It currently searches all revs of every matching file. I'll change
this soon so that it can still do this, but it will not be the default
behaviour.
Many options are unimplemented. There's only one output mode. Binary
files are not handled yet.
The original implementation of log walked backwards through history,
which had terrible behaviour. It took several minutes to view
complete kernel change history on a fast machine, for example.
The rewrite uses a windowed approach to walk hunks of history
forwards, while still giving results in reverse order. This reduces
run time from five minutes to five seconds on my system.
In addition, the rewrite uses our normal name handling mechanisms, so
you can run a command like "hg log net/ipv4/**.c" and get a useful
answer. It optimises for three different cases (no arguments, only
files, and anything goes), so it performs well in all circumstances
I've tested.
Based on a patch by Fuming Wang
- add util.binary which decides whether a file is binary if it has any NUL
characters in the first 1K.
- teach mdiff.unidiff to print "binary file <x> has changed" for binary files
- add text flag to cause unidiff and dodiff to treat all files as text
- add -a and --text flags (like diff(1)) to hg diff and export
- update docs
- rename mode to branch_merge
- use explicit update mode
- use negative mtime for updates that set mtime
- expand some cryptic variable names
- elaborate merge dirstate comments
- remove redundant manifest lookup for non-merge case
- remove impossible merge case
- fix up test cases
- use new changed list to track what files actually got new revlog entries
at commit/rawcommit time
- when updating a file during a merge, don't mark it as completely unchanged
- handle file not being in parent 1 in status
Since switching to the multihead approach, we've been creating
excessive file-level merges where files are marked as merged with
their ancestors.
This explicitly checks at commit time whether the two parent versions
are linearly related, and if so, reduces the file check-in to a
non-merge. Then the file is compared against the remaining parent,
and, if equal, skips check-in of that file (as it's not changed).
Since we're not checking in all files that were different between
versions, we no longer need to mark so many files for merge. This
removes most of the 'm' state marking as well.
Finally, it is possible to do a tree-level merge with no file-level
changes. This will happen if one user changes file A and another
changes file B. Thus, if we have have two parents, we allow commit to
proceed even if there are no file-level changes.
- allow passing a repo object instead of a path
(if we get a string, we construct a repo object)
- hg serve: pass options via repo.ui.setconfig
- add --style option
- get default name from repo.root rather than getcwd()
- remove template argument to hgweb()
- reduce create_server from 8 args to 1
Right now, mercurial doesn't take daylight saving time into account
because it uses time.timezone inconditionally, instead of resorting to
time.altzone if daylight saving time is on.
Example:
% hg log | head -12
changeset: 791:41440890e57d2dbacde03a2a114e5114a031ea4a
tag: tip
user: Samuel Tardieu <sam@rfc1149.net>
date: Mon Jul 25 13:24:15 2005 +0200
summary: Parse various date formats when accepting user specified dates
[...]
changeset: 787:ba5c2021364ee876dae170e8f153e43bd0b2bd73
parent: 785:31a80e3723cee0a5991340facb2707e71983eb41
user: Bryan O'Sullivan <bos@serpentine.com>
date: Sun Jul 24 00:39:24 2005 -0800
summary: Get "forget" command to use new walk code.
[...]
This merges a bunch of the ideas for a multi-repository CGI script and
simplifies things a bit.
It uses PATH_INFO so it generally wants paths of the form:
http://host/hg.cgi/virtualpath
This gets mapped via a simple config file to the real path. All info
about the repo is read from the repo's hgrc.
Most read operations in hg don't need locks because we order reads and
writes for consistency. Clone is an exception to this as we're copying
entire file histories and could end up with more file history copied
than we have commits.
For now, make clone take a lock on the source repo. Non-hardlinked
clone should eventually be changed to use lockless pull.
named branching is when you identify a head by a tag on an earlier revision.
This patch adds repo.branchlookup for searching through the tree to
find branch tags for heads.
hg update -b tag is added to checkout based on branch tags
hg heads -b is added to print the tag associated with each head
Temporary fix to allow Mercurial to build on HP-UX 11, as the C
compiler on HP-UX 11 doesn't support 'inline' qualifier. The
'__inline' qualifier seemed to be supported, but not without
first resolving other associated issues.
Prior to this, a file that did not exist was reported as showing up in
the filesystem, as were files of unsupported types (such as fifos).
Now, an error message is printed and nothing is returned in such cases.
This change also moves the commands.pathto function to the util module,
as the walk code needs it to print non-confusing error messages.
a list of files to look at. These end up with a src of 'f' from
walk() but no longer exist on the filesystem.
Index: mine/mercurial/commands.py
===================================================================
The repo walking code introduces a number of calls to dirstate.map.copy(),
significantly slowing down the walk on large trees. When a list of
files is passed to the walking code, we should only look at map entries
relevant to the file list passed in.
dirstate.filterfiles() is added to return a subset of the dirstate map.
The subset includes in files passed in, and if one of the files requested
is actually a directory, it includes any files inside that directory tree.
This brings the time for hg diff Makefile down from 1.7s to .3s on
a linux kernel repo.
Also, the diff command was unconditionally calling makewalk, leading
to an extra pass through repo.changes. This patch avoids the call
to makewalk when commands.diff isn't given a list of patterns, cutting
the time for hg diff (with no args) in half.
Index: mine/mercurial/hg.py
===================================================================
The dirstate has to match what is in the repository (what would be
checked out with 'hg update -C'), because the resulting file may be
identical to the new parent, or it may be completely different.
Previously the dirstate wasn't updated, so if you changed the file to
look like the original parent, it might be considered unmodified
relative to the new parent.
On the other OS, it seems that case insensitivity for
environment vars can bite users when using some unknown
combination of python 2.4.1 and win2kSP4+minsys (and
probably other vversions of these softwares).
The best way to avoid problems in those weird cases is to
ignore OSError exception during env var deletion.
Interface is not yet changed.
Non-regular files are considered to be removed or deleted.
Removed the unneeded match(fn) call for adding to the unknown list.
This patch allows propagates the template map though recursively
though all the templates.
This allows for some hgweb template cleanup patches as well as it makes
writing new skins/themes for hgweb much much easier. (I'm planing to
write several basic ones.)
The first consequence of this is that absolute and relative paths now
all work in the same way. The second is that paths that lie outside
the repository now cause an error to be reported, instead of something
arbitrary and expensive being done.
Internally, all of the serious work is in the util package. The new
canonpath function takes an arbitrary path and either returns a
canonical path or raises an error. Because it needs to know where the
repository root is, it must be fed a repository or dirstate object, which
has given commands.matchpats and friends a new parameter to pass along.
The util.matcher function uses this to canonicalise globs and relative
path names.
Meanwhile, I've moved the Abort exception from commands to util, and
killed off the redundant util.CommandError exception.
Show command aliases in 'hg help something', unless in quiet mode.
Show short command description with 'hg help -q something'.
Show global options in verbose mode of command help.
State that Mercurial is a Distributed SCM.
When hg is executed without parameters ( or hg is executed with the
'help shortlist' command ) the short help list is displayed.
If the 'help' command is passed on the commandline the full help list
is displayed.
In both the cases if the switch '-v' is passed, the help contains also
the aliases of the short/full commands list and the global switches.
Added options -m, -a, -r and u to select files corresponding
to status M, A, R and ? respectively. If none of these
options are specified, files of all status will be shown.
I ran into a bug while importing a large repository into mercurial.
The diff algorithm does not allocate a big enough array of hunks
for some test cases. This results in memory corruption, and possibly,
as in my case, a seg fault.
You should be able to reproduce this problem with any case of more
than a few lines that follows this pattern:
a b
= =
1 1
2
2 3
4
3 5
.
4 .
.
5
.
.
.
I.e., "a" has blank lines on every other line that have been removed in
"b". In this case, the number of matching hunks is equal to the number
of lines in "b". This is more than ((an + bn)/4 + 2). I'm not sure what
motivates this formula, but when I changed it to the smaller of an or
bn (+ 1), it works.
[comment added by mpm]
commands.walk has been refactored. commands.walk's behaviour remains as
it was, but there is a new function, commands.makewalk. This returns
results that can be passed along to other internal code, in our case
dodiff.
This is an improvement in behaviour, but the walk and changes code still
has some flaws that make sorted name presentation difficult:
- changes returns tuples of names, instead of a sorted list of (name,
status) pairs.
- walk yields deleted names after all others.
New function: commands.pathto returns the relative path from one path
to another. For example, given foo/bar and baz/quux, it will return
../../baz/quux. This new function is used by the walk and status code
to print relative paths correctly.
New command: debugwalk exercises the walk code without doing anything
more.
hg.dirstate.walk now yields normalised names. For example, if you're
in the baz directory and you ask it to walk ../foo/bar/.., it will yield
names starting with foo/.
As a result of this change, all of the other walk and changes methods
in this module also return normalised names.
The util.matcher function now normalises globs and path names, so that
it will match normalised names properly.
Finally, util.matcher uses the non-glob prefix of a glob to tell walk
which directories to scan. Perviously, a glob like foo/* would scan
everything, but only return matches for foo/*. Now, foo/* only scans
under foo (using the globprefix function), which is much faster.
The code is slightly complicated by the need to commit all outstanding
changes in the repository if no file names are given (other commands
operate on the current directory and its subdirectories in this case).
localrepository.changes has acquired an optional match parameter, to let
it filter out include/exclude options.
When we switched to the new walk code for commands, we no longer passed a
list of specific files to the repo or dirstate walk or changes methods.
This meant that we always walked and attempted to match everything,
which was not efficient.
Now, if we are given any patterns to match, or nothing at all, we still
walk everything. But if we are given only file names that contain no
glob characters, we only walk those.
With no names, it now recursively forgets everything, as is the default
behaviour of other commands. And prints the names of all files it
hasn't specifically been told to forget.
When I rewrote addremove, I lazily put a call to repo.changes in,
which was unnecessary and slow. This is a new rewrite, preserving the
file name behaviour, but replacing the call to repo.changes with a walk,
which is much cheaper, and avoids calls to os.stat on all but files that
have probably been deleted.
Commit would overwrite undo.dirstate unconditionally, so an undo after
an aborted commit would restore the dirstate from the aborted commit
and not the prior transaction.
This copies dirstate to journal.dirstate and moves it after a
successful transaction.
It is now more verbose than it used to be. If given file names, it
prints nothing, as before. But if given patterns or nothing, it prints
the names of the files it is operating on, to remove that air of mystery.
It also now operates at or below the current directory.
For 3-way merges, we were marking files as 'm', which should only be
used for branch merges. This was causing files to show up in 'hg
status' that aren't actually modified.
you geta pretty obscure error (zlib: uknown compression type). The
attached patch modifies hgweb.py and hg.py to supply and check a
'Content-type: application/hg-0.1' HTTP header for the branches,
between and changegroup commands, so that we know it's a proper hg
repo before snarfing the input. Comments appreciated!