David Goulet authored Jan 11, 2023 and Nick Mathewson committed Jan 11, 2023

Signed-off-by: David Goulet <dgoulet@torproject.org>

Our previous rule had been that all u8 arrays needed a `truncated:`
section.  That's fine if the arrays have a length, but not if their
length extends to the end of the structure: such arrays can't be
truncated.

This time I also add a test to make sure change works.

This reverts commit 6d24296e.

I was modifying an _encode_ function when I should have modified the
_decode_ function.

Our previous rule had been that all u8 arrays needed a `truncated:`
section.  That's fine if the arrays have a length, but not if their
length extends to the end of the structure: such arrays can't be
truncated.

To run it, write a trunnel description for what you want to fuzz,
and run python -m trunnel.SeedFuzzer foo.trunnel .  The
subdirectories of "fuzzing-inputs" will fill up with strings that
conform to that description, suitable for consumption by afl or
libfuzzer.

Bumping to Trunnel 1.5.1

This way, trunnel-local can do things like setting _FILE_OFFSET_BITS
other things that need to happen before the standards headers.

The C standard says that you can never say memcpy(x, NULL, 0), even
though most libcs permit it.

Found with asan.

Instead, assert that it is less than 0.

Otherwise, Coverity issues frequent complaints about dead code.

teor authored Oct 31, 2014 and Nick Mathewson committed Oct 31, 2014

clang warns that a few implicit casts in trunnel truncate values from
64 to 32 bit. We do explicit casts to silence these warnings.

Back when spark was new, sticking patterns and productions in
docstrings was kinda reasonable.  But now that we have function
decorators in Python, I should use them: doing so will let me
actually put documentation on the methods of the lexer and parser.