Withdraw the change that Fred just checked in -- it was a poorly
documented feature, not a bug, to ignore I/O errors in read().
The new docstring explains the reason for the feature:
"""
this is designed so that you can specifiy a list of potential
configuration file locations (e.g. current directory, user's home
directory, systemwide directory), and all existing configuration files
in the list will be read.
"""
Also add a lower-level function, readfp(), which takes an open file
object (and optionally a filename).
XXX There are some other problems with this module, but I don't have
time to dig into these; in particular, there are complaints that the
%(name)s substitution from the [DEFAULTS] section doesn't work
correctly.
global options table.
Every Command instance now has its own copies of the global options,
which automatically fallback to the Distribution instance. Changes:
- initialize them in constructor
- added '__getattr__()' to handle the fallback logic
- changed every 'self.distribution.{verbose,dry_run}' in Command to
'self.{verbose,dry_run}'.
- filesystem utility methods ('copy_file()' et al) don't take 'update'
parameter anymore -- instead we pass 'not force' to the underlying
function as 'update'
Changed parsing of command line so that global options apply to all
commands as well -- that's how (eg.) Command.verbose will be initialized.
Simplified 'make_file()' to use 'newer_group()' (from util module).
Deleted some cruft.
Some docstring tweaks.
the code a bit and should make it work under Windows even with trailing
backslash.
Fixed a couple of docstrings.
Added comment about 'make_file()' possibly being redundant and unnecessary.
- change how we call it
- added methods 'library_dir_option()', 'library_option()', and
'find_library_file()' that it calls
Added 'force' flag; it's automatically "respected", because this class
always rebuilds everything! (Which it to say, "force=0" is not respected.)
Catch up with changes in 'gen_lib_options()':
- change how we call it
- added methods 'library_dir_option()', 'library_option()', and
'find_library_file()' that it calls
Added 'force' flag and changed compile/link methods to respect it.
has a directory component, then we only search for the library in
that one directory, ie. ignore the 'library_dirs' lists for that
one library.
Changed calling convention to 'gen_lib_options()' again: now, it takes
a CCompiler instance and calls methods on it instead of taking
format strings. Also implemented the new "library name" semantics
using the 'find_library_file()' method in the CCompiler instance.
Added 'force' flag to CCompiler; added to constructor and 'new_compiler()'.
Added 'warn()' method.
Catch up with renamed 'platdir' -> 'build_platlib' option in 'build'.
Don't call 'set_final_options()' in 'run()' anymore -- that's now
guaranteed to be taken care of for us by the Distribution instance.
If 'include_dirs' is a string, split it on os.pathsep (this is half-
hearted -- support for setting compile/link options on the command
line is totally lame and probably won't work at all).
Added 'get_source_files()' for use by 'dist' command.
Added code to 'build_extensions()' to figure out the "def file" to use
with MSVC++ and add it to the linker command line as an "extra_postarg".
Don't call 'set_final_options()' in 'run()' anymore -- that's now
guaranteed to be taken care of for us by the Distribution instance.
Rearranged to bit to allow outsiders (specifically, the 'dist' command)
to find out what modules we would build:
- 'find_modules()' renamed to 'find_package_modules()'
- most of 'build_modules()' abstracted out to 'find_modules()'
- added 'get_source_files()' (for the 'dist' command to use)
- drastically simplified 'build_modules()' -- now just a wrapper around
'find_modules()' and 'build_module()'
Tweaked some help strings to be consistent with documentation.
Don't call 'set_final_options()' in 'run()' anymore -- that's now
guaranteed to be taken care of for us by the Distribution instance.
- add 'extra_preargs' and 'extra_postargs' parameters (and use them!)
- got rid of 'build_info' kludge parameter
- added 'compiler_type' class attribute
- respect reordered arguments to 'gen_lib_options()'
Also added 'output_dir' parameter (catching up with older change in
CCompiler) -- BUT this is presently ignored by all methods!
Deleted some more docstrings redundant with CCompiler.
Dropped generated of "/DEF:" argument --- that's now done by
the 'build_ext' command.
- add 'extra_preargs' and 'extra_postargs' parameters (and use them!)
- added 'compiler_type' class attribute
- respect reordered arguments to 'gen_lib_options()'
which allowed us to get rid of the 'build_info' used in some places
(a temporary kludge to support MSVC++ "def" files).
Deleted big comment whining about that kludge.
Added 'compiler_type' class attribute.
Overhauled 'new_compiler()': now takes 'compiler' argument along with
'plat' (both optional with sensible defaults), and looks them both up
in the new 'default_compiler' and 'compiler_class' dictionaries to
figure out where to get the concrete compiler class from.
Reordered arguments to 'gen_lib_options()' to match the order in
which the arguments are generated (ie. -L before -l).
they make sure that 'set_final_options()' has been called, but isn't
called redundantly.
Changed Distribution to call 'ensure_ready()' where it used to call
'set_final_options()', and in a few extra places as well.
Lots of comment/docstring revisions and additions in both classes.
New one-liner utility methods in Command: 'find_peer()', 'spawn()'.
'alias_options' table and getting rid of some hairy code in the
Distribution constructor.
Resurrected the distribution options that describe the modules present
in the module distribution ('py_modules', 'ext_modules'), and added
a bunch more: 'packages', 'package_dir', 'ext_package', 'include_dirs',
'install_path'.
Updated some comments.
Added 'warn()' method to Command.
'Command.get_command_name()' now stores generated command name in
self.command_name.
Added global cache PATH_CREATED used by 'mkpath()' to ensure it doesn't
try to create the same path more than once in a session (and, more
importantly, to ensure that it doesn't print "creating X" more than
once for each X per session!).
distributions their own directory (and .pth file).
Overhauled how we determine installation directories in
'set_final_options()' to separate platform-dependence and take
'install_path' option into account.
Added 'create_path_file()' to create path config file when 'install_path'
given.
Only run 'install_py' and 'install_ext' when, respectively, there are
some pure Python modules and some extension modules in the distribution.
- rename 'dir' to 'build_dir'
- take 'package' from distribution option 'ext_package'
- take 'extensions' from distribution option 'ext_modules'
- take 'include_dirs' from distribution
Name keyword args explictly when calling CCompiler methods.
Overhauled how we generate extension filenames (in 'extension_filename()
and 'build_extension()') to take 'package' option into account.
It breaks Mailman, it was actually documented in the docstring, so it
was an intentional deviation from the usual del semantics. Let's
document the original behavior in Doc/lib/librfc822.tex.
Changed those two methods to only compile/link if necessary (according
to simplistic timestamp checks).
Added 'output_dir' to 'object_filenames()' and 'shared_object_filename()'.
much breakage (esp. in JPython which holds absolute path names in
co_filename already). This implementation uses os.path.abspath() as a
slightly better way to canonicalize path names. It implements a
cache.
the file that a function is defined on. Non-portable to Windows and
JPython. Instead, new find_function() uses re module on a similar
(simple-minded) pattern.
attributes, etc. Biggest change was to the Distribution constructor
-- it now looks for an 'options' attribute, which contains values
(options) that are explicitly farmed out to the commands. Also,
certain options supplied to Distribution (ie. in the 'setup()' call in
setup.py) are now "command option aliases", meaning they are dropped
right into a certain command rather than being distribution options.
This is handled by a new Distribution class attribute,
'alias_options'.
Various comment changes to reflect the new way-of-thinking.
Added 'get_command_name()' method to Command -- was assuming its
existence all along as 'command_name()', so changed the code that
needs it to call 'get_command_name()'.
now provided (minus the leading underscore) by the ccompiler module.
Fix 'compile()' to return the list of object files generated.
Cosmetic tweaks/delete cruft.
Added big comment about the kludginess of passing 'build_options'
to the link methods and how to fix it.
Added 'gen_preprocess_options()' and 'gen_lib_options()' convenience
functions -- the two cases are very similar for Unix C Compilers and
VC++, so I figured I might as well unify the implementations.
- fix some broken abstract methods
- kludge: add 'build_info' parameter to link methods
- add 'object_name()' and 'shared_library_name()'
- support for MSVCCompiler class on NT/Win95
"""
Added some optional arguments to the XMLParser __init__ method to
specify that selected non-standard constructs are to be accepted.
Also removed the documentation for handle_entityrefs since it isn't
used.
"""
The version is incremented to 0.3.
"""
Extended chunk so that it can also handle formats that are almost
according to EA IFF 85. In particular, added options to handle
little-endian and to handle formats that include the header size in
the chunk size value.
Fixed a bug where the header size was included in the chunk size, which
it isn't according to EA IFF 85.
Added a new method getsize() to get the size of the chunk (excluding
header).
Fixed chunk documentation (TIFF doesn't look like it uses chunks).
Converted wave to use chunk. Wave uses EA IFF 85 chunks except that
it uses little-endian encoding of integer data.
Removed __del__ methods from aifc and wave since I got an
AttributeError there upon exit.
"""
and Toplevel class constructors. This means that if the window
manager closes the window, the Python-side Tkinter data structures
will be destroyed correctly. (Most apps do this anyway, and it's
recommended practice; I see no reason why making it the default
behavior could be bad.)
Added 'verbose' and 'dry_run' parameters to constructor.
Changed 'compile()', 'link_*()' to default lists arguments to None
rather than empty list.
Added implementations of the filename-mangling methods mandated by
the CCompiler interface.
'new_compiler()' factory function.
Added 'runtime_library_dirs' list (for -R linker option) and methods
to manipulate it.
Deleted some obsolete comments.
Added all the filename manglign methods: 'object_filenames()',
'shared_object_filename()', 'library_filename()',
'shared_library_filename()'.
Added 'spawn()' method (front end to the "real" spawn).
- did away with 'comment_re' option -- it's just not that simple anymore
- heavily revised the main logic in 'readline()' to accomodate this
Beefed up 'warn()': 'line' can be list or tuple, and 'msg' is
automatically converted to a string.
"""
If the filename being complained about contains a space, enclose the
file-name in quotes.
The reason is simply that when I try and parse tabnanny's output, filenames
with spaces make it very difficult to determine where the filename stops
and the linenumber begins!
"""
Tim approves.
I slightly changed the patch (use 'in' instead of string.find()) and
arbitrarily bumped the __version__ variable up to 6.
bindings to a dictionary _tagcommands which was otherwise unused.
(This was checked in accidentally with rev. 1.125 and not deleted with
rev. 1.127 when the other half of this code was removed -- although
even as originally checked in the _tagcommands variable was never
used.)
(PR#40, reported by Peter Stoehr)
"""
Here's a patch for the ForkingMixIn which will prevent the server from
forking itself into the ground. Note: I've tested a very similar patch
(subclassed ForkingMixIn) but not actually tested this one. As you might
surmise, this was done out of necessity...
If the maximum number of children are already running, block while waiting
for a child to exit.
"""
(I added that last sentence as a comment to the code --GvR.)
for gotonext() pushing self.pos past the end of the string. This can
happen if the message has a To field like "To: :" and you call
msg.getaddrlist('to').
In splithost, accept empty host part in URLs. This is required for
file URLs that can have an empty host part. For such URLs, we should
not return the initial 2 slashes as part of the file name.
1. Fix incorrect file open mode on Win32 platforms (use "rb" instead
of "r").
2. Add shallow parameter to cmp.cmp(). If false, deep file
comparisons are made.
The module should be 100 percent backwards compatible.
Urllib makes the URL of the opened file available through the geturl
method of the returned object. For local files, this consists of
file: plus the name of the file. This results in an invalid URL if
the file name was relative. This patch fixes this so that the
returned URL is just a relative URL in that case. When the file name
is absolute, the URL returned is of the form file:///absolute/path.
[I guess that a URL of the form "file:foo.html" is illegal... GvR]
Introduce a new builtin exception, UnboundLocalError, raised when ceval.c
tries to retrieve or delete a local name that isn't bound to a value.
Currently raises NameError, which makes this behavior a FAQ since the same
error is raised for "missing" global names too: when the user has a global
of the same name as the unbound local, NameError makes no sense to them.
Even in the absence of shadowing, knowing whether a bogus name is local or
global is a real aid to quick understanding.
Example:
D:\src\PCbuild>type local.py
x = 42
def f():
print x
x = 13
return x
f()
D:\src\PCbuild>python local.py
Traceback (innermost last):
File "local.py", line 8, in ?
f()
File "local.py", line 4, in f
print x
UnboundLocalError: x
D:\src\PCbuild>
Note that UnboundLocalError is a subclass of NameError, for compatibility
with existing class-exception code that may be trying to catch this as a
NameError. Unfortunately, I see no way to make this wholly compatible
with -X (see comments in bltinmodule.c): under -X, [UnboundLocalError
is an alias for NameError --GvR].
[The ceval.c patch differs slightly from the second version that Tim
submitted; I decided not to raise UnboundLocalError for DELETE_NAME,
only for DELETE_LOCAL. DELETE_NAME is only generated at the module
level, and since at that level a NameError is raised for referencing
an undefined name, it should also be raised for deleting one.]
always lowercasing the option name, call a method optionxform() which
can be overridden. Also make the regexps SECTRE and OPTRE non-private
variables so they can also be overridden.
mode attribute of the file object (if it has one), otherwise
use 'rb'.
The documentation should still show this as required until
there's a new release.
Added support for unseekable files.
(I use unqualified excepts since we don't know why the seek/tell might
fail. In my case it was because of an AttributeError.)
Problem: rfc822.py in 1.5.2 final loses the quotes around
quoted local-part names.
The fix is to preserve the quotes around a local-part
name in an address.
Test:
import rfc822
a = rfc822.AddrlistClass('(Comment stuff) "Quoted
name"@somewhere.com')
a.getaddrlist()
The correct result is:
[('Comment stuff', '"Quoted name"@somewhere.com')]
Hammond: record top-level functions (as Function instances, a simple
subclass of Class). You must use the new interface readmodule_ex() to
get these, though.
__init__.py it isn't read. (Sjoerd just came up with this, so it's
not heavily tested.)
Other (yet unsolved) package problems noted by Sjoerd:
- If you have a package and a module inside that or another package
with the same name, module caching doesn't work properly since the
key is the base name of the module/package.
- The only entry that is returned when you readmodule a package is a
__path__ whose value is a list which confuses certain class browsers
that I wrote. (Hm, this could be construed as a feature.)
I've found two places where smtplib.py sends an extra trailing space
on command lines to the SMTP server. I don't know if this ever causes
any problems, but I'd prefer to be on the safe side. The enclosed
patch removes the extra space.
all processing instruction target names containing 'xml' were
rejected, instead (as the standard rejects) only the name 'xml' itself
(or case variants thereof).
I guess in 1.5.2 a new module, whichdb, was added that attempts to
divine the nature of a database file. This module doesn't know anything
about Berkeley DB v2 files. In v2, Sleepycat added a 12-byte null pad
in front of the old magic numbers (at least for hash and btree files).
I've been using v2 for awhile and upgrading to 1.5.2 broke all my
anydbm.open calls. I believe the following patch corrects the problem.
appreciably. Triple-quoted strings no longer confuse it, nor nested
classes or defs, nor comments starting in column 1. Chews thru
Tkinter.py in < 3 seconds for me; doctest.py no longer confuses it; no
longer missing methods in PyShell.py; etc. Also captures defs
starting in column 1 now, but ignores them; an interface should be
added so that IDLE's class browser can show the top-level functions
too.
'install_site_lib' and install_site_platlib' on non-POSIX platforms.
Should at least work for NT, as this is adopted from Amos Latteier's NT
patches. Also added extensive comments bitching about the inadequacy of
the current model, both under POSIX and NT (and probably other) systems.
'run_command()' to refer to it before attempting to run a command --
that way, command classes can freely invoke other commands without fear
of duplicate execution.
Beefed up some comments and docstrings.
It wasn't hard to speed pyclbr by a factor of 3, and I'll attach an
experimental patch for that (experimental because barely tested). Uncomment
the new "String" stuff and it will deal with strings correctly (pyclbr
currently ignores the possibility), but that slows it down a lot. Still
faster in the end than current pyclbr, but-- frankly --I'd rather have the
dramatic speedup!
content-type to application/x-www-form-urlencoded only when the method
is POST. Ditto for when the content-type is unrecognized -- only
fall back to urlencoded with POST.
I noticed while watching (with lsof) my forking SocketServer app running
that I would get multiple processes listening to the socket. For the most
part, this doesn't hurt things, but if you terminate the server, this can
prevent it from restarting because it cannot bind to the port due to any
running children which also have the socket open. The following one-liner
fixes this.
meaningful return values: respectively, whether the copy was done, and
the list of files that were copied. This meant some trivial changes in
core.py as well: the Command methods that mirror 'copy_file()' and
'copy_tree()' have to pass on their return values.
of the 'install_py' command rather than 'build_py'. Obviously, this
meant that the 'build_py' and 'install_py' modules had to change; less
obviously, so did 'install' and 'build', since these higher-level
commands must make options available to control the lower-level
commands, and some compilation-related options had to migrate with the
code.
(1) Fix reference to pwd.error to be KeyError -- there is no pwd.error
and pwd.getpwnam() raises KeyError on failure.
(2) Add cookie support, by placing the 'Cookie:' header, if present,
in the HTTP_COOKIE environment variable.
Two problems: The SMTPRecipientsRefused class should not inherit
SMTPResponseException, since it doesn't provide the smtp_code and
smtp_error attributes. My patch for not adding an extra CRLF was
apparently forgotten. The enclosed patch fixes these two problems.
terminated; this makes the final assert in the self-test code fail if
the parent runs faster than the children. Fix this by calling wait()
on the remaining children instead.
1. Jack Jansen reports that on the Mac, the time may be negative, and
solves this by adding a write32u() function that writes an unsigned
long.
2. On 64-bit platforms the CRC comparison fails; I've fixed this by
casting both values to be compared to "unsigned long" i.e. modulo
0x100000000L.
unsupported format string. (I guess this is because the logic for
deciding whether to reallocate the buffer or not has been improved.)
This caused the test code to crash on result[0]. Fix this by assuming
an empty result also means the format is not supported.
than was worth it: when deleting a canvas item, it would try to
automatically delete the bindings for that item. Since there's
nothing that says you can't reuse the tag and still have the bindings,
this is not correct. Also, it broke at least one demo
(Demo/tkinter/matt/rubber-band-box-demo-1.py).
so the preferred name for them is tag_lower, tag_raise
(similar to tag_bind, and similar to the Text widget);
unfortunately can't delete the old ones yet (maybe in 1.6)
Per writes:
"""
The application where Signum Support uses smtplib needs to be able to
report good error messages to the user when sending email fails. To
help in diagnosing problems it is useful to be able to report the
entire message sent by the server, not only the SMTP error code of the
offending command.
A lot of the functions in sendmail.py unfortunately discards the
message, leaving only the code. The enclosed patch fixes that
problem.
The enclosed patch also introduces a base class for exceptions that
include an SMTP error code and error message, and make the code and
message available on separate attributes, so that surrounding code can
deal with them in whatever way it sees fit. I've also added some
documentation to the exception classes.
The constructor will now raise an exception if it cannot connect to
the SMTP server.
The data() method will raise an SMTPDataError if it doesn't receive
the expected 354 code in the middle of the exchange.
According to section 5.2.10 of RFC 1123 a smtp client must accept "any
text, including no text at all" after the error code. If the response
of a HELO command contains no text self.helo_resp will be set to the
empty string (""). The patch fixes the test in the sendmail() method
so that helo_resp is tested against None; if it has the empty string
as value the sendmail() method would invoke the helo() method again.
The code no longer accepts a -1 reply from the ehlo() method in
sendmail().
[Text about removing SMTPRecipientsRefused deleted --GvR]
"""
and also:
"""
smtplib.py appends an extra blank line to the outgoing mail if the
`msg' argument to the sendmail method already contains a trailing
newline. This patch should fix the problem.
"""
The Dragon writes:
"""
Mostly I just re-added the SMTPRecipientsRefused exception
(the exeption object now has the appropriate info in it ) [Per had
removed this in his patch --GvR] and tweaked the behavior of the
sendmail method whence it throws the newly added SMTPHeloException (it
was closing the connection, which it shouldn't. whatever catches the
exception should do that. )
I pondered the change of the return values to tuples all around,
and after some thinking I decided that regularizing the return values was
too much of the Right Thing (tm) to not do.
My one concern is that code expecting an integer & getting a tuple
may fail silently.
(i.e. if it's doing :
x.somemethod() >= 400:
expecting an integer, the expression will always be true if it gets a
tuple instead. )
However, most smtplib code I've seen only really uses the
sendmail() method, so this wouldn't bother it. Usually code I've seen
that calls the other methods usually only calls helo() and ehlo() for
doing ESMTP, a feature which was not in the smtplib included with 1.5.1,
and thus I would think not much code uses it yet.
"""
splitunc() parses UNC paths. The contributor of the UNC parsing in
splitdrive() doesn't like it, but I haven't heard a good reason to
keep it, and it causes some problems. (I think there's a
philosophical problem -- to me, the split*() functions are purely
syntactical, and the fact that \\foo is not a valid path doesn't mean
that it shouldn't be considered an absolute path.)
Also (quite separately, but strangely related to the philosophical
issue above) fix abspath() so that if win32api exists, it doesn't fail
when the path doesn't actually exist -- if GetFullPathName() fails,
fall back on the old strategy (join with getcwd() if neccessary, and
then use normpath()).
and dry-run flags consistently painless): 'execute()', 'mkpath()',
'copy_file()', 'copy_tree()', 'make_file()', and stub for 'make_files()'
(not sure yet if it's useful).
that wrap them in the Command class).
Fixed 'copy_file()' to use '_copy_file_contents()', not 'copyfile()'
from shutil module -- no reference to shutil anymore.
Added "not copying" announcement in 'copy_file()'.
Wee comment fix.
If you send something like "PUT / HTTP/1.0" to something derived from
BaseHTTPServer that doesn't define do_PUT, you will get a response
that begins like this:
HTTP/1.0 501 Unsupported method ('do_PUT')
Server: SimpleHTTP/0.3 Python/1.5
Date: Tue, 30 Mar 1999 18:53:53 GMT
The server should complain about 'PUT' instead of 'do_PUT'. This
patch should fix the problem.
"""
- It needlessly used the makefile() method for each response that is
read from the SMTP server.
- If the remote SMTP server closes the connection unexpectedly the
code raised an IndexError. It now raises an SMTPServerDisconnected
exception instead.
- The code now checks that all lines in a multiline response actually
contains an error code.
"""
The Dragon approves.
allow using the 'a' flag as a mode for opening a GzipFile. gzip
files, surprisingly enough, can be concatenated and then decompressed;
the effect is to concatenate the two chunks of data.
If we support it on writing, it should also be supported on reading.
This *wasn't* trivial, and required rearranging the code in the
reading path, particularly the _read() method.
Raise IOError instead of RuntimeError in two cases, 'Not a gzipped file'
and 'Unknown compression method'
If a filename on Windows starts with \\, it is converted to a URL
which starts with ////. If this URL is passed to urlparse.urlparse
you get a path that starts with // (and an empty netloc). If you pass
the result back to urlparse.urlunparse, you get a URL that starts with
//, which is parsed differently by urlparse.urlparse. The fix is to
add the (empty) netloc with accompanying slashes if the path in
urlunparse starts with //. Do this for all schemes that use a netloc.
Pathnames of files on other hosts in the same domain
(\\host\path\to\file) are not translated correctly to URLs and back.
The URL should be something like file:////host/path/to/file.
Note that a combination of drive letter and remote host is not
possible.
netloc from the base url as the default netloc for the resulting url
even if the schemes differ.
Once upon a time, when the web was wild, this was a valuable hack
because some people had a URL referencing an ftp server colocated with
an http server without having the host in the ftp URL (so they could
replicate it or change the hostname easily).
More recently, after the file: scheme got added back to the list of
schemes that accept a netloc, it turns out that this caused weirdness
when joining an http: URL with a file: URL -- the resulting file: URL
would always inherit the host from the http: URL because the file:
scheme supports a netloc but in practice never has one.
There are two reasons to get rid of the old, once-valuable hack,
instead of removing the file: scheme from the uses_netloc list. One,
the RFC says that file: uses the netloc syntax, and does not endorse
the old hack. Two, neither netscape 4.5 nor IE 4.0 support the old
hack.
Guido van Rossum
Fred Drake
Greg Ward
Barry Warsaw
Andrew M. Kuchling
Jeremy Hylton