Avery Pennarun [Mon, 11 Jan 2010 20:18:35 +0000 (15:18 -0500)]
Merge branch 'cygwin'
* cygwin:
Assorted cleanups to Luke's cygwin fixes.
Makefile: work with cygwin on different windows versions.
.gitignore sanity.
Makefile: On Windows, executable files must end with .exe.
client.py: Windows files don't support ':', so rename cachedir.
index.py: os.rename() fails on Windows if dstfile already exists.
Don't try to rename tmpfiles into existing open files.
helpers.py: Cygwin doesn't support `hostname -f`, use `hostname`.
cmd-index.py: Retry os.open without O_LARGEFILE if not supported.
Makefile: Build on Windows under Cygwin.
Avery Pennarun [Mon, 11 Jan 2010 20:06:03 +0000 (15:06 -0500)]
Assorted cleanups to Luke's cygwin fixes.
There were a few things that weren't quite done how I would have done them,
so I changed the implementation. Should still work in cygwin, though.
The only actual functional changes are:
- index.Reader.close() now actually sets m=None rather than just closing it
- removed the "if rename fails, then unlink first" logic, which is
seemingly not needed after all.
- rather than special-casing cygwin to use "hostname" instead of "hostname
-f", it turns out python has a socket.getfqdn() that does what we want.
Avery Pennarun [Mon, 11 Jan 2010 19:57:23 +0000 (14:57 -0500)]
Makefile: work with cygwin on different windows versions.
Just check the CYGWIN part; don't depend on the fact that it's NT 5.1. (Of
course, uname isn't supposed to report such things by default anyway... but
that's cygwin for you.)
Lukasz Kosewski [Sun, 10 Jan 2010 09:04:17 +0000 (04:04 -0500)]
Don't try to rename tmpfiles into existing open files.
Linux and friends have no problem with this, but Windows doesn't allow
this without some effort, which we can avoid by... not needing to write
to an already-open file.
Give index.Reader a 'close' method which identifies and closes any open
mmaped files, and make cmd-index.py use this before trying to close a
index.Writer instance (which renames a tmpfile into the same file the
Reader has mmaped).
Lukasz Kosewski [Sun, 10 Jan 2010 08:57:42 +0000 (03:57 -0500)]
cmd-index.py: Retry os.open without O_LARGEFILE if not supported.
Python under Cygwin doesn't have os.O_LARGEFILE, so if we receive an
'AttributeError' exception trying to open something, just remove
O_LARGEFILE and try again.
Lukasz Kosewski [Sun, 10 Jan 2010 08:52:52 +0000 (03:52 -0500)]
Makefile: Build on Windows under Cygwin.
- Python modules have to end with .dll instead .so to load into Python
via 'import'.
- GCC under Windows builds all programs with -fPIC, and doesn't accept
this command-line option.
- libpython2.5.dll is found in /usr/bin under Cygwin (wtf?), so we need
to add this to the LDFLAGS path.
- 'make clean' should remove .dll files too.
Avery Pennarun [Sun, 10 Jan 2010 06:13:10 +0000 (01:13 -0500)]
This adds the long-awaited indexfile feature, so you no longer have to feed
your backups through tar.
Okay, 'bup save' is still a bit weak... but it could be much worse.
Merge branch 'indexfile'
* indexfile:
Minor fix for python 2.4.4 compatibility.
cmd-save: completely reimplement using the indexfile.
Moved some reusable index-handling code from cmd-index.py to index.py.
A bunch of wvtests for the 'bup index' command.
Start using wvtest.sh for shell-based tests in test-sh.
cmd-index: default indexfile path is ~/.bup/bupindex, not $PWD/index
cmd-index: skip merging the index if nothing was written to the new one.
cmd-index: only update if -u is given; print only given file/dirnames.
cmd-index: correct reporting of deleted vs. added vs. modified status.
Generalize the multi-index-walking code.
cmd-index: indexfiles should start with a well-known header.
cmd-index: eliminate redundant paths from index update command.
cmd-index: some handy options.
index: add --xdev (--one-file-system) option.
Fix some bugs with indexing '/'
cmd-index: basic index reader/writer/merger.
Avery Pennarun [Sun, 10 Jan 2010 03:43:48 +0000 (22:43 -0500)]
cmd-save: completely reimplement using the indexfile.
'bup save' no longer walks the filesystem: instead it walks the indexfile
(which is much faster) and doesn't bother opening any files that haven't had
an attribute change, since it can just reuse their sha1 from before. That
makes it *much* faster in the common case.
Avery Pennarun [Sun, 10 Jan 2010 00:27:26 +0000 (19:27 -0500)]
cmd-index: only update if -u is given; print only given file/dirnames.
cmd-index now does two things:
- it updates the index with the given names if -u is given
- it prints the index if -p, -s, or -m are given.
In both cases, if filenames are given, it operates (recursively) on the
given filenames or directories. If no filenames are given, -u fails (we
don't want to default to /; it's too slow) but -p/s/m just prints the whole
index.
Avery Pennarun [Sun, 10 Jan 2010 00:07:05 +0000 (19:07 -0500)]
cmd-index: correct reporting of deleted vs. added vs. modified status.
A file with an all-zero sha1 is considered Added instead of Modified, since
it has obviously *never* had a valid sha1. (A modified file has an old
sha1, but IX_HASHVALID isn't set.)
We also now don't remove old files from the index - for now - so that we can
report old files with a D status. This might perhaps be useful eventually.
Furthermore, we had a but where reindexing a particular filename would
"sometimes" cause siblings of that file to be marked as deleted. The
sibling entries should never be updated, because we didn't check them and
thus have no idea of their new status. This bug was mostly caused by the
silly way we current pass dirnames and filenames around...
Avery Pennarun [Thu, 7 Jan 2010 23:54:40 +0000 (18:54 -0500)]
cmd-index: eliminate redundant paths from index update command.
If someone asks to update "/etc" and "/etc/passwd", the latter is redundant
because it's included in the first. Don't bother updating the file twice
(and thus causing two index merges, etc).
Ideally we would only do one merge for *any* number of updates (etc /etc and
/var). This should be possible as long as we sort the entries correctly
(/var/ and then /etc/), since a single sequential indexfile could just have
one appended to the other. But we don't do that yet.
Avery Pennarun [Thu, 7 Jan 2010 23:43:02 +0000 (18:43 -0500)]
cmd-index: some handy options.
New options:
--modified: print only files that aren't up to date
--status: prefix printouts with status chars
--fake-valid: mark all entries as up to date
--indexfile: override the default index filename
Avery Pennarun [Wed, 6 Jan 2010 21:42:54 +0000 (16:42 -0500)]
splitting to a remote server would cause "already busy" errors.
Specifically:
client.ClientError: already busy with command 'receive-objects'
That's because recent changes removed the call to onclose() from
PackWriter_Remote. Now it's back, plus I added an extra unit test to reveal
the problem.
Avery Pennarun [Wed, 6 Jan 2010 18:03:23 +0000 (13:03 -0500)]
client: enhance the PATH when searching for the 'bup' binary.
Automatically adds the *local* $PWD to the *remote* $PATH before trying to
run 'bup server'. That way, if you build the source in exactly the same
folder on two machines - or if those two machines are actually the same
machine and you're just doing a test against localhost - it'll work.
I hereby curse both "sh -c <command>" and "ssh hostname -- <command>" for
not allowing a sensible way to just set argv[] without doing any stupid
quoting. Nasty.
Avery Pennarun [Wed, 6 Jan 2010 17:07:59 +0000 (12:07 -0500)]
Much more user-friendly error messages when bup can't exec the server.
...which happens unfortunately often, including in 'make test' when PATH
doesn't include bup. I'll fix that next. But it makes sense to fix the
error messages first :)
Avery Pennarun [Wed, 6 Jan 2010 05:19:11 +0000 (00:19 -0500)]
split: Prevent memory drain from excessively long shalists.
This avoids huge RAM usage when you're splitting a really huge object, plus
git probably doesn't work too well with single trees that contain millions
of objects anyway.
Avery Pennarun [Wed, 6 Jan 2010 04:42:15 +0000 (23:42 -0500)]
Split packs around 100M objects or 1G bytes.
This will make pruning much easier later, plus avoids any problems with
packs >= 2GB (not that we've had any of those yet, but...), plus avoids
wasting RAM with an overly full MultiPackIndex.also{} dictionary.
Avery Pennarun [Wed, 6 Jan 2010 04:50:41 +0000 (23:50 -0500)]
OOPS! Was writing one byte at a time to the server.
_raw_write() expects a list, not a string, so it was iterating over it
character by character. Magically it worked anyway. Which is sort of cool,
and yet not.
Avery Pennarun [Wed, 6 Jan 2010 03:21:18 +0000 (22:21 -0500)]
Fix compatibility with git 1.5.4.3 (Ubuntu Hardy).
Thanks to Andy Chong for reporting the problem.
Basically it comes down to two things that are missing in that version but
exist in git 1.5.6:
- git init --bare doesn't work, but git --bare init does.
- git cat-file --batch doesn't exist in that version.
Unfortunately, the latter problem is pretty serious; bup join is really slow
without it. I guess it might be time to implement an internal version of
cat-file.
Avery Pennarun [Mon, 4 Jan 2010 16:48:38 +0000 (11:48 -0500)]
Fix two bugs reported by dcoombs.
test-sh was assuming 'bup' was on the PATH. (It wasn't *supposed* to be
assuming that, but the "alias bup=whatever" line wasn't working,
apparently.)
randomgen.c triggered a warning in some versions of gcc about the return
value of write() being ignored. It really doesn't bother me if some of my
random bytes don't get written, but whatever; I'll assert instead, which
should shut it up.
Avery Pennarun [Sun, 3 Jan 2010 11:17:30 +0000 (06:17 -0500)]
Support incremental backups to a remote server.
We now cache the server's packfile indexes locally, so we know which objects
he does and doesn't have. That way we can send him a packfile with only the
ones he's missing.
cmd-split supports this now, but cmd-save still doesn't support remote
servers.
The -n option (set a ref correctly) doesn't work yet either.
Avery Pennarun [Sun, 3 Jan 2010 10:00:38 +0000 (05:00 -0500)]
Extremely basic 'bup server' support.
It's enough to send a pack to the remote end with 'bup split', though 'bup
save' doesn't support it yet, and we're not smart enough to do incremental
backups, which means we generate the gigantic pack every single time.
Avery Pennarun [Sat, 2 Jan 2010 09:16:25 +0000 (04:16 -0500)]
Write git pack files instead of loose object files.
This causes much, much less disk grinding than creating zillions of files,
plus it's even more disk space efficient.
We could theoretically make it go even faster by generating the .idx file
ourselves, but for now, we just call "git index-pack" to do it. That
helpfully also confirms that the data was written in a git-compatible way.
Avery Pennarun [Sat, 2 Jan 2010 06:46:06 +0000 (01:46 -0500)]
'bup split': speed optimization for never-ending blocks.
For blocks which never got split (eg. huge endless streams of zeroes) we
would constantly scan and re-scan the same sub-blocks, making things go
really slowly. In such a bad situation, there's no point in being so careful;
just dump the *entire* input buffer to a chunk and move on. This vastly
speeds up splitting of files with lots of blank space in them, eg.
VirtualBox images.
Also add a cache for git.hash_raw() so it doesn't have to stat() the same
blob files over and over if the same blocks (especially zeroes) occur more
than once.
Avery Pennarun [Wed, 30 Dec 2009 23:18:35 +0000 (18:18 -0500)]
Add a '-t' option to 'bup split' and make 'bup join' support it.
This lets you generate a git "tree" object with the list of hashes in a
particular file, so you can treat the file as a directory as far as git is
concerned. And 'bup join' knows how to take a tree and concatenate it
together to reverse the operation.
Avery Pennarun [Wed, 30 Dec 2009 08:28:36 +0000 (03:28 -0500)]
Use t# instead of et# for hashsplit parameter type.
This lets us work with any kind of buffer object, which means there's no
unnecessary data copying when coming into our module. Causes a bit of
speedup.
Also refactored the checksum code for easier experimentation.
Avery Pennarun [Wed, 30 Dec 2009 07:33:35 +0000 (02:33 -0500)]
Add a C module to do the rolling checksum.
This is about 80x faster than the old speed (27megs/sec instead of 330k/sec)
but still quite a lot slower than the 60+megs/sec I get *without* the
checksum stuff. There are a few inefficiencies remaining, but not such easy
ones as before...
Avery Pennarun [Wed, 30 Dec 2009 06:08:27 +0000 (01:08 -0500)]
datagen.c: a quick program to generate a repeatable series of bytes.
Useful for testing. Note that we *don't* see the random number generator,
so every time you generate the bytes, you get the same sequence.
This is also vastly faster than /dev/urandom, since it doesn't try to be
cryptographically secure. It generates about 200 megs/sec on my computer,
which is much faster than a disk and thus useful for testing the speed of
hashsplit.