| 1 | """
|
| 2 | const.py
|
| 3 | """
|
| 4 |
|
| 5 | DEFAULT_INT_WIDTH = 3 # 24 bits
|
| 6 |
|
| 7 | # 2^24 - 1 is used as an invalid/uninitialized value for ASDL integers.
|
| 8 |
|
| 9 | # Why? We have a few use cases for invalid/sentinel values:
|
| 10 | # - span_id, line_id. Sometimes we don't have a span ID.
|
| 11 | # - file descriptor: 'read x < f.txt' vs 'read x 0< f.txt'
|
| 12 | #
|
| 13 | # Other options for representation:
|
| 14 | #
|
| 15 | # 1. ADSL could use signed integers, then -1 is valid.
|
| 16 | # 2. Use a type like fd = None | Some(int fd)
|
| 17 | #
|
| 18 | # I don't like #1 because ASDL is lazily-decoded, and then we have to do sign
|
| 19 | # extension on demand. (24 bits to 32 or 64). As far as I can tell, sign
|
| 20 | # extension requires a branch, at least in portable C (on the sign bit).
|
| 21 | #
|
| 22 | # The second option is semantically cleaner. But it needlessly
|
| 23 | # inflates the size of both the source code and the data. Instead of having a
|
| 24 | # single "inline" integer, we would need a reference to another value.
|
| 25 | #
|
| 26 | # We could also try to do some fancy thing like fd = None |
|
| 27 | # Range<1..max_fd>(fd), with smart encoding. But that is overkill for these
|
| 28 | # use cases.
|
| 29 | #
|
| 30 | # Using InvalidInt instead of -1 seems like a good compromise.
|
| 31 |
|
| 32 | NO_INTEGER = (1 << (DEFAULT_INT_WIDTH * 8)) - 1
|
| 33 |
|
| 34 | # NOTE: In Python: 1 << (n * 8) - 1 is wrong! I thought that bit shift would
|
| 35 | # have higher precedence.
|