Issue 3681 - Add suport for "Fixes XXXX - ..." commit messages

sitescripts/hg/README.md

+# Mercurial hooks
+
+`sitescripts/hg` contains Mercurial hooks for integration with other components
+of our infrastructure.
+
+## IRC integration
+
+`irchook.py` contains a commit hook that posts the information about new pushed
+commits to an IRC channel. 
+
+## Trac integration
+
+`update_issues.py` contains two hooks: `changegroup_hook` and `pushkey_hook`.
+They will recognise issue references in commit messages and update referenced
+issues in Adblock Plus issue tracker.

[Felix Dahlke, 2016/05/17 20:12:05]

Nit: "in the" maybe?

[Vasily Kuznetsov, 2016/05/18 08:30:44]

Done.

+
+The format of the commit messages is "ISSUE-REFERENCE - MESSAGE"
+where ISSUE-REFERENCE is one of "Noissue", "Issue NUMBER" or "Fixes NUMBER".
+Several "Issue" and "Fixes" references separated by commas can be present
+in the same commit message (for example: "Issue 1, Fixes 2 - Two issues").
+Such commit will affect all the referenced issues.
+
+* `changegroup_hook` will post a comment with the link to the commit into the
+  issue if the issue is referenced from the commit message.
+* `pushkey_hook` will close the issues and assign milestones to them (based
+  on their module) when `master` bookmark passes over the commits that fix

[Felix Dahlke, 2016/05/17 20:12:06]

Nit: "when the"?

[Vasily Kuznetsov, 2016/05/18 08:30:44]

Done.

+  them. It will not assign a milestone if the issue already has one.
+
+### Configuring the repository
+
+`changegroup_hook` should be installed as `changegroup` or
+`pretxnchangegroup` hook. `pushkey_hook` should be installed as
+`pushkey` or `prepushkey` hook. For example (in `.hg/hgrc`):
+
+    [hooks]
+    pretxnchangegroup = python:.../update_issues.py:changegroup_hook
+    pushkey = python:.../update_issues.py:pushkey_hook
+
+### Configuring the hooks
+
+The hooks are configured via `sitescripts.ini` in `hg` and
+`hg_module_milestones` sections. For example:
+
+    [hg]
+    trac_xmlrpc_url=https://abpbot:abpbot@issues.adblockplus.org/login/xmlrpc
+    issue_url_template=https://issues.adblockplus.org/ticket/{id}
+
+    [hg_module_milestones]
+    platform=adblock-plus(-[\d\.]+)?-for-chrome-opera-safari(-next)?
+    Adblock-Plus-for-Firefox=adblock-plus(-[\d\.]+)?-for-firefox(-next)?
+
+`hg.track_xmlrpc_url` key from is used to determine the address of XMLRPC
+interface of Trac and `hg.issue_url_template` as a template for producing links
+to the referenced issues that are displayed in the log.
+
+The keys of the `hg_module_milestones` section are module names and the values
+are corresponding milestone regular expressions (they are matched
+case-insensitively). The first open milestone that matches the regular
+expression of issue's module will be assigned to the issue when the `master`

[Felix Dahlke, 2016/05/17 20:12:06]

Nit: "of an"?

[Vasily Kuznetsov, 2016/05/18 08:30:44]

Should not it be "of the issue's module"? We're talking about specific issue and
module here, right? I changed it to this.

[Felix Dahlke, 2016/05/18 08:37:22]

Acknowledged.

+bookmark passes a commit that fixes it.
+
+### Master bookmark
+
+What exactly does it mean when we say _`master` bookmark is passing a commit_?
+The idea is that if the `master` bookmark _passed_ a commit we will have
+those changes in our working copy when we do `hg checkout master`.

[Felix Dahlke, 2016/05/17 20:12:05]

Nit: `checkout` is just an alias for `update`, which is the normal command to
use. Technically it's correct though, so I don't mind much.

[Vasily Kuznetsov, 2016/05/18 08:30:44]

I think I wrote it this way thinking about a universal command that git users
would also understand. Perhaps it doesn't make much sense since we're talking
about a mercurial hook, but I have a mild preference for this more inclusive
approach. So I'll leave it as is for now.

[Felix Dahlke, 2016/05/18 08:37:22]

Acknowledged.

+
+Let's first look at a simple case, linear commit history like this:
+
+    one <- two <- three
+
+Here `one` is a parent commit of `two` and `two` is a parent of `three`. If
+the `master` bookmark was on `one` and then moved to `three`, we say that it
+passed `two` and `three`. This would happen naturally if we clone the
+repository that contains `one` with the `master` bookmark pointing to it,
+check out `master`, author two commits and then push them.
+
+A somewhat similar thing happens when we have branches:
+
+    one <---- two <---- three
+         \            /
+          \-- dos <--/
+
+Here `one` is a parent commit of `two` and `dos` and they are both parents
+of `three`. If the `master` bookmark was on `two` and now is on `three`,
+we say that it passed `three` and `dos`. What happened here is that by `three`
+we've merged a branch containing `dos` into the master branch that was going
+through `one` and `two`.