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