Tim Kientzle [Sun, 15 Jun 2008 03:20:28 +0000 (23:20 -0400)]
Back out change 143007; go back to using platform getopt_long
if it exists.
(I intend to re-add this getopt_long implementation at a later
date, but I'm trying to stabilize the libarchive 2.5 release
and this is a bit disruptive.)
Tim Kientzle [Sat, 14 Jun 2008 21:27:26 +0000 (17:27 -0400)]
Fix the link resolver to never match dirs as hardlinks.
In particular, this seems to fix some recent port build failures
(the use of tar in these ports is rather bizarre, but it did work
with tar's old link-matching code).
Thanks to: Kris Kennaway
Tim Kientzle [Sat, 14 Jun 2008 15:11:37 +0000 (11:11 -0400)]
On some platforms, the GID bit is inherited by subdirectories.
This caused a lot of false failures in the libarchive test
harness. Be a little more careful: the GID bit on created
directories should only be checked if ARCHIVE_EXTRACT_PERM
was used when the dir was created.
Submitted by: Carey Evans
Tim Kientzle [Sat, 14 Jun 2008 14:45:42 +0000 (10:45 -0400)]
Flush stdout after printing the name of the test;
this ensures that any errors from that test appear after the
name of the test. <sigh>
Submitted by: Carey Evans
Tim Kientzle [Sat, 14 Jun 2008 14:36:30 +0000 (10:36 -0400)]
Use wctomb(), which is present in C90, instead of wcrtomb(),
which wasn't introduced until C99. In particular, this allows
the new locale support to work on FreeBSD 4, which doesn't
implement C99 wide character functions.
Include an autoconf check for wctomb(); if it's not present at
all, use the built-in UTF-8 functions instead so that pax
extended headers (which are stored in UTF-8) can still be minimally
supported.
Tim Kientzle [Sat, 7 Jun 2008 18:23:50 +0000 (14:23 -0400)]
Somewhere I got the idea that wcrtomb(NULL, ...) cleared the
provided shift state. This is wrong; memset() seems to be
the correct portable way to clear the shift state.
Submitted by: NAKAJIMA Michihiro
MFP4 after: 3 days
Tim Kientzle [Thu, 5 Jun 2008 22:41:21 +0000 (18:41 -0400)]
Use Gregory Pietsch's public domain getopt_long() implementation
instead of relying on the platform one.
Having the tar program use getopt() on platforms that didn't
support getopt_long() was a nice idea, but it's turning out to
be a real headache. This way, I can just assume long option
support and simplify chunks of the test harness.
Tim Kientzle [Wed, 28 May 2008 22:33:11 +0000 (18:33 -0400)]
Mis-integrate; the p4 version doesn't need this, since archive.h isn't
a built file.
Code in CVS needs this until archive.h.in gets renamed to archive.h.
Between all the false cases found by Coverity, one is actual
relevant: after closing child_stdin, set it to -1, keeping
child_stdout alone and not the other way around.
Some systems don't have vfork, only fork. It is good enough for
our purposes, so use either.
Include archive.h if the program support is not present to fix
compilation.
Be more persistent when trying to get blocking behavior back for
the pipes. If the child switched the descriptor back to
non-blocking, set it back to blocking and call poll/select.
Make the poll/select code filter out -1 as descriptor for this
purpose.
Move the opening of the file to write down into
write_entry_backend, if it isn't already open.
Use archive_entry_sourcepath for this purpose and allow
writing deferred entries for newc format that way.
Tim Kientzle [Wed, 21 May 2008 20:53:40 +0000 (16:53 -0400)]
Follow-on to change 141950: Enforce the non-zero minimum BEFORE
checking to ensure that the buffer really is larger. Otherwise,
doubling zero gives us zero.
Pointy hat: /me
Store mtree option as individual strings in a single linked
list. Scan when adding options for an entry for duplicate
keywords and override them. Implement /set and /unset support.
Change the behaviour for multiple full pathes to match the
NetBSD mtree behaviour: only the last line is used.
Match NetBSD mtree behaviour and always parse the permissions
as octal. Actually return a warning when the mode is symbolic.
Handle the current directory as full path, this seems to match
the behaviour of NetBSD mtree, but is not documented.
Tim Kientzle [Tue, 20 May 2008 21:45:46 +0000 (17:45 -0400)]
Fix the broken mtree test. Joerg recently made the mtree parser
a bit stricter; 'type' is now a required keyword, in keeping
with the expectations of mtree(8).
As specified for nbpax, when the optional keyword is present and
the on-disk item doesn't exist or has a different type, skip it.
Fix the error message for the mismatch to not include errno.
Claim copyright for parts of this file.
Skip over entries where mtree and on-disk type disagree.
For directories and files, fstat the on-disk item, otherwise
lstat it. Copy device, gid/uid, mtime, nlink, permissions and
size if they have not been explicitly specified.
Skip entries that don't specify the type -- mtree requires this.
Input archive and target archive might disagree about the entry
size. E.g. the entry to write is a hardlink and the tar writer
reset the entry size to 0, skip the data on the input side.
Tim Kientzle [Thu, 15 May 2008 22:14:57 +0000 (18:14 -0400)]
Overhaul this test a bit so that the right entries get stored
as hardlinks, regardless of what order the underlying filesystem
returns the files in. This basically just involves segregating
the different groups of entries into separate dirs and specifying
those dirs -- in the correct order -- on the command line,
instead of relying entirely on the tree traversal.
This should fix a persistent test failure that's been reported
on Linux when using newer filesystems.
Tim Kientzle [Thu, 15 May 2008 22:11:25 +0000 (18:11 -0400)]
Failing to write a header is not just a "WARN" problem,
it's a "FAILED" operation. In particular, this fixes a
relatively obscure problem writing ustar archives with bsdtar.
Prior to this, bsdtar would report additional bogus errors
because it kept trying to write the body even after the
header failed.
Actually recognize the '-s' option by adding it to the getopt
string. Refactor substitution code by integrating it in
edit_pathname. This also makes the subst code apply to -c.
Add a very basic test case.
Add support to sparsify files on write. This scans output blocks
for NUL bytes and seeks if a block is full of NULs. This is
optional, but the penalty is very small.
optional and ignore are pure keywords without arguments.
Currently ignore them both, for optional it might be better to
mask errors later though.
Add device keyword support, this doesn't do any mapping though.
Fix link keyword, it always specifies a symlink independent of
whether the type keyword was processed already.
Fix BSD ar writer to not increase the size of the archive_entry
for long filenames. This breaks applications that use
archive_entry_size to decide how much to write.
Ignore the various checksums supported by NetBSD's mtree.
Add support for flags keyword.
Add support for nlink keyword.
Sort time keyword correctly.
Ignore tags keyword. This should deal with include/exclude
later.
Add commented out entries for device, ignore and optional.
Refactor stat cache handling into a function. Consistenly use
lstat(2) as no symlink sohuld ever be referenced at this point.
After unlink(2)/rmdir(2) operations ensure that the stat cache
is reset.
Fix the extraction of archives where hardlinks and the main file
disagree on permissions. This can be observed with base.tgz of
an unprivileged NetBSD build.
After the change permissions are restored only for files that
have content attached. This works for almost ustar. For pax
archives with optional data in the second entry and cpio this
will fail for unprivileged user if the first entry doesn't allow
writing for the current user.
Tim Kientzle [Thu, 8 May 2008 20:48:01 +0000 (16:48 -0400)]
Just skip entries where --strip-components removes everything.
Previously, this was failing for dir entries that had exactly
the right number of components, e.g. "a/" with --strip-components=1,
leading to odd warnings about "Invalid empty pathname"
PR: bin/121158
Tim Kientzle [Thu, 8 May 2008 20:28:36 +0000 (16:28 -0400)]
If there are leftover inclusions (command-line patterns) after
extraction, warn about them:
$ bsdtar xfv tar.tgz foo bar
-rw-r--r-- 0 tim tim 0 May 8 19:50 foo
bsdtar: bar: Not found in archive
bsdtar: Error exit delayed from previous errors.
Tim Kientzle [Wed, 7 May 2008 08:54:46 +0000 (04:54 -0400)]
More Windows cleanup: Add a copyright notice (as discussed with
the author, Kees Zeelenberg) to archive_windows.c, include a comment
explaining the history and intent for this file. Remove a couple
of unnecessary functions.
Tim Kientzle [Tue, 6 May 2008 22:55:46 +0000 (18:55 -0400)]
Add some additional examples and clarification about using
mtree descriptions to generate tar archives using
tar -cf output.tar @input.mtree
Obtained from: Vincent Zweije
projects / thirdparty / libarchive.git / log
