This'll work fine with 2.0 or 1.5.2, but is less than ideal for
1.6a1/a2. But the code to accomodate 1.6a1/a2 was released with
Distutils 0.9, so it can go away now.
allows the caller to execute the various tests in pseudo-random order -
default is still to execute tests in the order returned by findtests().
* moved initialization of the various flag variables to the main() function
definition, making it possible to execute regrtest.main() interactively
and still override default behavior.
-- added pickling support (only works if sre is imported)
-- fixed wordsize problems in engine
(instead of casting literals down to the character size,
cast characters up to the literal size (same as the code
word size). this prevents false hits when you're matching
a unicode pattern against an 8-bit string. (unfortunately,
this broke another test, but I think the test should be
changed in this case; more on that on python-dev)
-- added sre.purge function
(unofficial, clears the cache)
group reset problem. in the meantime, I added some
optimizations:
- added "inline" directive to LOCAL
(this assumes that AC_C_INLINE does what it's
supposed to do). to compile SRE on a non-unix
platform that doesn't support inline, you have
to add a "#define inline" somewhere...
- added code to generate a SRE_OP_INFO primitive
- added code to do fast prefix search
(enabled by the USE_FAST_SEARCH define; default
is on, in this release)
errors in some of the hash algorithms. For exmaple, in float_hash and
complex_hash a certain part of the value is not included in the hash
calculation. See Tim's, Guido's, and my discussion of this on
python-dev in May under the title "fix float_hash and complex_hash for
64-bit *nix"
(2) The hash algorithms that use pointers (e.g. func_hash, code_hash)
are universally not correct on Win64 (they assume that sizeof(long) ==
sizeof(void*))
As well, this patch significantly cleans up the hash code. It adds the
two function _Py_HashDouble and _PyHash_VoidPtr that the various
hashing routine are changed to use.
These help maintain the hash function invariant: (a==b) =>
(hash(a)==hash(b))) I have added Lib/test/test_hash.py and
Lib/test/output/test_hash to test this for some cases.
get_starttag_text(): New method.
Return the text of the most recently parsed start tag, from
the '<' to the '>' or '/'. Not really useful for structure
processing, but requested for Web-related use. May also be
useful for being able to re-generate the input from the parse
events, but there's no equivalent for end tags.
attrfind: Be a little more forgiving of unquoted attribute values.
(those semantics are weird...)
- got rid of $Id$'s (for the moment, at least). in other
words, there should be no more "empty" checkins.
- internal: some minor cleanups.
(test_sre still complains about split, but that's caused by
the group reset bug, not split itself)
- added more mark slots
(should be dynamically allocated, but 100 is better than 32.
and checking for the upper limit is better than overwriting
the memory ;-)
- internal: renamed the cursor helper class
- internal: removed some bloat from sre_compile
accidentally wiped out by Ping's patch (which shouldn't have affected
this file at all, had Ping done a cvs update).
This checkin restores Gordon's version, with Fredrik's change merged
back in.
tests in sre_patch back to previous version
- fixed return value from findall
- renamed a bunch of functions inside _sre (way too
many leading underscores...)
</F>
Changed 'prune_file_list()' so it also prunes out RCS and CVS directories.
Added 'is_regex' parameter to 'select_pattern()', 'exclude_pattern()',
and 'translate_pattern()', so that you don't have to be constrained
by the simple shell-glob-like pattern language, and can escape into
full-blown regexes when needed. Currently this is only available
in code -- it's not exposed in the manifest template mini-language.
Added 'prune' option (controlled by --prune and --no-prune) to determine
whether we call 'prune_file_list()' or not -- it's true by default.
Fixed 'negative_opt' -- it was misnamed and not being seen by dist.py.
Added --no-defaults to the option table, so it's seen by FancyGetopt.
Testing: test_array.py was also extended to check that one can set the
full range of values for each of the integral signed and unsigned
array types.
This closes SourceForge patch #100506.
This patch adds the openpty() and forkpty() library calls to posixmodule.c,
when they are available on the target
system. (glibc-2.1-based Linux systems, FreeBSD and BSDI at least, probably
the other BSD-based systems as well.)
Lib/pty.py is also rewritten to use openpty when available, but falls
back to the old SGI method or the "manual" BSD open-a-pty
code. Openpty() is necessary to use the Unix98 ptys under Linux 2.2,
or when using non-standard tty names under (at least) BSDI, which is
why I needed it, myself ;-) forkpty() is included for symmetry.
<skip@mojam.com>. Revisions to the markup to make it pass LaTeX, added
an index entry and a reference from the sys.exitfunc documentation.
This closes SourceForge patch #100620.
methods (but not 'link_executable()', hmmm). Currently only used by
BCPPCompiler; it's a dummy parameter for UnixCCompiler and MSVCCompiler.
Also added 'bcpp' to compiler table used by 'new_compiler()'.
Two major points:
* lots of overlap with MSVCCompiler; the common code really should be
factored out into a base class, say WindowsCCompiler
* it doesn't work: weird problem spawning the linker (see comment for
details)
to 'msvc_prelink_hack()', adding the parameters that it actually needs,
and only calling it for MSVC compiler objects. Generally gave up on the
idea of a general "hook" mechanism: deleted the empty 'precompile_hook()'.
that a particular compiler system depends on. This consists of the
'set_executables()' and 'set_executable()' methods, and a few lines in
the constructor that expect implementation classes to provide an
'executables' attribute, which we use to initialize several instance
attributes. The default implementation is somewhat biased in favour of
a Unix/DOS "command-line" view of the world, but it shouldn't be too
hard to override this for operating systems with a more sophisticated
way of representing programs-to-execute.
meant playing along with the new "dictionary of executables" scheme
added to CCompiler by adding the 'executables' class attribute, and
changing all the compile/link/etc. methods to use the new attributes
(which encapsulate both the program to run and its standard arguments,
so it was a *little* bit more than just changing some names).
Unix shell-like syntax (eg. in Python's Makefile, for one thing -- now that
I have this function, I'll probably allow quoted strings in config files too.
the "install_data" command to the installation base, which is usually just
sys.prefix. (Any setup scripts out there that specify data files will have
to set the installation directory, relative to the base, explicitly.)
in the module of the command classes that have command-specific
help options. This lets us keep the principle of lazily importing
the ccompiler module, and also gets away from defining non-methods
at class level.
These two fixes were approved by me.
Peter Kropf:
There's a problem with the xmllib module when used with JPython. Specifically,
the JPython re module has trouble with the () characters in strings passed into
re.compile.
Spiros Papadimitriou:
I just downloaded xmllib.py ver. 0.3 from python.org and there
seems to be a slight typo: Line 654 ("tag = self.stack[-1][0]"
in parse_endtag), is indented one level more than it should be.
I just thought I'd let you know...
major ports of GCC to Windows. Contributed by Rene Liebscher, and quite
untested by me. Apparently requires tweaking Python's installed config.h
and adding a libpython.a to build extensions.
'try_cpp()', 'search_cpp()', and 'check_header()'. This is enough that
the base config is actually useful for implementing a real config
command, specifically one for mxDateTime.
it in UnixCCompiler. Still needs to be implemented in MSVCCompiler (and
whatever other compiler classes are lurking out there, waiting to be
checked in).
this patch adds a fast _flatten function to the _tkinter
module, and imports it from Tkinter.py (if available).
this speeds up canvas operations like create_line and
create_polygon. for example, a create_line with 5000
vertices runs about 50 times faster with this patch in
place.
'remove_tree()' can cooperate with 'mkpath()' in the maintenance of
the PATH_CREATED cache: specifically, if a directory is created
with 'mkpath()', later removed with 'remove_tree()', and 'mkpath()'
is again requested to create it, then it would erroneously think
the directory already existed, because it was in the PATH_CREATED
cache. The patch (slightly tweaked by me) fixes that.
Fixed some tests to not cause the script to fail, but rather
output a warning (which then is caught by regrtest.py as wrong
output). This is needed to make test_unicode.py run through
on JPython.
Thanks to Finn Bock.
template into a new method 'prune_file_list()', called from
'get_file_list()' rather than 'read_manifest()' -- this keeps
'read_manifest()' more general.
Deleted the redundant call to 'exclude_pattern()' in 'make_distribution()'
-- this had the same intention as 'prune_file_list()', but was incomplete
(only pruned the release tree, not the build tree) and in the wrong
place (the prune wouldn't be reflected in the manifest file).
directly printing to stdout. This was a bit more work than it sounds like
it should have been:
* turned 'select_pattern()' and 'exclude_pattern()' from functions into
methods, so they can refer to 'self' to access the method
* commented out the *other* 'exclude_pattern()' method, which appears
to be vestigial code that was never cleaned up when the
'exclude_pattern()' function was created
* changed the one use of the old 'exclude_pattern()' method to use the
new 'exclude_pattern()' (same behaviour, slightly different args)
* some code and docstring reformatting
* and, of course, changed all the debugging prints to 'debug_print()' calls
Added/tweaked some regular ('self.announce()') output for better runtime
feedback.
Added support to set the default encoding of strings
at startup time to the values defined by the C locale.
The sys.setdefaultencoding() API is deleted after having
set up the encoding, so that user code cannot subsequentely
change the setting. This effectively means that only site.py
may alter the default setting.
values that "--foo" can take for various commands: eg. what formats for
"sdist" and "bdist", what compilers for "build_ext" and "build_clib".
I have *not* reviewed this patch; I'm checking it in as-is because it also
fixes a paper-bag-over-head bug in bdist.py, and because I won't have
time to review it properly for several days: so someone else can
test it for me, instead!
Look for personal config file in /home/greg on Windows, too: users will have
to set /home/greg to use this, so it's not something that many people will
use. But if python-dev comes up with the "right way" to divine a
home directory on Windows, we can use that to set /home/greg and poof! --
personal Distutils config files on Windows.
one doesn't *do* anything by default; it's just there as a conduit for data
(eg. include dirs, libraries) from the user to the "build" commands.
However, it provides a couple of Autoconf-ish methods ('try_compile()',
'try_link()', 'try_run()') that derived, per-distribution "config" commands
can use to poke around the target system and see what's available.
Initial experimenst with mxDateTime indicate that higher-level methods are
necessary: analogs of Autoconf's AC_CHECK_HEADER, AC_CHECK_LIB will be
needed too (and that's just to probe the C/C++ system: how to probe the
Python system is wide open, and someday we'll have to worry about probing a
Java system too).
Half-fixed RPM 2 compatibility:added 'rpm_base' option, which must be set
(to eg. /usr/src/redhat on a stock Red Hat system) if rpm2_mode is on.
Still not quite working, though.
Fills in question marks in help
Reads scripts in from files rather than strings
Adds RPM 2 compatibility mode (untested). Use of this mode requires that
--bdist-base be specified because bdist_rpm has no way of detecting where
RPM wants to find spec files and source files. An unmodified RedHat 5.0
system would require '--bdist-base=/usr/src/RedHat'. (You would also have
to be root.) If the rpmrc file has been modified to allow RPMs to be built
by normal users then --build-base would need to be changed accordingly.
Formats the changelog.
GPW: tweaked formatting, added some editorial comments.
prep/build/etc. scripts, doc files, dependency info) from a config file
rather than the dedicated "package_info" file. (The idea is that
developers will provide RPM-specific info in the "[bdist_rpm]" section of
setup.cfg, but of course it could also be supplied in the other config
files, on the command line, or in the setup script -- or any mix of the
above.)
Major changes:
* added a boatload of options to 'user_options' and
'initialize_options()': 'distribution_name', 'group', 'release', ...
* added 'finalize_package_data()', which takes the place of
'_get_package_data()' -- except it's called from 'finalize_options()',
not 'run()', so we have everything figured out before we actually run
the command
* added 'ensure_string()', 'ensure_string_list()', 'ensure_filename()';
these take the place of '_check_string()' and friends. (These actually
look like really useful type-checking methods that could come in handy
all over the Distutils; should consider moving them up to Command and
using them in other command classes' 'finalize_options()' method for
error-checking).
* various cleanup, commentary, and adaptation to the new way of
storing RPM info in '_make_spec_file()'
* added "--bdist-base" option to parameterize where we build
the RPM (comes from "bdist" by default: "build/bdist.<plat>")
* simplified/cleaned up some code in 'run()' in the process of
removing (most) hard-coded directory names
* if "--spec-only", drop spec file in "dist" rather than "redhat"
(directory name still hard-coded, though)
* use 'reinitialize_command()' to fetch the "sdist" object to
tweak before running "sdist" command
* use 'self.copy_file()' method rather than 'copy_file()' function
* cosmetic tweaks to comments, error messages
* help strings start with lowercase
* added affirmative version of '--no-clean' and '--no-rpm-opt-flags',
which are the default (thus the attributes that correspond to
the options are now 'clean' and 'use_rpm_opt_flags')
setup script) to be a list of Extension instances, rather than a list of of
(ext_name, build_info) tuples. This is mostly a simplification, but
'check_extension_list()' got a lot more complicated because of the need to
convert the old-style tuples to Extension instances.
Temporarily dropped support for defining/undefining macros in the
'extensions' list -- I want to change the interface, but haven't yet made
the required changes in CCompiler and friends to support this nicely.
Also neatened up the code that merges 'extra_compile_flags' and the CFLAGS
environment variable.
Added 'reinitialize_command()' -- lets us "push" option values in
a controlled, safe way; this is a small change to the code, but
a big change to the Distutils philosophy of passing option values
around. The preferred mode is still definitely to "pull" options
from another command (eg. "install" fetches the base build directory
from "build"), but it is now feasible to "push" options onto another
command, when you know what's best for it. One possible application
will be a "config" command, which pokes around the system and pushes
values (eg. include and library directories) onto the "build" command.
Added 'dump_option_dicts()' method (for debugging output).
* Command method 'find_peer()' -> 'get_finalized_command()'
* Command method 'run_peer()' -> 'run_command()'
Also deleted the 'get_command_option()' method from Command, and
fixed the one place where it was used (in "bdist_dumb").
directories after all is said and done, so we don't accidentally include
those files in the source distribution.
(This is the quick and easy way to fix this; Andrew says: "Changing
findall() looked like it was going to be messy, so I tried this instead.
The only problem is that redundant directory traversals are being done,
walking through build/ only to throw out all the files found at the end.").
* 'headers' entry added to all the install schemes
* '--install-headers' option added
* 'install_headers' added to 'sub_commands'
* added 'dist_name' to configuration variables (along with a few
others that seem handy: 'dist_version', 'dist_fullname', and
'py_version'
* in 'finalize_unix()', make sure 'install_headers' defined if
user specified 'install_base' and/or 'install_platbase'
* added 'has_headers()'
* a few other small changes
dictionaries in 'self.command_options' to 'get_option_dict()'.
Simplified code in 'parse_config_files()' and 'parse_command_line()'
accordingly.
Fixed code in constructor that processes the 'options' dictionary
from the setup script so it actually works: uses the new
'self.command_options' dictionary rather than creating command
objects and calling 'set_option()' on them.
attempt to verify the bold assertions in the documentation):
* entries for the "root package" in 'package_dir' didn't work --
fixed by improving the fall-through code in 'get_package_dir()'
* __init__.py files weren't installed when modules-in-packages
were listed individually (ie. in 'py_modules' in the setup script);
fixed by making 'check_package()' return the name of the __init__
file if it exists, and making 'find_modules()' add an entry to
the module list for __init__ if applicable
This patch adds a comment about quoting to the doc string,
and also checks that the 'flags' argument to the STORE command
is appropriately enclosed inside parentheses to avoid quoting.
* 'first_line_re' loosened up
* command description improved
* replaced '_copy_files()' and '_adjust_files()' with one method
that does everything, 'copy_scripts()' -- this should be more
efficient than Bastian's version, should behave better in
dry-run mode, and does timestamp dependency-checking
necessary to support it.
Details:
- build command additionally calls build_scripts
- build_scripts builds your scripts in 'build/scripts' and adjusts the
first line if it begins with "#!" and ends with "python", optionally
ending with commandline options (like -O, -t ...). Adjusting means we
write the current path to the Python interpreter in the first line.
- install_scripts copies the scripts to the install_scripts dir
- install_data copies your data_files in install_data. You can
supply individual directories for your data_files:
data_files = ['doc/info.txt', # copy this file in install_scripts dir
('testdata', ['a.dat', 'b.dat']), # copy these files in
# install_scripts/testdata
('/etc', ['packagerc']), # copy this in /etc. When --root is
# given, copy this in rootdir/etc
]
So you can use the --root option with absolute data paths.
often, ftp URLs hang in the final close. Further analysis suggests
that this is because the close hook in addclosehook() calls the hook
before acually closing the connection. The hook, in this case, waits
for the '226 Transfer complete' status from the server on the command
socket. However, more and more ftp servers only send this status when
the data socket has actually been closed -- causing a deadlock.
The fix is simple: in addclosehook.close(), call addbase.close()
*before* calling the closehook.
* 'get_command_obj()' now sets command attributes based on
the 'command_options' dictionary
* some typos fixed
* kludged 'parse_config_files()' to re-initialize the ConfigParser
instance after each file, so we know for sure which config
file each option comes form
* added lots of handy debugging output
command-line parsing code, splitting it up into several methods (new
methods: '_parse_command_opts()', '_show_help()') and making it put options
into the 'command_options' dictionary rather than instantiating command
objects and putting them there.
Lots of other little changes:
* merged 'find_command_class()' and 'create_command_obj()' and
called the result 'get_command_class()'
* renamed 'find_command_obj()' to 'get_command_obj()', and added
command object creation and maintenance of the command object cache to
its responsibilities (taken over from 'create_command_obj()')
* parse config files one-at-a-time, so we can keep track of the
filename for later error reporting
* tweaked some help messages
* fixed up many obsolete comments and docstrings
is no index.htm[l] file, and when it is called, it also spits out the
headers. When an index.htm[l] file is present, the regular (file
access) path is followed. Also, when the guessed content-type matches
text/*, open the file in text mode; otherwise in binary mode.
objects, it now has method names.
Added three methods, 'has_lib()', 'has_scripts()', and 'has_data()'
to determine if we need to run each of the three possible sub-commands.
Added 'get_sub_commands()' to take care of finding the methods named
in 'sub_commands', running them, and interpreting the results to
build a list of sub-commands that actually have to be run.
