Backed out changeset 41989e55fa375de4376e7e64b17e38312e8ec140

Backed out changeset 66a17364c30ef0804153093cfaccf830e563928a

Check for at least having a final release of python 2.3.0 in setup.py Mercurial uses features that are not in 2.3a1.

test for git empty new files

git patch: create empty added files

zsh: tab-complete status results by directory Filter diff and revert for modified files only, unless -r is given.

filelog: rename filelog.size() to filelog.len() filelog.size() is misleading for users of revlog.size() since it doesn't return the size of the revision.

revlog: more nullrev fixes

revlog: return 0 as offset for nullid

doc: clone doesn't create a working directory in remote repos

revlog.size() fix handling of rev == -1

use workingcontext.files() to detect if the repo is unclean

allow update to switch to a different branch if the repo is clean

Explicitly use REVLOGV0 instead of 0 in bundlerepo.

Define and use nullrev (revision of nullid) instead of -1.

Correct optimization from 3464f5e77f34; add a test. If there are two (or more) heads that point to the same .hgtags node, we can safely skip parsing the file in all but the last head. (In 3464f5e77f34, we were parsing the file in the first head and skipping all the others.)

log speedup: don't call revlog.node if we don't have to

log speedup: use list.extend to avoid an explicit loop

log speedup: use parentrevs in log and changeset_printer.show

log speedup: set quiet/verbose/debugflag in the fake ui object

log speedup: use only ui.write in changeset_printer.show - ui.status is not very useful, since this code is never executed when ui.quiet is true. - explicitly checking for ui.debug allows a regular hg log to run without looking at the manifest index - the calls to ui.note were converted for consistency

log speedup: walkchangerevs: filter the files only if we need them This speeds up hg log and significantly reduces memory usage (max RSS goes from ~92MB to ~21MB on the kernel repo), since we no longer store all the revisions in the cache.

Inform the user about the new URL when being redirected via http. Additionally the changed code ensures that the new URL is used, even if it doesn't end with the query string.

Stripping of query string (since 88b4755fa48f) stripped too much (issue327) rstrip(qs) not only strips qs from the right, but it continues stripping every char at the end of the URL that occurs on qs.

Merge with upstream

Turn bundle file into a string for http push, for urllib2 over proxies.

Force the content-length header to be a string in httprepo.unbundle. (Needed for Python2.3 compatibility)

Use the new UnexpectedOutput exception in httprepo, too.

New UnexpectedOutput exception to catch server errors in localrepo.stream_in If the unexpected is a string, the empty string will be mentioned, and long strings are cut to at most 400 chars.

Some clarifications for pull/push protocols: - List file:// and static-http:// - Mention that you can pull from bundle files - List http and https as valid push targets, but emphasize that this feature has to be enabled.

httprepo: record the url after a request, makes pull + redirect works POST+redirect doesn't work in python, as a workaround we record the url from the previous GETs so that when we do a POST it uses the redirected url fix issue327

add AIX to the list of compilers that don't have inline keyword

zsh: support remote URLs

zsh: suppress mq completion error messages outside of repository

zsh: suppress error message for status completion outside of repository

Turn of "Not trusting file" logging when running hgweb and hgwebdir (hg serve still shows the warning)

use untrusted settings in hgwebdir

use untrusted settings in hgweb The only exceptions are web.static and web.templates, since they can be used to get any file that is readable by the user running the CGI script. Other options can be (ab)used to increase the use of the cpu (allow_bz2) or of the bandwidth (server.uncompressed), but they're trusted anyway.

add untrusted argument to patch.diffopts

add --untrusted to showconfig

save settings from untrusted config files in a separate configparser This untrusted configparser is a superset of the trusted configparser, so that interpolation still works. Also add an "untrusted" argument to ui.config* to allow querying ui.ucdata. With --debug, we print a warning when we read an untrusted config file, and when we try to access a trusted setting that has one value in the trusted configparser and another in the untrusted configparser.

Only read .hg/hgrc files from trusted users/groups The list of trusted users and groups is specified in the [trusted] section of a hgrc; the current user is always trusted; "*" can be used to trust all users/groups. Global hgrc files are always read. On Windows (and other systems that don't have the pwd and grp modules), all .hg/hgrc files are read. This is essentially the same patch that was previously applied as revision 494521a3f142.

zsh: add revrange completion

zsh: add qbase and qtip to qapplied completions

zsh: use standard tags where possible

mq: support qimport -

hgcommand.vim: doc install delete into "black hole", g: prefix global vars.

merge with crew

Corrected hg synopsis.

zsh: complete qguard with +/- guard prefixes

hgcommand.vim: remove deprecated -b option from hg parents.

zsh: complete qapplied, qdiff, qfold, qguard, qimport, qnext, qprev as well as qrename, qselect, qseries, qunapplied and qtop.

Merge with mpm

zsh: make option lists more compact

Merge with mpm

zsh: rewrite This version of zsh completion handles global options more correctly, completes options for command abbreviations, and does much more lazy evaluation.

small refactoring of path normalization in dirstate.statwalk

Make util.unique return a list

Simplify ignore logic in dirstate.walk - kill blockignore hack - pull pconvert out of imatch - skip imatch when not ignoring

shortname: truncate at '.' too