checkunfinished: accommodate histedit quirk Turns out histedit actually intends for commits (but not other operations like update) to be possible during its operation.

tests: test-check-code-hg.t works for all files to check with xargs, backslashes are eaten up. Convert them to slashes therefore. This is only a problem with ls (on windows). hg manifest returns slashes. The pipe char is moved before the line end for telling check-code.py that sed does not modify the output.

check-code: do not abort on an unreadable file, only report this

heads: fix children/descendants in doc (issue3992)

paper: fix rendering of the first tab in a line in Webkit (issue3990) This fixes issue3990 for Webkit-browsers, and also older Opera versions. Rendering in firefox is not changed.

hgweb: fix duplication for some search queries Given that N is maximum revision number in a repo, than if a revision with number N-100n or N-100n+1 (for any integer n) is found with a hgweb search, this revision is duplicated in search results.

sslutil: force SSLv3 on Python 2.6 and later (issue3905) We can't (easily) force SSL version on older Pythons, but on 2.6 and later we can force SSLv3, which is safer and widely supported. This also appears to work around a bug in IIS detailed in issue 3905.

httpclient: update to revision 9517a2b56fe9 of httpplus (issue3905) Includes upstream change "socketutil: force SSLv3 by default, as it is safer" which should fix issue 3905.

hgweb: force connection close on early response Not all WSGI servers close the socket when an early response is sent to a large POST request, which can cause the server to interpret the already-sent request body as an incoming (but hopelessly invalid) request.

hgweb: fix incorrect revisions count in graph (issue3977) Actual amount of revisions is used now instead of their numbers in the repo before to deal with skipped numbers correctly.

hgweb: fix incorrect way to count revisions in log (issue3977) Actual amount of revisions is used now instead of their numbers in the repo before to deal with skipped numbers correctly. This iterates starting from the newest revision (which is shown on top) yielding up to the specified count, instead of the reversed order used before. Effect of this change on efficiency is negligible, when the same changesets are returned.

hgweb: always start graph with the revision in url It is the same fix for graph command, as was recently for log. This makes the specified revision be always on top of the graph view. Before the patch, for example with repo having revisions 0, 1, 2, 3 and revision in url being '2', all revisions were shown and the specified one wasn't the first.

merge with i18n

i18n-pt_BR: synchronized with bf6bc4681383

update: add tracking of interrupted updates (issue3113) This takes advantage of the new checkunfinished infrastructure

mq: add checkunfinished support (issue3955)

transplant: add checkunfinished (issue3955) Nobody loves transplant anymore, but it was the first command to have a --continue flag.

histedit: add checkunfinished support (issue3955) The tests contain a couple corner cases where workarounds are now required. Strictly speaking, these are behavior regressions, but of an extremely obscure and marginal sort. Commits or updates in the middle of a histedit would have almost always been fairly serious user error.

rebase: add checkunfinished support (issue3955)

rebase: reorder parent check and state storage This could cause a rebase to be 'in progress' even though it aborted.

commands: add checks for unfinished operations (issue3955) commands checked: backout bisect commit graft import

update: clear any clearable unfinished operations (issue3955) Unclearable operations will cause an abort.

cmdutil: core functionality to block during multistep commands (issue3955) This adds a registration point and check functions that will allow commands to check if multistep operations like an interrupted graft or rebase are in progress before proceeding.

histedit: refuse to edit history that contains merges (issue3962)

largefiles: overridematch() should replace the file path instead of extending (issue3934)

revlog: handle hidden revs in _partialmatch (issue3979) Looking up hidden prefixes could cause a no node exception Looking up unique non-hidden prefixes could be ambiguous

fileset: handle underbar in symbols This was documented, but not implemented.

heads: modernize documentation (issue3992) The old docs emphasized topological heads rather than branch heads and incorrectly defined branch heads as not having children rather than descendants.

convert: catch empty origpaths in svn gettags (issue3941)

template: fix tabindent docstring (issue2880)

merge default into stable for 2.7 code freeze

osutil: consider WindowsError's behaviour to support python 2.4 on Windows This change treat the ESRCH error as ENOENT like WindowsError class does in python 2.5 or later. Without this change, some try..execpt code which expects errno is ENOENT may fail. Actually hg command does not work with python 2.4 on Windows. CreateFile() will fail with error code ESRCH when parent directory of specified path is not exist, or ENOENT when parent directory exist but file is not exist. Two errors are same in the mean of "file is not exist". So WindowsError class treats error code ESRCH as ENOENT in python 2.5 or later, but python 2.4 does not. Actual results with python 2.4: >>> errno.ENOENT 2 >>> errno.ESRCH 3 >>> WindowsError(3, 'msg').errno 3 >>> WindowsError(3, 'msg').args (3, 'msg') And with python 2.5 (or later): >>> errno.ENOENT 2 >>> errno.ESRCH 3 >>> WindowsError(3, 'msg').errno 2 >>> WindowsError(3, 'msg').args (3, 'msg') Note that there is no need to fix osutil.c because it never be used with python 2.4.

churn: split email aliases from the right This splits churn email aliases from the right, to enable incorrectly-specified addresses that include equal signs to be mapped to correct addresses. This will enable aliasing of bad addresses (typically typos) such as: sername=myusername that appear in the churn output through a churn alias such as: sername=myusername = myusername whereas previously splitting from the left would not enable this behavior.

hgweb: highlight line which is linked to at annotate view

hgweb: always start log with searched revision This makes the specified revision be always on top of the list. Before the patch, for example with repo having revisions 0, 1, 2, 3 and user searching for '2', all revisions were shown and the specified one wasn't the first.

util: check if re2 works before using it (issue 3964)

run-tests: backout 4f32747879d1 line endings change It made the windows buildbot sad.

hgweb: highlight line which is linked to at comparison view

hgweb: change highlighted line color to be different from 'inserted' color This changes line highlight color from a fain yellow (#ffff99) to a faint blue (#bfdfff), because yellow color is used in comparison view for inserted lines. This new color is okay for people with different forms of color blindness (tested with a simulator): a) this color looks quite different from other used backgrounds b) text doesn't lose distinction on this color

convert: fix bad conversion of copies when hg.startrev is specified The 'copynode' was looked up in self.keep as if it was a changeset node. It is however a filelog node, and self.keep would thus fail if it actually looked at its parameter ... which it only did if a startrev was specified. Instead we now don't check the copy node - we don't have to. It must have been copied from one of the parents, and we already check whether one of the parents have the copy source. We could perhaps use linkrev to see if the corresponding changeset was converted ... but that would sometimes be wrong. The existing test of this was wrong - now it is better, but it seems like it exposes a 'log' issue.

test-convert-hg-startrev: fix test for copy removal on missing parent The test assumed that 'b' was missing just because the changeset that introduced 'b' wasn't converted ... but 'b' was also not removed before the start revision, and the parent was thus not missing at all. Instead we introduce a file 'f' in rev 0 and remove it in rev 1 so the copy source really doesn't exist in the repo that is converted from rev 1.

hgweb: make stripes in diffstat with CSS

hgweb: make stripes in file log with CSS

hgweb: make stripes in graph with CSS

hgweb: make stripes in log and search with CSS

hgweb: make stripes in branch list with CSS

hgweb: make stripes in repo list with CSS

hgweb: make stripes in file annotate view with CSS

hgweb: make stripes in tag list with CSS

hgweb: make stripes in directory view with CSS

hgweb: make stripes in bookmark list with CSS

tests: update for commit --secret

gpg: show "Unknown key ID xxxxxxxx" when the status is ERRSIG

gpg: add shortkey() to convert from long id to short

gpg: getkeys() removes unused returning value "err"

gpg: treat "ERRSIG" as a valid key id but no fingerprint

commit: enable --secret option At the moment, creating secret commits is slightly cumbersome. They can either be created by changing the default commit phase to secret or by doing `hg phase --secret --force`. Both of these make secret commits appear to be like some kind of advanced feature. Secret commits, however, should be a convenient feature for people who want to work on a private branch without affecting anyone else. There should therefore be a prominent and convenient method for creating secret commits. Since the default phase is draft and there is no need to use --force to go from a secret phase to any other phase, this patch intentionally does not add --draft and --public options.

merge: deprecate the --force option The --force option in merge does not make what people think it does so it may not be visible to everyone. I have local changes and want to pull one's changes which made 2 heads. The --force option in help says -f --force force a merge with outstanding changes so I can expect that I can use it to force the merge and commit it in my local repository without taking my local changes into account. But merging with -f keeps local changes and "add" them: they must be committed or reverted before doing the merge commit. The merge -f cannot be reverted so it leads my repository in a bad state: cannot commit merge and don't want to revert/commit local changes yet. Message in help have been updated to emphasize the fact that local changes are included in the merge.

run-tests: revert previous commit, run() waits after a timeout

run-tests: reap timed-out zombies