that required explicitly calling LazyList.clear() in the two tests that
use LazyList (I added a LazyList Fibonacci generator too).
A real bitch: the extremely inefficient first version of the 2-3-5 test
*looked* like a slow leak on Win98SE, but it wasn't "really": it generated
so many results that the heap grew over 4Mb (tons of frames! the number
of frames grows exponentially in that test). Then Win98SE malloc() starts
fragmenting address space allocating more and more heaps, and the visible
memory use grew very slowly while the disk was thrashing like mad.
Printing fewer results (i.e., keeping the heap burden under 4Mb) made
that illusion vanish.
Looks like there's no hope for plugging the LazyList leaks automatically
short of adding frameobjects and genobjects to gc. OTOH, they're very
easy to break by hand, and they're the only *kind* of plausibly realistic
leaks I've been able to provoke.
Dilemma.
Implement sys.maxunicode.
Explicitly wrap around upper/lower computations for wide Py_UNICODE.
When decoding large characters with UTF-8, represent expected test
results using the \U notation.
Good news: Some of this stuff is pretty sophisticated (read nuts), and
I haven't bumped into a bug yet.
Bad news: If I run the doctest in an infinite loop, memory is clearly
leaking.
i_divmod: New and simpler algorithm. Old one returned gibberish on most
boxes when the numerator was -sys.maxint-1. Oddly enough, it worked in the
release (not debug) build on Windows, because the compiler optimized away
some tricky sign manipulations that were incorrect in this case.
Makes you wonder <wink> ...
Bugfix candidate.
Unfortunately, the std-mode bBhHIL codes don't do any range-checking; if
and when some of those get fixed, remove their letters from the
IntTester.BUGGY_RANGE_CHECK string. In the meantime, a msg saying that
range-tests are getting skipped is printed to stdout whenever one is
skipped.
This completes the q/Q project.
longobject.c _PyLong_AsByteArray: The original code had a gross bug:
the most-significant Python digit doesn't necessarily have SHIFT
significant bits, and you really need to count how many copies of the sign
bit it has else spurious overflow errors result.
test_struct.py: This now does exhaustive std q/Q testing at, and on both
sides of, all relevant power-of-2 boundaries, both positive and negative.
NEWS: Added brief dict news while I was at it.
native mode, and only when config #defines HAVE_LONG_LONG. Standard mode
will eventually treat them as 8-byte ints across all platforms, but that
likely requires a new set of routines in longobject.c first (while
sizeof(long) >= 4 is guaranteed by C, there's nothing in C we can rely
on x-platform to hold 8 bytes of int, so we'll have to roll our own;
I'm thinking of a simple pair of conversion functions, Python long
to/from sized vector of unsigned bytes; that may be useful for GMP
conversions too; std q/Q would call them with size fixed at 8).
test_struct.py: In addition to adding some native-mode 'q' and 'Q' tests,
got rid of unused code, and repaired a non-portable assumption about
native sizeof(short) (it isn't 2 on some Cray boxes).
libstruct.tex: In addition to adding a bit of 'q'/'Q' docs (more needed
later), removed an erroneous footnote about 'I' behavior.
be possible to provoke unbounded recursion now, but leaving that to someone
else to provoke and repair.
Bugfix candidate -- although this is getting harder to backstitch, and the
cases it's protecting against are mondo contrived.
random inputs: if you ran the test 100 times, you could expect it to
report a bogus failure. So loosened its expectations.
Also changed the way failing tests are printed, so that when run under
regrtest.py we get enough info to reproduce the failure.
exactly once. But the test code can't know that, as the number of times
__cmp__ is called depends on internal details of the dict implementation.
This is especially nasty because the __hash__ method returns the address
of the class object, so the hash codes seen by the dict can vary across
runs, causing the dict to use a different probe order across runs. I
just happened to see this test fail about 1 run in 7 today, but only
under a release build and when passing -O to Python. So, changed the test
to be predictable across runs.
name of the test, only write the output file if it already exists (and
tell the user to consider removing it). This avoids the generation of
unnecessary turds.
dictresize() was too aggressive about never ever resizing small dicts.
If a small dict is entirely full, it needs to rebuild it despite that
it won't actually resize it, in order to purge old dummy entries thus
creating at least one virgin slot (lookdict assumes at least one such
exists).
Also took the opportunity to add some high-level comments to dictresize.
Tim Peters
Martin v. Löwis
Barry Warsaw
Jack Jansen
Marc-André Lemburg
Fred Drake
Jeremy Hylton
Guido van Rossum