This updates the thread linking code to use ghost messages instead of
user metadata to link messages into threads.
In contrast with the old approach, this is actually correct.
Previously, thread merging updated only the thread IDs of message
documents, not thread IDs stored in user metadata. As originally
diagnosed by Mark Walters [1] and as demonstrated by the broken
T260-thread-order test, this can cause notmuch to fail to link
messages even though they're in the same thread. In principle the old
approach could have been fixed by updating the user metadata thread
IDs as well, but these are not indexed and hence this would have
required a full scan of all stored thread IDs. Ghost messages solve
this problem naturally by reusing the exact same thread ID and message
ID representation and indexing as regular messages.
Furthermore, thanks to this greater symmetry, ghost messages are also
algorithmically simpler. We continue to support the old user metadata
format, so this patch can't delete any code, but when we do remove
support for the old format, several functions can simply be deleted.
[1] id:8738h7kv2q.fsf@qmul.ac.uk
This updates the message abstraction to support ghost messages: it
adds a message flag that distinguishes regular messages from ghost
messages, and an internal function for initializing a newly created
(blank) message as a ghost message.
In emacs 24.4 the messages buffer starts being read-only, which kills
these tests. This seems to be the point of the variable
inihibit-read-only, which has existed at least since emacs 21.
After yet another variation in objdump output caused this test to fail
(on a Debian port, no less), I decided whatever putative benefit we
get from looking at the object files instead of the library isn't
worth the maintenence headache.
This version uses nm -P. nm -P should be portable, and fixed format.
It purposely doesn't use the -D argument, since that is non-POSIX and
nm on GNU/Linux seems do the right thing without it.
It still won't work out of the box on e.g. Mac OS/X. I think the right
thing to do there is to move some more configuration information into
sh.config.
(cherry picked from commit c34d6bad0f)
Previously, this was implemented using a horrible GDB script (because
there is no such thing as a non-horrible GDB script). This GDB script
often broke with newer versions of GDB for mysterious reasons. Port
the test script to GDB's Python API, which makes the code much cleaner
and, hopefully, more stable.
(cherry picked from commit cbbda62258)
Conflicts:
test/T380-atomicity.sh
Apparently README.rst overrides README, so this will show up instead
of our generic README on github.
If the user is already on github, then clicking a link for more
information is not a hardship.
This adds a variable to make starting in insert mode optional when
composing and replying to emails. I found it unusual to be started in
insert mode so I thought I'd make it optional as others may find this as
well.
Ian
This patch switches from reading .notmuch-config directly to using
the CLI the same way that emacs does it. It actually uses less code
and is probably less error prone.
Ian
Add --keep option to keep any remaining stuff in index or file. We
could distinguish between failures to index and failures to apply tags
or maildir sync, but for simplicity just have one.
Handle failures gracefully in add_file_to_database, renamed simply
add_file while at it. Add keep option to not remove the message from
database if tagging or tag syncing to maildir flags fails. Expand the
function documentation to cover the changes.
In the interest of robustness, avoid undefined behavior of
sortable_unserialise if the date value is missing. This shouldn't
happen now, but ghost messages will have blank date values.
This moves the code to retrieve and clear the metadata thread ID out
of _notmuch_database_link_message into its own function. This will
simplify future changes.
Previously, this was performed by notmuch_database_add_message. This
happens to be the only caller currently (which is why this was safe),
but we're about to introduce more callers, and it makes more sense to
put responsibility for ID compression in the lower-level function
rather than requiring each caller to handle it.
It blows things up by a factor of six or so, so it's worth giving
people a heads up. It won't effect e.g. Debian, that already builds
with -g and then strips.
For more details, see the commit message for 7f2cb3be (nmbug:
Translate to Python, 2014-10-03). I realized while writing this that
the 7f2cb3be commit message has:
* 'nmbug log' now execs 'git log', as there's no need to keep the
Python process around once we've launched Git there.
But we dropped that exec in favor of the subprocess approach between
v3 and v4, I just forgot to update the commit message [1].
[1]: id:e630b6763e9d0771718afee41ea15b29bb4a1de8.1409935538.git.wking@tremily.us
http://article.gmane.org/gmane.mail.notmuch.general/19007
Previously, this was implemented using a horrible GDB script (because
there is no such thing as a non-horrible GDB script). This GDB script
often broke with newer versions of GDB for mysterious reasons. Port
the test script to GDB's Python API, which makes the code much cleaner
and, hopefully, more stable.
In the future, tests may rely on debug symbols being present in
notmuch, so we plan to switch the default flags.
The main purpose of this test is to help explain the perhaps
mysterious failures of other tests which rely on symbols being
present.
This allows us to capture stdout and stderr separately, and do other
explicit subprocess manipulation without resorting to external
packages. It should be compatible with Python 2.7 and later
(including the 3.x series).
Most of the user-facing interface is the same, but there are a few
changes, where reproducing the original interface was too difficult or
I saw a change to make the underlying Git UI accessible:
* 'nmbug help' has been split between the general 'nmbug --help' and
the command-specific 'nmbug COMMAND --help'.
* Commands are no longer split into "most common", "other useful", and
"less common" sets. If we need something like this, I'd prefer
workflow examples highlighting common commands in the module
docstring (available with 'nmbug --help').
* 'nmbug commit' now only uses a single argument for the optional
commit-message text. I wanted to expose more of the underlying 'git
commit' UI, since I personally like to write my commit messages in
an editor with the notes added by 'git commit -v' to jog my memory.
Unfortunately, we're using 'git commit-tree' instead of 'git
commit', and commit-tree is too low-level for editor-launching. I'd
be interested in rewriting commit() to use 'git commit', but that
seemed like it was outside the scope of this rewrite. So I'm not
supporting all of Git's commit syntax in this patch, but I can at
least match 'git commit -m MESSAGE' in requiring command-line commit
messages to be a single argument.
* The default repository for 'nmbug push' and 'nmbug fetch' is now the
current branch's upstream (branch.<name>.remote) instead of
'origin'. When we have to, we extract this remote by hand, but
where possible we just call the Git command without a repository
argument, and leave it to Git to figure out the default.
* 'nmbug push' accepts multiple refspecs if you want to explicitly
specify what to push. Otherwise, the refspec(s) pushed depend on
push.default. The Perl version hardcoded 'master' as the pushed
refspec.
* 'nmbug pull' defaults to the current branch's upstream
(branch.<name>.remote and branch.<name>.merge) instead of hardcoding
'origin' and 'master'. It also supports multiple refspecs if for
some crazy reason you need an octopus merge (but mostly to avoid
breaking consistency with 'git pull').
* 'nmbug log' now execs 'git log', as there's no need to keep the
Python process around once we've launched Git there.
* 'nmbug status' now catches stderr, and doesn't print errors like:
No upstream configured for branch 'master'
The Perl implementation had just learned to avoid crashing on that
case, but wasn't yet catching the dying subprocess's stderr.
* 'nmbug archive' now accepts positional arguments for the tree-ish
and additional 'git archive' options. For example, you can run:
$ nmbug archive HEAD -- --format tar.gz
I wish I could have preserved the argument order from 'git archive'
(with the tree-ish at the end), but I'm not sure how to make
argparse accept arbitrary possitional arguments (some of which take
arguments). Flipping the order to put the tree-ish first seemed
easiest.
* 'nmbug merge' and 'pull' no longer checkout HEAD before running
their command, because blindly clobbering the index seems overly
risky.
* In order to avoid creating a dirty index, 'nmbug commit' now uses
the default index (instead of nmbug.index) for composing the commit.
That way the index matches the committed tree. To avoid leaving a
broken index after a failed commit, I've wrapped the whole thing in
a try/except block that resets the index to match the pre-commit
treeish on errors. That means that 'nmbug commit' will ignore
anything you've cached in the index via direct Git calls, and you'll
either end up with an index matching your notmuch tags and the new
HEAD (after a successful commit) or an index matching the original
HEAD (after a failed commit).
Although this manual is far from complete, it may be helpful for
someone. In particular building it as part of the standard build
process makes it easier to find problems when editing the
notmuch-emacs-manual.
In Xapian, closing a database implicitly aborts any outstanding
transaction and commits changes. For historical reasons,
notmuch_database_close had grown to almost, but not quite duplicate
this behavior. Before closing the database, it would explicitly (and
unnecessarily) commit it. However, if there was an outstanding
transaction (ie atomic section), commit would throw a Xapian
exception, which notmuch_database_close would unnecessarily print to
stderr, even though notmuch_database_close would ultimately abort the
transaction anyway when it called close.
This patch simplifies notmuch_database_close to explicitly abort any
outstanding transaction and then just call Database::close. This
works for both read-only and read/write databases, takes care of
committing changes, unifies the exception handling path, and codifies
aborting outstanding transactions. This is currently the only way to
abort an atomic section (and may remain so, since it would be
difficult to roll back things we may have cached from rolled-back
modifications).
Make the function calls make more sense as independent building blocks
of the big picture, with clear inputs and outputs. Split up
write_message into two. Improve function documentation. Cleanup and
clarify the error paths.
This will clean up the usage. There's the slight functional change of
potentially ending up doing extra gethostname and getpid calls, but
this is neglible.
Combine make_directory() and make_directory_and_parents() into a
single recursive mkdir_recursive() function. Clarify the code and
improve error handling. Improve error messages. Switch to using the
new function in maildir_create_folder(). Constify talloc context.
The copying has nothing to do with stdin, so call it copy_fd
instead. While at it, improve documentation and reverse the
parameters, as destination is traditionally the first parameter.
An "is something" predicate conveys the meaning better. While at it,
improve the function documentation and error message. Besides the
error message change, no functional changes.
