //:
138 //: 0. Initialize all primitive variables in functions and constructors.
139 //:
140 //: 1. Minimize use of pointers and pointer arithmetic. Avoid 'new' and
141 //: 'delete' as far as possible. Rely on STL to perform memory management to
142 //: avoid use-after-free issues (and memory leaks).
143 //:
144 //: 2. Avoid naked arrays to avoid out-of-bounds access. Never use operator[]
145 //: except with map. Use at() with STL vectors and so on.
146 //:
147 //: 3. Valgrind all the things.
148 //:
149 //: 4. Avoid unsigned numbers. Not strictly an undefined-behavior issue, but
150 //: the extra range doesn't matter, and it's one less confusing category of
151 //: interaction gotchas to worry about.
152 //:
153 //: Corollary: don't use the size() method on containers, since it returns an
154 //: unsigned and that'll cause warnings about mixing signed and unsigned,
155 //: yadda-yadda. Instead use this macro below to perform an unsafe cast to
156 //: signed. We'll just give up immediately if a container's ever too large.
157 //: Basically, Mu is not concerned about this being a little slower than it
158 //: could be. (https://gist.github.com/rygorous/e0f055bfb74e3d5f0af20690759de5a7)
159 //:
160 //: Addendum to corollary: We're going to uniformly use int everywhere, to
161 //: indicate that we're oblivious to number size, and since Clang on 32-bit
162 //: platforms doesn't yet support multiplication over 64-bit integers, and
163 //: since multiplying two integers seems like a more common situation to end
164 //: up in than integer overflow.
165 :(before "End Includes")
166 #define SIZE(X) (assert((X).size() < (1LL<<(sizeof(int)*8-2))), static_cast<int>((X).size()))
167 
168 //: 5. Integer overflow is guarded against at runtime using the -ftrapv flag
169 //: to the compiler, supported by Clang (GCC version only works sometimes:
170 //: http://stackoverflow.com/questions/20851061/how-to-make-gcc-ftrapv-work).
171 :(before "atexit(reset)")
172 initialize_signal_handlers();  // not always necessary, but doesn't hurt
173 //? cerr << INT_MAX+1 << '\n';  // test overflow
174 //? assert(false);  // test SIGABRT
175 :(code)
176 // based on https://spin.atomicobject.com/2013/01/13/exceptions-stack-traces-c
177 void initialize_signal_handlers() {
178   struct sigaction action;
179   bzero(&action, sizeof(action));
180   action.sa_sigaction = dump_and_exit;
181   sigemptyset(&action.sa_mask);
182   sigaction(SIGABRT, &action, NULL);  // assert() failure or integer overflow on linux (with -ftrapv)
183   sigaction(SIGILL,  &action, NULL);  // integer overflow on OS X (with -ftrapv)
184 }
185 void dump_and_exit(int sig, siginfo_t* /*unused*/, void* /*unused*/) {
186   switch (sig) {
187     case SIGABRT:
188       #ifndef __APPLE__
189         cerr << "SIGABRT: might be an integer overflow if it wasn't an assert() failure\n";
190         _Exit(1);
191       #endif
192       break;
193     case SIGILL:
194       #ifdef __APPLE__
195         cerr << "SIGILL: most likely caused by integer overflow\n";
196         _Exit(1);
197       #endif
198       break;
199     default:
200       break;
201   }
202 }
203 :(before "End Includes")
204 #include <signal.h>
205 
206 //: 6. Map's operator[] being non-const is fucking evil.
207 :(before "Globals")  // can't generate prototypes for these
208 // from http://stackoverflow.com/questions/152643/idiomatic-c-for-reading-from-a-const-map
209 template<typename T> typename T::mapped_type& get(T& map, typename T::key_type const& key) {
210   typename T::iterator iter(map.find(key));
211   if (iter == map.end()) {
212     cerr << "get couldn't find key '" << key << "'\n";
213     assert(iter != map.end());
214   }
215   return iter->second;
216 }
217 template<typename T> typename T::mapped_type const& get(const T& map, typename T::key_type const& key) {
218   typename T::const_iterator iter(map.find(key));
219   if (iter == map.end()) {
220     cerr << "get couldn't find key '" << key << "'\n";
221     assert(iter != map.end());
222   }
223   return iter->second;
224 }
225 template<typename T> typename T::mapped_type const& put(T& map, typename T::key_type const& key, typename T::mapped_type const& value) {
226   map[key] = value;
227   return map[key];
228 }
229 template<typename T> bool contains_key(T& map, typename T::key_type const& key) {
230   return map.find(key) != map.end();
231 }
232 template<typename T> typename T::mapped_type& get_or_insert(T& map, typename T::key_type const& key) {
233   return map[key];
234 }
235 template<typename T> typename T::mapped_type const& put_new(T& map, typename T::key_type const& key, typename T::mapped_type const& value) {
236   assert(map.find(key) == map.end());
237   map[key] = value;
238   return map[key];
239 }
240 //: The contract: any container that relies on get_or_insert should never call
241 //: contains_key.
242 
243 //: 7. istreams are a royal pain in the arse. You have to be careful about
244 //: what subclass you try to putback into. You have to watch out for the pesky
245 //: failbit and badbit. Just avoid eof() and use this helper instead.
246 :(code)
247 bool has_data(istream& in) {
248   return in && !in.eof();
249 }
250 
251 :(before "End Includes")
252 #include <assert.h>
253 
254 #include <iostream>
255 using std::istream;
256 using std::ostream;
257 using std::iostream;
258 using std::cin;
259 using std::cout;
260 using std::cerr;
261 #include <iomanip>
262 
263 #include <string.h>
264 #include <string>
265 using std::string;
266 
267 #include <algorithm>
268 using std::min;
269 using std::max;