1 | """
|
2 | mylib.py
|
3 | """
|
4 | from __future__ import print_function
|
5 |
|
6 | try:
|
7 | import cStringIO
|
8 | except ImportError:
|
9 | # Python 3 doesn't have cStringIO. Our yaks/ demo currently uses
|
10 | # mycpp/mylib.py with Python 3.
|
11 | cStringIO = None
|
12 | import io
|
13 |
|
14 | import sys
|
15 |
|
16 | from pylib import collections_
|
17 | try:
|
18 | import posix_ as posix
|
19 | except ImportError:
|
20 | # Hack for tangled dependencies.
|
21 | import os
|
22 | posix = os
|
23 |
|
24 | from typing import (Tuple, List, Dict, Optional, Iterator, Any, TypeVar,
|
25 | Generic, cast, TYPE_CHECKING)
|
26 | if TYPE_CHECKING:
|
27 | from mycpp import mops
|
28 |
|
29 | # For conditional translation
|
30 | CPP = False
|
31 | PYTHON = True
|
32 |
|
33 | # Use POSIX name directly
|
34 | STDIN_FILENO = 0
|
35 |
|
36 |
|
37 | def MaybeCollect():
|
38 | # type: () -> None
|
39 | pass
|
40 |
|
41 |
|
42 | def NewDict():
|
43 | # type: () -> Dict[str, Any]
|
44 | """Make dictionaries ordered in Python, e.g. for JSON.
|
45 |
|
46 | In C++, our Dict implementation should be ordered.
|
47 | """
|
48 | return collections_.OrderedDict()
|
49 |
|
50 |
|
51 | def log(msg, *args):
|
52 | # type: (str, *Any) -> None
|
53 | """Print debug output to stderr."""
|
54 | if args:
|
55 | msg = msg % args
|
56 | print(msg, file=sys.stderr)
|
57 |
|
58 |
|
59 | def print_stderr(s):
|
60 | # type: (str) -> None
|
61 | """Print a message to stderr for the user.
|
62 |
|
63 | This should be used sparingly, since it doesn't have location info, like
|
64 | ui.ErrorFormatter does. We use it to print fatal I/O errors that were only
|
65 | caught at the top level.
|
66 | """
|
67 | print(s, file=sys.stderr)
|
68 |
|
69 |
|
70 | #
|
71 | # Byte Operations avoid excessive allocations with string algorithms
|
72 | #
|
73 |
|
74 |
|
75 | def ByteAt(s, i):
|
76 | # type: (str, int) -> int
|
77 | """i must be in bounds."""
|
78 |
|
79 | # This simplifies the C++ implementation
|
80 | assert 0 <= i, 'No negative indices'
|
81 | assert i < len(s), 'No negative indices'
|
82 |
|
83 | return ord(s[i])
|
84 |
|
85 |
|
86 | def ByteEquals(byte, ch):
|
87 | # type: (int, str) -> bool
|
88 | assert len(ch) == 1, ch
|
89 | assert 0 <= byte < 256, byte
|
90 |
|
91 | return byte == ord(ch)
|
92 |
|
93 |
|
94 | def ByteInSet(byte, byte_set):
|
95 | # type: (int, str) -> bool
|
96 | assert 0 <= byte < 256, byte
|
97 |
|
98 | return chr(byte) in byte_set
|
99 |
|
100 |
|
101 | def JoinBytes(byte_list):
|
102 | # type: (List[int]) -> str
|
103 |
|
104 | return ''.join(chr(b) for b in byte_list)
|
105 |
|
106 |
|
107 | #
|
108 | # For SparseArray
|
109 | #
|
110 |
|
111 |
|
112 | def BigIntSort(keys):
|
113 | # type: (List[mops.BigInt]) -> None
|
114 | keys.sort(key=lambda big: big.i)
|
115 |
|
116 |
|
117 | #
|
118 | # Files
|
119 | #
|
120 |
|
121 |
|
122 | class File:
|
123 | """
|
124 | TODO: This should define a read/write interface, and then LineReader() and
|
125 | Writer() can possibly inherit it, with runtime assertions
|
126 |
|
127 | Then we allow downcasting from File -> LineReader, like we currently do in
|
128 | C++ in gc_mylib.h.
|
129 |
|
130 | Inheritance can't express the structural Reader/Writer pattern of Go, which
|
131 | would be better. I suppose we could use File* everywhere, but having
|
132 | fine-grained types is nicer. And there will be very few casts.
|
133 | """
|
134 | pass
|
135 |
|
136 |
|
137 | class LineReader:
|
138 |
|
139 | def readline(self):
|
140 | # type: () -> str
|
141 | raise NotImplementedError()
|
142 |
|
143 | def close(self):
|
144 | # type: () -> None
|
145 | raise NotImplementedError()
|
146 |
|
147 | def isatty(self):
|
148 | # type: () -> bool
|
149 | raise NotImplementedError()
|
150 |
|
151 |
|
152 | if TYPE_CHECKING:
|
153 |
|
154 | class BufLineReader(LineReader):
|
155 |
|
156 | def __init__(self, s):
|
157 | # type: (str) -> None
|
158 | raise NotImplementedError()
|
159 |
|
160 | def open(path):
|
161 | # type: (str) -> LineReader
|
162 |
|
163 | # TODO: should probably return mylib.File
|
164 | # mylib.open() is currently only used in yaks/yaks_main and
|
165 | # bin.osh_parse
|
166 | raise NotImplementedError()
|
167 |
|
168 | else:
|
169 | # Actual runtime
|
170 | if cStringIO:
|
171 | BufLineReader = cStringIO.StringIO
|
172 | else: # Python 3
|
173 | BufLineReader = io.StringIO
|
174 |
|
175 | open = open
|
176 |
|
177 |
|
178 | class Writer:
|
179 |
|
180 | def write(self, s):
|
181 | # type: (str) -> None
|
182 | raise NotImplementedError()
|
183 |
|
184 | def flush(self):
|
185 | # type: () -> None
|
186 | raise NotImplementedError()
|
187 |
|
188 | def isatty(self):
|
189 | # type: () -> bool
|
190 | raise NotImplementedError()
|
191 |
|
192 | def close(self):
|
193 | # type: () -> None
|
194 | raise NotImplementedError()
|
195 |
|
196 |
|
197 | class BufWriter(Writer):
|
198 | """Mimic StringIO API, but add clear() so we can reuse objects.
|
199 |
|
200 | We can also add accelerators for directly writing numbers, to avoid
|
201 | allocations when encoding JSON.
|
202 | """
|
203 |
|
204 | def __init__(self):
|
205 | # type: () -> None
|
206 | self.parts = []
|
207 |
|
208 | def write(self, s):
|
209 | # type: (str) -> None
|
210 | self.parts.append(s)
|
211 |
|
212 | def isatty(self):
|
213 | # type: () -> bool
|
214 | return False
|
215 |
|
216 | def write_spaces(self, n):
|
217 | # type: (int) -> None
|
218 | """For JSON indenting. Avoid intermediate allocations in C++."""
|
219 | self.parts.append(' ' * n)
|
220 |
|
221 | def getvalue(self):
|
222 | # type: () -> str
|
223 | return ''.join(self.parts)
|
224 |
|
225 | def clear(self):
|
226 | # type: () -> None
|
227 | del self.parts[:]
|
228 |
|
229 | def close(self):
|
230 | # type: () -> None
|
231 |
|
232 | # No-op for now - we could invalidate write()?
|
233 | pass
|
234 |
|
235 |
|
236 | def Stdout():
|
237 | # type: () -> Writer
|
238 | return sys.stdout
|
239 |
|
240 |
|
241 | def Stderr():
|
242 | # type: () -> Writer
|
243 | return sys.stderr
|
244 |
|
245 |
|
246 | def Stdin():
|
247 | # type: () -> LineReader
|
248 | return sys.stdin
|
249 |
|
250 |
|
251 | class switch(object):
|
252 | """Translates to C switch on int.
|
253 |
|
254 | with tagswitch(i) as case:
|
255 | if case(42, 43):
|
256 | print('hi')
|
257 | elif case(99):
|
258 | print('two')
|
259 | else:
|
260 | print('neither')
|
261 | """
|
262 |
|
263 | def __init__(self, value):
|
264 | # type: (int) -> None
|
265 | self.value = value
|
266 |
|
267 | def __enter__(self):
|
268 | # type: () -> switch
|
269 | return self
|
270 |
|
271 | def __exit__(self, type, value, traceback):
|
272 | # type: (Any, Any, Any) -> bool
|
273 | return False # Allows a traceback to occur
|
274 |
|
275 | def __call__(self, *cases):
|
276 | # type: (*Any) -> bool
|
277 | return self.value in cases
|
278 |
|
279 |
|
280 | class str_switch(object):
|
281 | """Translates to fast dispatch on string length, then memcmp()."""
|
282 |
|
283 | def __init__(self, value):
|
284 | # type: (str) -> None
|
285 | self.value = value
|
286 |
|
287 | def __enter__(self):
|
288 | # type: () -> switch
|
289 | return self
|
290 |
|
291 | def __exit__(self, type, value, traceback):
|
292 | # type: (Any, Any, Any) -> bool
|
293 | return False # Allows a traceback to occur
|
294 |
|
295 | def __call__(self, *cases):
|
296 | # type: (*Any) -> bool
|
297 | return self.value in cases
|
298 |
|
299 |
|
300 | class tagswitch(object):
|
301 | """A ContextManager that translates to switch statement over ASDL types."""
|
302 |
|
303 | def __init__(self, node):
|
304 | # type: (Any) -> None
|
305 | self.tag = node.tag()
|
306 |
|
307 | def __enter__(self):
|
308 | # type: () -> tagswitch
|
309 | return self
|
310 |
|
311 | def __exit__(self, type, value, traceback):
|
312 | # type: (Any, Any, Any) -> bool
|
313 | return False # Allows a traceback to occur
|
314 |
|
315 | def __call__(self, *cases):
|
316 | # type: (*Any) -> bool
|
317 | return self.tag in cases
|
318 |
|
319 |
|
320 | if TYPE_CHECKING:
|
321 | # Doesn't work
|
322 | T = TypeVar('T')
|
323 |
|
324 | class StackArray(Generic[T]):
|
325 |
|
326 | def __init__(self):
|
327 | self.items = [] # type: List[T]
|
328 |
|
329 | def append(self, item):
|
330 | # type: (T) -> None
|
331 | self.items.append(item)
|
332 |
|
333 | def pop(self):
|
334 | # type: () -> T
|
335 | return self.items.pop()
|
336 |
|
337 | # Doesn't work, this is only for primitive types
|
338 | #StackArray = NewType('StackArray', list)
|
339 |
|
340 |
|
341 | def MakeStackArray(item_type):
|
342 | # type: (TypeVar) -> StackArray[item_type]
|
343 | """
|
344 | Convenience "constructor" used like this:
|
345 |
|
346 | myarray = MakeStackArray(int)
|
347 |
|
348 | The idiom could also be
|
349 |
|
350 | myarray = cast('StackArray[int]', [])
|
351 |
|
352 | But that's uglier.
|
353 | """
|
354 | return cast('StackArray[item_type]', [])
|
355 |
|
356 |
|
357 | if TYPE_CHECKING:
|
358 | K = TypeVar('K')
|
359 | V = TypeVar('V')
|
360 |
|
361 |
|
362 | def iteritems(d):
|
363 | # type: (Dict[K, V]) -> Iterator[Tuple[K, V]]
|
364 | """Make translation a bit easier."""
|
365 | return d.iteritems()
|
366 |
|
367 |
|
368 | def split_once(s, delim):
|
369 | # type: (str, str) -> Tuple[str, Optional[str]]
|
370 | """Easier to call than split(s, 1) because of tuple unpacking."""
|
371 |
|
372 | parts = s.split(delim, 1)
|
373 | if len(parts) == 1:
|
374 | no_str = None # type: Optional[str]
|
375 | return s, no_str
|
376 | else:
|
377 | return parts[0], parts[1]
|
378 |
|
379 |
|
380 | def hex_lower(i):
|
381 | # type: (int) -> str
|
382 | return '%x' % i
|
383 |
|
384 |
|
385 | def dict_erase(d, key):
|
386 | # type: (Dict[Any, Any], Any) -> None
|
387 | """
|
388 | Ensure that a key isn't in the Dict d. This makes C++ translation easier.
|
389 | """
|
390 | try:
|
391 | del d[key]
|
392 | except KeyError:
|
393 | pass
|
394 |
|
395 |
|
396 | def str_cmp(s1, s2):
|
397 | # type: (str, str) -> int
|
398 | if s1 == s2:
|
399 | return 0
|
400 | if s1 < s2:
|
401 | return -1
|
402 | else:
|
403 | return 1
|
404 |
|
405 |
|
406 | class UniqueObjects(object):
|
407 | """A set of objects identified by their address in memory
|
408 |
|
409 | Python's id(obj) returns the address of any object. But we don't simply
|
410 | implement it, because it requires a uint64_t on 64-bit systems, while mycpp
|
411 | only supports 'int'.
|
412 |
|
413 | So we have a whole class.
|
414 |
|
415 | Should be used for:
|
416 |
|
417 | - Cycle detection when pretty printing, as Python's repr() does
|
418 | - See CPython's Objects/object.c PyObject_Repr()
|
419 | /* These methods are used to control infinite recursion in repr, str, print,
|
420 | etc. Container objects that may recursively contain themselves,
|
421 | e.g. builtin dictionaries and lists, should use Py_ReprEnter() and
|
422 | Py_ReprLeave() to avoid infinite recursion.
|
423 | */
|
424 | - e.g. dictobject.c dict_repr() calls Py_ReprEnter() to print {...}
|
425 | - In Python 2.7 a GLOBAL VAR is used
|
426 |
|
427 | - It also checks for STACK OVERFLOW
|
428 |
|
429 | - Packle serialization
|
430 | """
|
431 |
|
432 | def __init__(self):
|
433 | # 64-bit id() -> small integer ID
|
434 | self.addresses = {} # type: Dict[int, int]
|
435 |
|
436 | def Contains(self, obj):
|
437 | # type: (Any) -> bool
|
438 | """ Convenience? """
|
439 | return self.Get(obj) != -1
|
440 |
|
441 | def MaybeAdd(self, obj):
|
442 | # type: (Any) -> None
|
443 | """ Convenience? """
|
444 |
|
445 | # def AddNewObject(self, obj):
|
446 | def Add(self, obj):
|
447 | # type: (Any) -> None
|
448 | """
|
449 | Assert it isn't already there, and assign a new ID!
|
450 |
|
451 | # Lib/pickle does:
|
452 |
|
453 | self.memo[id(obj)] = memo_len, obj
|
454 |
|
455 | I guess that's the object ID and a void*
|
456 |
|
457 | Then it does:
|
458 |
|
459 | x = self.memo.get(id(obj))
|
460 |
|
461 | and
|
462 |
|
463 | # If the object is already in the memo, this means it is
|
464 | # recursive. In this case, throw away everything we put on the
|
465 | # stack, and fetch the object back from the memo.
|
466 | if id(obj) in self.memo:
|
467 | write(POP + self.get(self.memo[id(obj)][0]))
|
468 |
|
469 | BUT It only uses the numeric ID!
|
470 | """
|
471 | addr = id(obj)
|
472 | assert addr not in self.addresses
|
473 | self.addresses[addr] = len(self.addresses)
|
474 |
|
475 | def Get(self, obj):
|
476 | # type: (Any) -> int
|
477 | """
|
478 | Returns unique ID assigned
|
479 |
|
480 | Returns -1 if it doesn't exist?
|
481 | """
|
482 | addr = id(obj)
|
483 | return self.addresses.get(addr, -1)
|
484 |
|
485 | # Note: self.memo.clear() doesn't appear to be used
|
486 |
|
487 |
|
488 | def probe(provider, name, *args):
|
489 | # type: (str, str, Any) -> None
|
490 | """Create a probe for use with profilers like linux perf and ebpf or dtrace."""
|
491 | # Noop. Just a marker for mycpp to emit a DTRACE_PROBE()
|
492 | return
|
493 |
|
494 |
|
495 | if 0:
|
496 | # Prototype of Unix file descriptor I/O, compared with FILE* libc I/O.
|
497 | # Doesn't seem like we need this now.
|
498 |
|
499 | # Short versions of STDOUT_FILENO and STDERR_FILENO
|
500 | kStdout = 1
|
501 | kStderr = 2
|
502 |
|
503 | def writeln(s, fd=kStdout):
|
504 | # type: (str, int) -> None
|
505 | """Write a line. The name is consistent with JavaScript writeln() and Rust.
|
506 |
|
507 | e.g.
|
508 | writeln("x = %d" % x, kStderr)
|
509 |
|
510 | TODO: The Oil interpreter shouldn't use print() anywhere. Instead it can use
|
511 | writeln(s) and writeln(s, kStderr)
|
512 | """
|
513 | posix.write(fd, s)
|
514 | posix.write(fd, '\n')
|
515 |
|
516 | class File(object):
|
517 | """Custom file wrapper for Unix I/O like write() read()
|
518 |
|
519 | Not C I/O like fwrite() fread(). There should be no flush().
|
520 | """
|
521 |
|
522 | def __init__(self, fd):
|
523 | # type: (int) -> None
|
524 | self.fd = fd
|
525 |
|
526 | def write(self, s):
|
527 | # type: (str) -> None
|
528 | posix.write(self.fd, s)
|
529 |
|
530 | def writeln(self, s):
|
531 | # type: (str) -> None
|
532 | writeln(s, fd=self.fd)
|