Fred Drake [Mon, 11 Jun 2001 15:21:48 +0000 (15:21 +0000)]
Fixed parameter order for os.popen2(), os.popen3(), and os.popen(4). Added
a reference to these functions and popen() from the "Process Management"
section.
Tim Peters [Sun, 10 Jun 2001 23:40:19 +0000 (23:40 +0000)]
Initial support for 'q' and 'Q' struct format codes: for now, only in
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.
Tim Peters [Sun, 10 Jun 2001 18:58:26 +0000 (18:58 +0000)]
SF bug 431772: traceback.print_exc() causes traceback
Patch from Michael Hundson.
format_exception_only() blew up when trying to report a SyntaxError
from a string input (line is None in this case, but it assumed a string).
Bugfix candidate.
Tim Peters [Sat, 9 Jun 2001 09:26:21 +0000 (09:26 +0000)]
SF bug 430991: wrong co_lnotab
Armin Rigo pointed out that the way the line-# table got built didn't work
for lines generating more than 255 bytes of bytecode. Fixed as he
suggested, plus corresponding changes to pyassem.py, plus added some
long overdue docs about this subtle table to compile.c.
Bugfix candidate (line numbers may be off in tracebacks under -O).
Martin v. Löwis [Sat, 9 Jun 2001 07:34:05 +0000 (07:34 +0000)]
Patch #424475: Speed-up tp_compare usage, by special-casing the common
case of objects with equal types which support tp_compare. Give
type objects a tp_compare function.
Also add c<0 tests before a few PyErr_Occurred tests.
Fred Drake [Fri, 8 Jun 2001 16:24:58 +0000 (16:24 +0000)]
Text from Tim & Guido discussing floating point arithmetic and what users
need to understand about the binary & decimal fp, so that representation
weirdness is documented somewhere. This makes it easier to repond to "bug"
reports caused by user confusion & ignorance of the issues.
Fred Drake [Fri, 8 Jun 2001 05:04:19 +0000 (05:04 +0000)]
In the section on extending the profiler, add some additional discussion
about setting up the dispatch table, and update the OldProfile and
HotProfile classes to the current implementations, showing the adjusted
construction for the dispatch table.
Fred Drake [Fri, 8 Jun 2001 04:33:09 +0000 (04:33 +0000)]
call_trace(): Add an additional parameter -- pointer to a PyObject*
that should be used to cache an interned version of the event
string passed to the profile/trace function. call_trace() will
create interned strings and cache them in using the storage
specified by this additional parameter, avoiding a lot of string
object creation at runtime when using the profiling or tracing
functions.
All call sites are modified to pass the additional parameter, and four
static PyObject* variables are allocated to cache the interned string
objects.
Fred Drake [Fri, 8 Jun 2001 04:25:24 +0000 (04:25 +0000)]
Performance improvements to the profiler:
Ensure that all the default timers are called as functions, not an
expensive method wrapper around a variety of different functions.
Agressively avoid dictionary lookups.
Modify the dispatch scheme (Profile.trace_dispatch_*(), where * is not
'call', 'exception' or 'return') so that the callables dispatched to
are simple functions and not bound methods -- this reduces the number
of layers of Python call machinery that gets touched.
Remove a couple of duplicate imports from the "if __name__ == ..."
section.
Acknowledge the existence of the 'platform' file, generated by the
Makefile and used by the setup.py script. Ignore it in .cvsignore;
remove it in "make clobber".
Tim Peters [Mon, 4 Jun 2001 21:00:21 +0000 (21:00 +0000)]
Store the mask instead of the size in dictobjects. The mask is more
frequently used, and in particular this allows to drop the last
remaining obvious time-waster in the crucial lookdict() and
lookdict_string() functions. Other changes consist mostly of changing
"i < ma_size" to "i <= ma_mask" everywhere.
Tim Peters [Sun, 3 Jun 2001 04:54:32 +0000 (04:54 +0000)]
lookdict: stop more insane core-dump mutating comparison cases. Should
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.
Tim Peters [Sun, 3 Jun 2001 04:14:43 +0000 (04:14 +0000)]
lookdict: Reduce obfuscating code duplication with a judicious goto.
This code is likely to get even hairier to squash core dumps due to
mutating comparisons, and it's hard enough to follow without that.
Fred Drake [Sun, 3 Jun 2001 03:12:57 +0000 (03:12 +0000)]
Explained more differences between PyList_SetItem() and PyList_SET_ITEM().
In particular, the affect on existing list content was not sufficiently
explained.
Tim Peters [Sat, 2 Jun 2001 08:02:56 +0000 (08:02 +0000)]
Coredumpers from Michael Hudson, mutating dicts while printing or
converting to string.
Critical bugfix candidate -- if you take this seriously <wink>.
Tim Peters [Sat, 2 Jun 2001 05:27:19 +0000 (05:27 +0000)]
New collision resolution scheme: no polynomials, simpler, faster, less
code, less memory. Tests have uncovered no drawbacks. Christian and
Vladimir are the other two people who have burned many brain cells on the
dict code in recent years, and they like the approach too, so I'm checking
it in without further ado.
Tim Peters [Tue, 29 May 2001 22:18:09 +0000 (22:18 +0000)]
This division test was too stringent in its accuracy expectations for
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.
Tim Peters [Tue, 29 May 2001 21:14:32 +0000 (21:14 +0000)]
BadDictKey test: The output file expected "raising error" to be printed
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.
Fred Drake [Tue, 29 May 2001 19:53:46 +0000 (19:53 +0000)]
New solution to the "Someone stuck a colon in that filename!" problem:
Allow colons in the labels used for internal references, but do not
expose them when generating filename.
Fred Drake [Tue, 29 May 2001 18:51:41 +0000 (18:51 +0000)]
Users of PySequence_GET_FAST() should get the length of the sequence using
PySequence_Size(), not PyObject_Size(): the later considers the mapping
methods as well as the sequence methods, which is not needed here. Either
should be equally fast in this case, but PySequence_Size() offers a better
conceptual match.
Fred Drake [Tue, 29 May 2001 18:13:06 +0000 (18:13 +0000)]
readlink() description: Added note that the return value may be either
absolute or relative.
remove(), rename() descriptions: Give more information about the cross-
platform behavior of these functions, so single-platform developers
can be aware of the potential issues when writing portable code.
Jeremy Hylton [Tue, 29 May 2001 17:46:19 +0000 (17:46 +0000)]
Change cascaded if stmts to switch stmt in vgetargs1().
In the default branch, keep three ifs that are used if level == 0, the
most common case. Note that first if here is a slight optimization
for the 'O' format.
Jeremy Hylton [Tue, 29 May 2001 17:37:05 +0000 (17:37 +0000)]
Internal refactoring of convertsimple() and friends.
Note that lots of code was re-indented.
Replace two-step of convertsimple() and convertsimple1() with
convertsimple() and helper converterr(), which is called to format
error messages when convertsimple() fails. The old code did all the
real work in convertsimple1(), but deferred error message formatting
to conversimple(). The result was paying the price of a second
function call on every call just to format error messages in the
failure cases.
Factor out of the buffer-handling code in convertsimple() and package
it as convertbuffer().
Add two macros to ease readability of Unicode coversions,
UNICODE_DEFAULT_ENCODING() and CONV_UNICODE, an error string.
The convertsimple() routine had awful indentation problems, primarily
because there were two tabs between the case line and the body of the
case statements. This patch reformats the entire function to have a
single tab between case line and case body, which makes the code
easier to read (and consistent with ceval). The introduction of
converterr() exacerbated the problem and prompted this fix.
Also, eliminate non-standard whitespace after opening paren and before
closing paren in a few if statements.
Fred Drake [Tue, 29 May 2001 17:10:51 +0000 (17:10 +0000)]
runtest(): When generating output, if the result is a single line with the
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.
Fred Drake [Tue, 29 May 2001 16:10:07 +0000 (16:10 +0000)]
Hack to make this play nicer with *old* versions of Python: os.path.abspath()
was not available in Python 1.5.1. (Yes, a user actually tried to use this
with that version of Python!)
Fred Drake [Tue, 29 May 2001 16:02:35 +0000 (16:02 +0000)]
Bring the notes on the relationship between __cmp__(), __eq__(), and
__hash__() up to date (re: use of objects which define these methods
as dictionary keys).
Fred Drake [Tue, 29 May 2001 15:34:06 +0000 (15:34 +0000)]
Removed information on the old third parameter to _PyTuple_Resize().
Added information on PyIter_Check(), PyIter_Next(),
PyObject_Unicode(), PyString_AsDecodedObject(),
PyString_AsEncodedObject(), and PyThreadState_GetDict().
Fred Drake [Tue, 29 May 2001 15:13:00 +0000 (15:13 +0000)]
Do not start API descriptions with "Does the same, but ..." -- actually
state *which* other function the current one is like, even if the
descriptions are adjacent.
Revise the _PyTuple_Resize() description to reflect the removal of the
third parameter.
Tim Peters [Tue, 29 May 2001 04:27:01 +0000 (04:27 +0000)]
Patch from Gordon McMillan.
updatecache(): When using imputil, sys.path may contain things other than
strings. Ignore such things instead of blowing up.
Hard to say whether this is a bugfix or a feature ...
Thomas Wouters [Mon, 28 May 2001 13:11:02 +0000 (13:11 +0000)]
_PyTuple_Resize: take into account the empty tuple. There can be only one.
Instead of raising a SystemError, just create a new tuple of the desired
size.
Tim Peters [Sun, 27 May 2001 07:39:22 +0000 (07:39 +0000)]
Implement an old idea of Christian Tismer's: use polynomial division
instead of multiplication to generate the probe sequence. The idea is
recorded in Python-Dev for Dec 2000, but that version is prone to rare
infinite loops.
The value is in getting *all* the bits of the hash code to participate;
and, e.g., this speeds up querying every key in a dict with keys
[i << 16 for i in range(20000)] by a factor of 500. Should be equally
valuable in any bad case where the high-order hash bits were getting
ignored.
Also wrote up some of the motivations behind Python's ever-more-subtle
hash table strategy.
Jack Jansen [Sat, 26 May 2001 20:01:41 +0000 (20:01 +0000)]
When reading from stdin (with the dialog box) use any partial line on
stdout as the prompt. This makes raw_input() and print "xxx", ; sys.stdin.readline() work a bit more palatable.
Tim Peters [Sat, 26 May 2001 05:28:40 +0000 (05:28 +0000)]
roundupsize() and friends: fiddle over-allocation strategy for list
resizing.
Accurate timings are impossible on my Win98SE box, but this is obviously
faster even on this box for reasonable list.append() cases. I give
credit for this not to the resizing strategy but to getting rid of integer
multiplication and divsion (in favor of shifting) when computing the
rounded-up size.
For unreasonable list.append() cases, Win98SE now displays linear behavior
for one-at-time appends up to a list with about 35 million elements. Then
it dies with a MemoryError, due to fatally fragmented *address space*
(there's plenty of VM available, but by this point Win9X has broken user
space into many distinct heaps none of which has enough contiguous space
left to resize the list, and for whatever reason Win9x isn't coalescing
the dead heaps). Before the patch it got a MemoryError for the same
reason, but once the list reached about 2 million elements.
Haven't yet tried on Win2K but have high hopes extreme list.append()
will be much better behaved now (NT & Win2K didn't fragment address space,
but suffered obvious quadratic-time behavior before as lists got large).
For other systems I'm relying on common sense: replacing integer * and /
by << and >> can't plausibly hurt, the number of function calls hasn't
changed, and the total operation count for reasonably small lists is about
the same (while the operations are cheaper now).
Fred Drake [Fri, 25 May 2001 04:24:37 +0000 (04:24 +0000)]
Add descriptions of {}.iteritems(), {}.iterkeys(), and {}.itervalues()
in the table of mapping object operations. Re-numbered the list of
notes to reflect the move of the "Added in version 2.2." note to the list
of notes instead of being inserted into the last column of the table.
Tim Peters [Thu, 24 May 2001 16:26:40 +0000 (16:26 +0000)]
dictresize(): Rebuild small tables if there are any dummies, not just if
they're entirely full. Not a question of correctness, but of temporarily
misplaced common sense.
Tim Peters [Wed, 23 May 2001 23:33:57 +0000 (23:33 +0000)]
Jack Jansen hit a bug in the new dict code, reported on python-dev.
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.
Barry Warsaw [Wed, 23 May 2001 16:59:45 +0000 (16:59 +0000)]
write(): Do two levels of sorting: first sort the individual location
tuples by filename/lineno, then sort the catalog entries by their
location tuples.
Guido van Rossum [Wed, 23 May 2001 13:24:30 +0000 (13:24 +0000)]
When Tim untabified this file, his editor accidentally assumed 4-space
tabs. The title was centered using 8-byte tabs, however, and the
result looked strange. Fixed this.
Tim Peters [Wed, 23 May 2001 07:46:36 +0000 (07:46 +0000)]
Remove test_doctest's expected-output file.
Change test_doctest and test_difflib to pass regrtest's notion of
verbosity on to doctest.
Add explanation for a dozen "new" things to test/README.