type for all functions. However many function call PyArg_Parse() and
need a 0. This is so that when they didn't change anything, the can
do Py_INCREF(args); return args. Reverted this back. For atof(),
there's no reason not to use PyArg_ParseTuple(), so I changed the code
(atoi and atol already used that).
it seems harmless for other platforms. It plays tricks with the name
of the library used to link with. Apparently DG/UX really wants a
shared library to link with if it wants shared modules to use symbols
from the library. I'm not sure why this wasn't an issue with 1.4;
DG/UX seems to be the only platform where moving to a single library
made things harder!
BTW This adds a target to create libpython$(VERSION).so; however this
target is *only* for DG/UX.
- Loading non-binary string pickles checks for insecure
strings. This is needed because cPickle (still)
uses a restricted eval to parse non-binary string pickles.
This change is needed to prevent untrusted
pickles like::
"S'hello world'*2000000\012p0\012."
from hosing an application.
- User-defined types can now support unpickling without
executing a constructor.
The second value returned from __reduce__ can now be None,
rather than an argument tuple. On unpickling, if the second
value returned from __reduce__ during pickling was None, then
rather than calling the first value returned from __reduce__,
directly, the __basicnew__ method of the first value returned
from __reduce__ is called without arguments.
- New option -x, to skip first line of script
- Use the correct platform-specific delimiter and library location in
the usage message
(Also removed two blank lines and moved one line around so that each
part of the usage message is again under 512 bytes and the whole usage
message still fits in 23 lines.)
the default build on Linux (because it requires -lcrypt which isn't
availabel everywhere).
Some improvements to the _tkinter build line suggested by Case Roole.
maxsplit which is implemented in string.py but wasn't here. The
reference manual doesn't define what happens when maxsplit is negative
or larger than the number of occurrences, but in either case, I
implemented this as all get replaced. Default value is zero which
replaces all occurrences.
signal handlers in a fork()ed child process when Python is compiled with
thread support. The bug was reported by Scott <scott@chronis.icgroup.com>.
What happens is that after a fork(), the variables used by the signal
module to determine whether this is the main thread or not are bogus,
and it decides that no thread is the main thread, so no signals will
be delivered.
The solution is the addition of PyOS_AfterFork(), which fixes the signal
module's variables. A dummy version of the function is present in the
intrcheck.c source file which is linked when the signal module is not
used.
to inside floatsleep(). This is necessary because floatsleep() does
the error handling and it must have grabbed the interpreter lock and
thread state before it can do so.
save and restore the tstate, but explicitly calling
PyEval_SaveThread() does reset it! While I think about how to fix
this for real, here's a fix that avoids getting a fatal error.
(1) Use PyErr_NewException("module.class", NULL, NULL) to create the
exception object.
(2) Remove all calls to Py_FatalError(); instead, return or
ignore the errors -- the import code now checks PyErr_Occurred()
after calling a module's init function, so it's no longer a
fatal error for the initialization to fail.
Also did some small cleanups, e.g. removed unnecessary test for
"already initialized" from initfpectl(), and unified
initposix()/initnt().
I haven't checked this very thoroughly, so while the changes are
pretty trivial -- beware of untested code!
This one works! However it requires using a modified version of
tclNotify.c (provided), which requires access to the Tcl source
to compile it. In order to enable this hack, add the following
to the Setup line for _tkinter:
tclNotify.c -DHAVE_PYTCL_WAITUNTILEVENT -I$(TCL)/generic
where TCL points to the source tree of Tcl 8.0. Other versions
of Tcl are not supported.
The tclNotify.c file is copyrighted by Sun Microsystems; the
licensing terms are in the file license.terms. According to this
file, no further permission to distribute this is required,
provided the file license.terms is included. Hence, I am checking
that in, too.
maps errno numbers to errno names (e.g. EINTR), and errorcode maps
them to message strings. (The latter is redundant because
the new call posix.strerror() now does the same, but alla...)
set_completer(function)
parse_and_bind(string)
read_init_file(filename)
The first is the most exciting feature: with an appropriate Python
completer function, it can do dynamic completion based on the contents
of your namespace!
Added 'p' format character for Pascal string (i.e. leading length
byte). This uses the count prefix line 's' does, except that the
count includes the length byte; i.e. '10p' takes 10 bytes packed but
has space for a length byte and 9 data bytes.
1. Fix bug in (de)compression objects. The final string resize used
zst.total_out to determine the length of the string, but the
(de)compression object will output data a little bit at a time, which
means total_out is not the string size. Fix: save original value of
total_out at the start of the call.
2. Be sure to Py_DECREF the result value if you exit with an
exception.
3. Use PyInt_FromLong instead of Py_BuildValue
4. include more constants from the zlib header file
5. Use PyErr_Format instead of using a local buffer and sprintf.
dealloc() functions contained code to free/DECREF the buffer
(there were differences between I and O objects but the logic bug was
the same). Fixed this be setting the buffer pointer to NULL and
testing for that. (This also makes it safe to call close() more than
once.)
XXX Worry: what if you try to read() or write() once the thing is
closed?
the executable must have that suffix. Note that there is no
corresponding support in the top-level Makefile because I'm not sure
that the install targets there make sense under these circumstances.
getpagesize() -- #ifdef doesn't work, Linux has conflicting decls in
its headers. Choice: only declare the return type, not the argument
prototype, and not on Linux.
-- initialize length to DEFAULTALLOC and not 0
-- resize string before returning (to remove '\000' padding)
Also converted some compression routines to use PyString instead of
buffer.
Change default alloc size for uncompressing to 16K.
Remove comment about core dumps when an invalid window sizes is used.
This bug has been fixed in zlib 1.0.4.
Two new optional arguments to decompress, wbits and bufsize. wbits
specifies the window size and bufsize specifies the initial output
string size.
In decompression code -- decompress and decompressobj methods -- use a
Python string (and _PyString_Resize) to collect the uncompressed
stream. Replaces a separate buffer that was copied into a string.
Fix bug in decompress that caused it to always realloc the buffer when
it was finished decompressing.
Modernized handling of optional arguments to compressobj.
Updated doc strings.
Removed handling of \e, \cX escapes, following a string-SIG discussion.
Fixed minor typos in re.py
re.error is now set equal to reop.error.
Move definition of constants like NORMAL and CHARCLASS into reop, which
exports them; re.py was changed to import them from reop.
Added C equivalents of _expand and expand_escape to reop, and changed
re.py to use them.
while one thread is blocked in mainloop(). Also, handle signals (not
just interrupts) as soon as they happen.
Cleanup: remove support for Tcl/Tk versions 7.4/4.0. (I've confirmed
that it works for 7.5/4.1 and 7.6/4.2, as well as 8.0b2.)
Coding style change: instead of ``func (args)'', write ``func(args)''
everywhere.
Minor functionality change: use PyArg_ParseTuple everywhere. This
should only affect the errors reported for bad argument lists; in
particular, deletefilehandler() is much clearer about what's going
on.
(XXX Still to do: Mac and Win ports to 8.0b2.)
it can be placed in the library.
Other, related changes:
- Moved the inspection of some environment variables to
Py_Initialize().
- Got rid of -s option.
- Moved Py_GetProgramName() and related logic to pythonrun.c; call
Py_SetProgramName() instead.
- Print the version header *after* successful initialization.
Not sure why this is generated, but this fixes a problem with a
particular file that was received with the following final line:
F-WE<-*A5]AY]%7>8'&!!(_Y<F*55_"*%46"<OFG=>_5(F/\'``!@
Ahlstrom:
Arraymodule.c has static functions H_getitem and h_getitem, and a
few others which differ only in case. These are a problem on
Windows 3.1, since a case-sensitive link causes Winsock to fail
(hey, it's not my fault). Please convert H_etc to HH_etc etc.
otherwise laudible attempt to rationalize the argument parsing): it
would save a copy of the original string instead of a reference to it.
Go back to saving a reference, but keep the "s#" format (using a hack
that involves two argument parsing steps, first using "O", then using
"s#").
cgen.py. Now that cgen.py and cstubs have been quickly renamed, check
in the actual output. This has some "old-style" names left in
(getilongarg etc.) but these are now take care of by macros in
cgensupport.h (which is now specific to glmodule.c).
This used to be done whenever stdin was interactive. Now we only do
it when the -i flag is given. Also (and this is the real reason for
this fix) we explicitly allocate a buffer -- this seems to be
necessary on Windows.
string_append(). These must be artifacts of GvR's rewrite.
Fixed some typos in the leading comment (and re-filled the
paragraphs).
Hope you don't mind, Guido.
This completes the getpath.c checkin. Note that to enable this in an
existing build tree, you'll have to edit your Setup and Setup.in file
to remove the $(DESTLIB) from all the PATH variable definitions.
Idea and first three implementation rounds due to Barry -- after that
I spent another day on it, hopefully it's enough for now :-)
(Wait for the checkin to Setup.in.)
even if it isn't. Changes:
- set the global flag Py_InteractiveFlag when -i is given
- call Py_FdIsInteractive() instead of isatty()
- make stdin unbuffered, too, when using -u
- make stdin and stdout line buffered, when stdin is interactive and not -u
Note that setting the environment variable PYTHONINSPECT does not have
these extra effects of -i. (Should it?)
Unlike Lee's changes, I don't set change the prompt to go to stderr
when -i is given.
initparser()) instead of statically (in the initializer). The static
initialization, using the address of an object in a different DLL, is
too much for the Microsoft VC++ compiler, and we want to be able to
build this module as a separate DLL (it's nice to have but we don't
want to increase the core DLL's size by 25K). This same trick has
been applied to a number of modules, e.g. NumPy and _tkinter.
to Python. Minimal documentation is included in comments at the top
of the file, and in the Misc/PURIFY.README file. Note that this
module must be statically linked since Pure doesn't provide shared
stubs libraries.
(Setup.in): Added commented template for pure module
(syslogmodule.c): ins() function wasn't declared static.
unsigned. This fixes the 8bit-char-in-key platform incompatibility.
I also removed the old backwards compatibility code, and the commented
lisp rotor code. I retained the lisp docstrings as comments preceding
each function.
- Quieted gcc -Wall by removing unused local variables.
- Added some choice parentheses around assignments in conditional
tests.
- Removed an unused (and seemingly unreachable) err label in
load_short_binstring().
- in Unpickler_load(), removed \. in string format.
- init_stuff() was declared to return an int, but had these
problems:
- it was returning NULL instead of 0 or 1 in some cases
- it was falling of the end of the routine without returning
anything
- the call of init_stuff() in initcPickle() was never checking
the return value anyway.
I changed all this by returning 1 in the case of errors, 0 when
no error occurred. Then in initcPickle(), if init_stuff()
returns non-zero, I call Py_FatalError().
Suppressing my urge to reformat according to Python coding standards!
:-)
- 'delete' is a C++ keyword; use 'del_table' instead
- apply Py_CHARMASK() to del_table[i] before using it as an index
*** this fixes a bug that was just reported on the list ***
- if the translation didn't make any changes, INCREF and return
the original string
- when del_table is empty or omitted, don't copy the translation
table to a table of ints (should be a bit faster)
Rewrote maketrans() to avoid copying the table (2-3% faster).
performance hit. Urg. Reverted.
strop_joinfields(): re-instate optimizations for lists and tuples, but
support arbitrary other kinds of sequences as well.
- split_whitespace(): slightly better memory ref handling when errors
occur.
- strop_joinfields(): First argument can now be any sequence-protocol
conformant object.
- strop_find(), strop_rfind(): Use PyArg_ParseTuple for optional
arguments
- strop_lower(), strop_upper(): Factor logic into a common function
do_casechange().
- strop_atoi(), strop_atol(): Use PyArg_ParseTuple.
- strop_maketrans(): arguments used to be optional, although the
documentation doesn't reflect this. Make the source conform to the
docs. Arguments are required, but two empty strings will return the
identity translation table.
- General pass fixing up formatting, and checking for return values.
int/long types, and use the new PyLong_FromUnsignedLong() and
PyLong_AsUnsignedLong() interfaces instead.
Semantic change: the 'I' format will now always return a long int.
- Conform to standard Python C coding styles.
- All static symbols were renamed and shorted.
- Eyeballed all return values and memory references.
- Fixed a bug in signal.pause() so that exceptions raised in signal
handlers are now properly caught after pause() returns.
- Removed SIGCPU and SIGFSZ. We surmise that these were typos for the
previously missing SIGXCPU and SIGXFSZ.
-- The whole implementation is now more table-driven.
-- Unsigned integers. Format characters 'B', 'H', 'I' and 'L'
mean unsigned byte, short, int and long. For 'I' and 'L', the return
value is a Python long integer if a Python plain integer can't
represent the required range (note: this is dependent on the size of
the relevant C types only, not of the sign of the actual value).
-- A new format character 's' packs/unpacks a string. When given a
count prefix, this is the size of the string, not a repeat count like
for the other format characters; e.g. '10s' means a single 10-byte
string, while '10c' means 10 characters. For packing, the string is
truncated or padded with null bytes as appropriate to make it fit.
For unpacking, the resulting string always has exactly the specified
number of bytes. As a special case, '0s' means a single, empty
string (while '0c' means 0 characters).
-- Various byte order options. The first character of the format
string determines the byte order, size and alignment, as follows:
First character Byte order size and alignment
'@' native native
'=' native standard
'<' little-endian standard
'>' big-endian standard
'!' network (= big-endian) standard
If the first character is not one of these, '@' is assumed.
Native byte order is big-endian or little-endian, depending on the
host system (e.g. Motorola and Sun are big-endian; Intel and DEC are
little-endian).
Native size and alignment are determined using the C compiler's sizeof
expression. This is always combined with native byte order.
Standard size and alignment are as follows: no alignment is required
for any type (so you have to use pad bytes); short is 2 bytes; int and
long are 4 bytes. In this mode, there is no support for float and
double.
Note the difference between '@' and '=': both use native byte order,
but the size and alignment of the latter is standardized.
The form '!' is available for those poor souls who can't remember
whether network byte order is big-endian or little-endian.
There is no way to indicate non-native byte order (i.e. force
byte-swapping); use the appropriate choice of '<' or '>'.
non-checked error return values, and where appropriate,
PyArg_ParseTuple() style argument parsing.
I also changed some function names and converted all malloc/free calls
to PyMem_NEW/PyMem_DEL.
Some stylistic changes and formatting standardization.
- Where optional arguments were being used, converted to
PyArg_ParseTuple() style instead of nested PyArg_Parse() style.
- Check for and handle many potential error conditions that were never
being tested.
- internal reg_* functions renamed to regobj_* (makes it easier to
figure out which are global regex functions and which are for regex
objects).
- reg_group (now regobj_group) was quite extensively reworked. it no
longer recurses to do its job (by factoring core functionality into
a separate function that knows about string and integer indexes).
- some minor formatting fixes.
- regex_set_syntax() now invalidates the cache. Without this change
(in the example below), the second search would produce different
output depending on whether the first search were performed or not
(since performing the first search would cache the compiled object
with RE_SYNTAX_EMACS, causing the second test to unexpectedly fail).
regex.search('(a+)|(b+)', 'cdb')
prev = regex.set_syntax(RE_SYNTAX_AWK)
regex.search('(a+)|(b+)', 'cdb')
This is needed because if a configure option has " as its value,
it will be rendered as {"}; after stripping one level of quoting it's
just ", on which splitlist will barf.
SIGCPU and SIGFSZ but we're (Jeremy and I) are actually unsure whether
these were typos or if there are systems that use these alternate
names. We've checked Solaris, SunOS, and IRIX; they contain only the
SIGX* names.
Guido van Rossum
Ken Manheimer
Barry Warsaw
Fred Drake
Jeremy Hylton
Roger E. Masse