Script to match open Phabricator reviews with potential reviewers

[kristof.beyls]

Summary

At the last EuroLLVM, I gave a lightning talk about code review
statistics on Phabricator reviews and what we could derive from that
to try and reduce waiting-for-review bottlenecks. (see
https://llvm.org/devmtg/2018-04/talks.html#Lightning_2).

One of the items I pointed to is a script we've been using internally
for a little while to try and match open Phabricator reviews to people
who might be able to review them well. I received quite a few requests
to share that script, so here it is.

Warning: this is prototype quality!

The script uses 2 similar heuristics to try and match open reviews with
potential reviewers:

• If there is overlap between the lines of code touched by the patch-under-review and lines of code that a person has written, that person may be a good reviewer.
• If there is overlap between the files touched by the patch-under-review and the source files that a person has made changes to, that person may be a good reviewer.

The script provides a percentage for each of the above heuristics and
emails a summary. For example, this morning, I received the following
summary from the script in my inbox for patches-under-review where some
change was made in the past 24 hours:

SUMMARY FOR [e-mail] (found 8 reviews):
[3.37%/41.67%] https://reviews.llvm.org/D46018 '[GlobalISel][IRTranslator] Split aggregates during IR translation' by Amara Emerson
[0.00%/100.00%] https://reviews.llvm.org/D46111 '[ARM] Enable misched for R52.' by Dave Green
[0.00%/50.00%] https://reviews.llvm.org/D45770 '[AArch64] Disable spill slot scavenging when stack realignment required.' by Paul Walker
[0.00%/40.00%] https://reviews.llvm.org/D42759 '[CGP] Split large data structres to sink more GEPs' by Haicheng Wu
[0.00%/25.00%] https://reviews.llvm.org/D45189 '[MachineOutliner][AArch64] Keep track of functions that use a red zone in AArch64MachineFunctionInfo and use that instead of checking for noredzone in the MachineOutliner' by Jessica Paquette
[0.00%/25.00%] https://reviews.llvm.org/D46107 '[AArch64] Codegen for v8.2A dot product intrinsics' by Oliver Stannard
[0.00%/12.50%] https://reviews.llvm.org/D45541 '[globalisel] Update GlobalISel emitter to match new representation of extending loads' by Daniel Sanders
[0.00%/6.25%] https://reviews.llvm.org/D44386 '[x86] Introduce the pconfig/enclv instructions' by Gabor Buella

The first percentage in square brackers is the percentage of lines in
the patch-under-review that changes lines that I wrote. The second
percentage is the percentage of files that I made at least some
changes to out of all of the files touched by the patch-under-review.

Both the script and the heuristics are far from perfect, but I've
heard positive feedback from the few colleagues the script has been
sending a summary to every day - hearing that this does help them to
quickly find patches-under-review they can help to review.

The script takes quite some time to run (I typically see it running
for 2 to 3 hours on weekdays when it gets started by a cron job early
in the morning). There are 2 reasons why it takes a long time:

• The REST api into Phabricator isn't very efficient, i.e. a lot of uninteresting data needs to be fetched. The script tries to reduce this overhead partly by caching info it has fetched on previous runs, so as to not have to refetch lots of Phabricator state on each run.
• The script uses git blame to find for each line of code in the patch who wrote the original line of code being altered. git blame is sloooowww....

Anyway - to run this script:

• First install a virtualenv as follows (using Python2.7 - Python3 is almost certainly not going to work):

$ virtualenv venv
$ . ./venv/bin/activate
$ pip install Phabricator

• Then to run the script, looking for open reviews that could be done by [e-mail], run (in the venv):

$ python ./find_interesting_reviews.py [e-mail]

Please note that "[e-mail]" needs to be the exact email address
(capitalization is important) that the git LLVM repository knows the
person as. Multiple email addresses can be specified on the command
line. Note that the script as is will email the results to all email
addresses specified on the command line - so be careful not to spam
people accidentally!

[lebedev.ri]

This should probably go into llvm/utils, not the root.

It would also be good to have the opposite script - "given git diff / commits, whom should i add as reviewers?"
It could evaluate CODE_OWNERS.txt, git log / git shortlog -sne of the changed files, git blame.

[philip.pfaffe]

Do you plan on actually commiting this to the repository, or would you like any feedback nontheless?

One usability nit that I see is send_emails(). It would be great to be able to configure the SMTP connection, to disable the feature altogether, and to seperately specify the recipient of that e-mail: I use E-Mail in my local network, but do not allow for any outgoing traffic.

[JDevlieghere]

In D46192#1082049, @lebedev.ri wrote:

This should probably go into llvm/utils, not the root.

It would also be good to have the opposite script - "given git diff / commits, whom should i add as reviewers?"

I once wrote the following one-liner to do this. It works pretty well but it can be *very* slow. It will suggest a top 10 of reviewers for your staged changes.

git diff --name-only --cached | xargs -n 1 git blame --porcelain | grep \"^author \" | sort | uniq -c | sort -nr | head -10

It could evaluate CODE_OWNERS.txt, git log / git shortlog -sne of the changed files, git blame.

[javed.absar]

Hi Kristof. Thanks for this.

One more heuristic that might help getting quick reviews:

3. If there is overlap between the files touched by the patch-under-review and and a 'reviewer' of recent changes to same file, that person may be a good reviewer.

This heuristic is helpful where original implementor may have moved on, but new active reviewers in same area are happy to review.
Some commit messages have 'reviewers' listed and that could be used.

[kristof.beyls]

In D46192#1082214, @philip.pfaffe wrote:

Do you plan on actually commiting this to the repository, or would you like any feedback nontheless?

I definitely welcome any feedback.
I wasn't sure on committing to the repo, but with the comments coming in and the new feature requests popping up already, it seems like I should prepare to commit it to the repository, so that new features can be developed and shared incrementally.
As @lebedev.ri suggested, it probably should go somewhere in llvm/utils. Given there seems to be scope for other reviewing-related scripts, maybe it should go in a separate directory, e.g. llvm/utils/Reviewing.

One usability nit that I see is send_emails(). It would be great to be able to configure the SMTP connection, to disable the feature altogether, and to seperately specify the recipient of that e-mail: I use E-Mail in my local network, but do not allow for any outgoing traffic.

Agreed, more flexibility is needed on the emailing-feature of the script. I assume that the script is also too hard-coded in some other locations as this is really a prototype containing what was needed in my environment.

I hope to find some time next week to look both into finding a good location to commit this script and the email-feature improvements you suggest .

[kristof.beyls]

In D46192#1083794, @javed.absar wrote:

Hi Kristof. Thanks for this.

One more heuristic that might help getting quick reviews:

3. If there is overlap between the files touched by the patch-under-review and and a 'reviewer' of recent changes to same file, that person may be a good reviewer.

This heuristic is helpful where original implementor may have moved on, but new active reviewers in same area are happy to review.
Some commit messages have 'reviewers' listed and that could be used.

Interesting idea, thanks for sharing, Javed!
I'll need a bit of time to think on how this could be best implemented.

Thanks!

Kristof

[kristof.beyls]

In D46192#1082638, @JDevlieghere wrote:

In D46192#1082049, @lebedev.ri wrote:

This should probably go into llvm/utils, not the root.

It would also be good to have the opposite script - "given git diff / commits, whom should i add as reviewers?"

I once wrote the following one-liner to do this. It works pretty well but it can be *very* slow. It will suggest a top 10 of reviewers for your staged changes.

git diff --name-only --cached | xargs -n 1 git blame --porcelain | grep \"^author \" | sort | uniq -c | sort -nr | head -10

It could evaluate CODE_OWNERS.txt, git log / git shortlog -sne of the changed files, git blame.

Thanks for sharing, Jonas!

Your one-liner could become the start of a new "suggest_reviewers" script?

[MaskRay]

Can this script be made Python 3 compatible? Here is one place where running it under python3 will fail

    people_to_look_for = [e.decode('utf-8') for e in args.email_addresses]
AttributeError: 'str' object has no attribute 'decode'

[kristof.beyls]

Updating diff to make it add the script to llvm/utils/Reviewing.
I chose to put it in a new folder "Reviewing", since ideas for further scripts related to reviewing have been made on this thread.

For all of the feature requests raised so far on this review thread: I think they all make sense, but also think it's best to add them incrementally. In other words: commit the script as is and iterate through the feature requests and add them as individual commits.

It'd be nice if one of the reviewers could share an opinion on whether the above sounds like a good plan. If so, I'll go ahead with the above plan.

Thanks,

Kristof

[lebedev.ri]

utils/Reviewing/find_interesting_reviews.py

2

Can #!/usr/bin/env python be prepended as the first line?

2

Also, the standard licensing blurb at the top is missing.

[kristof.beyls]

utils/Reviewing/find_interesting_reviews.py

2

Added in latest updated patch.

2

I tried to find an example in any of the other python scripts under util, but I couldn't see any licensing blurbs in any of the existing python files.
Maybe best not to invent a licensing blurb in Python syntax here and just leave it as is?

[JDevlieghere]

There's a few small inconsistencies (such as spaces after print, between operators, etc) that could be automatically solved by running a formatter (e.g. YAPF) over the source code.

utils/Reviewing/find_interesting_reviews.py

4

Nit: can we sort the imports alphabetically?

85

Nit: You have different approaches to string formatting throughout this file. Personally I prefer the "".format() approach but as long as we're consistent I'd be happy.

150

Nit: We usually make comments full sentences in LLVM.

433

I'm guessing the commented out code is for logging? Maybe we could use the logger for this with a low verbosity?

533

Maybe turn this into a constant?

590

if __name__ == "__main__":
    main()

[kristof.beyls]

Many thanks for the detailed review, Jonas!
I think I've addressed all your comments - very useful feedback!

[kristof.beyls]

Oh - and I indeed also ran the source code through YAPF - thanks for that pointer, I didn't know that existed.

[JDevlieghere]

Awesome, thanks Kristof! Two more nits, but overall this looks good to me.

utils/Reviewing/find_interesting_reviews.py

9

Usually import and from are grouped/sorted separately:

import a
import b 
import c
from a import foo
from b import bar

94

If this is a constant I'd make it all caps. Same for the class below.

utils/Reviewing/find_interesting_reviews.py

+#!/usr/bin/env python
+
+import argparse
+from datetime import datetime, timedelta
+import email.mime.multipart
+import email.mime.text
+import logging
+import os.path
+from phabricator import Phabricator
+import pickle
+import re
+import smtplib
+import subprocess
+import sys
+
+# Setting up a virtualenv to run this script can be done by running the
+# following commands:
+# $ virtualenv venv
+# $ . ./venv/bin/activate
+# $ pip install Phabricator
+
+GIT_REPO_METADATA = (("llvm", "https://llvm.org/git/llvm.git"), )
+
+# The below PhabXXX classes represent objects as modelled by Phabricator.
+# The classes can be serialized to disk, to try and make sure that we don't
+# needlessly have to re-fetch lots of data from Phabricator, as that would
+# make this script unusably slow.
+
+
+class PhabObject:
+    object_kind = None
+
+    def __init__(self, id):
+        self.id = id
+
+
+class PhabObjectCache:
+    def __init__(self, PhabObjectClass):
+        self.PhabObjectClass = PhabObjectClass
+        self.most_recent_info = None
+        self.oldest_info = None
+        self.id2PhabObjects = {}
+
+    def get_name(self):
+        return self.PhabObjectClass.object_kind + "sCache"
+
+    def get(self, id):
+        if id not in self.id2PhabObjects:
+            self.id2PhabObjects[id] = self.PhabObjectClass(id)
+        return self.id2PhabObjects[id]
+
+    def get_ids_in_cache(self):
+        return self.id2PhabObjects.keys()
+
+    def get_objects(self):
+        return self.id2PhabObjects.values()
+
+    DEFAULT_DIRECTORY = "PhabObjectCache"
+
+    def _get_pickle_name(self, directory):
+        file_name = "Phab" + self.PhabObjectClass.object_kind + "s.pickle"
+        return os.path.join(directory, file_name)
+
+    def populate_cache_from_disk(self, directory=DEFAULT_DIRECTORY):
+        """
+        FIXME: consider if serializing to JSON would bring interoperability
+        advantages over serializing to pickle.
+        """
+        try:
+            f = open(self._get_pickle_name(directory), "rb")
+        except IOError as err:
+            print("Could not find cache. Error message: {0}. Continuing..."
+                  .format(err))
+        else:
+            with f:
+                try:
+                    d = pickle.load(f)
+                    self.__dict__.update(d)
+                except EOFError as err:
+                    print("Cache seems to be corrupt. " +
+                          "Not using cache. Error message: {0}".format(err))
+
+    def write_cache_to_disk(self, directory=DEFAULT_DIRECTORY):
+        if not os.path.exists(directory):
+            os.makedirs(directory)
+        with open(self._get_pickle_name(directory), "wb") as f:
+            pickle.dump(self.__dict__, f)
+        print("wrote cache to disk, most_recent_info= {0}".format(
+            datetime.fromtimestamp(self.most_recent_info)
+            if self.most_recent_info is not None else None))
+
+
+class PhabReview(PhabObject):
+    object_kind = "Review"
+
+    def __init__(self, id):
+        PhabObject.__init__(self, id)
+
+    def update(self, title, dateCreated, dateModified, author):
+        self.title = title
+        self.dateCreated = dateCreated
+        self.dateModified = dateModified
+        self.author = author
+
+    def setPhabDiffs(self, phabDiffs):
+        self.phabDiffs = phabDiffs
+
+
+class PhabUser(PhabObject):
+    object_kind = "User"
+
+    def __init__(self, id):
+        PhabObject.__init__(self, id)
+
+    def update(self, phid, realName):
+        self.phid = phid
+        self.realName = realName
+
+
+class PhabHunk:
+    def __init__(self, rest_api_hunk):
+        self.oldOffset = int(rest_api_hunk["oldOffset"])
+        self.oldLength = int(rest_api_hunk["oldLength"])
+        # self.actual_lines_changed_offset will contain the offsets of the
+        # lines that were changed in this hunk.
+        self.actual_lines_changed_offset = []
+        offset = self.oldOffset
+        inHunk = False
+        hunkStart = -1
+        contextLines = 3
+        for line in rest_api_hunk["corpus"].split("\n"):
+            if line.startswith("+"):
+                # line is a new line that got introduced in this patch.
+                # Do not record it as a changed line.
+                if inHunk is False:
+                    inHunk = True
+                    hunkStart = max(self.oldOffset, offset - contextLines)
+                continue
+            if line.startswith("-"):
+                # line was changed or removed from the older version of the
+                # code. Record it as a changed line.
+                if inHunk is False:
+                    inHunk = True
+                    hunkStart = max(self.oldOffset, offset - contextLines)
+                offset += 1
+                continue
+            # line is a context line.
+            if inHunk is True:
+                inHunk = False
+                hunkEnd = offset + contextLines
+                self.actual_lines_changed_offset.append((hunkStart, hunkEnd))
+            offset += 1
+        if inHunk is True:
+            hunkEnd = offset + contextLines
+            self.actual_lines_changed_offset.append((hunkStart, hunkEnd))
+
+        # The above algorithm could result in adjacent or overlapping ranges
+        # being recorded into self.actual_lines_changed_offset.
+        # Merge the adjacent and overlapping ranges in there:
+        t = []
+        lastRange = None
+        for start, end in self.actual_lines_changed_offset + \
+                [(sys.maxsize, sys.maxsize)]:
+            if lastRange is None:
+                lastRange = (start, end)
+            else:
+                if lastRange[1] >= start:
+                    lastRange = (lastRange[0], end)
+                else:
+                    t.append(lastRange)
+                    lastRange = (start, end)
+        self.actual_lines_changed_offset = t
+
+
+class PhabChange:
+    def __init__(self, rest_api_change):
+        self.oldPath = rest_api_change["oldPath"]
+        self.hunks = [PhabHunk(h) for h in rest_api_change["hunks"]]
+
+
+class PhabDiff(PhabObject):
+    object_kind = "Diff"
+
+    def __init__(self, id):
+        PhabObject.__init__(self, id)
+
+    def update(self, rest_api_results):
+        self.revisionID = rest_api_results["revisionID"]
+        self.dateModified = int(rest_api_results["dateModified"])
+        self.dateCreated = int(rest_api_results["dateCreated"])
+        self.changes = [PhabChange(c) for c in rest_api_results["changes"]]
+
+
+class ReviewsCache(PhabObjectCache):
+    def __init__(self):
+        PhabObjectCache.__init__(self, PhabReview)
+
+
+class UsersCache(PhabObjectCache):
+    def __init__(self):
+        PhabObjectCache.__init__(self, PhabUser)
+
+
+reviews_cache = ReviewsCache()
+users_cache = UsersCache()
+
+
+def init_phab_connection():
+    phab = Phabricator()
+    phab.update_interfaces()
+    return phab
+
+
+def update_cached_info(phab, cache, phab_query, order, record_results,
+                       max_nr_entries_per_fetch, max_nr_days_to_cache):
+    q = phab
+    LIMIT = max_nr_entries_per_fetch
+    for query_step in phab_query:
+        q = getattr(q, query_step)
+    results = q(order=order, limit=LIMIT)
+    most_recent_info, oldest_info = record_results(cache, results, phab)
+    oldest_info_to_fetch = datetime.fromtimestamp(most_recent_info) - \
+        timedelta(days=max_nr_days_to_cache)
+    most_recent_info_overall = most_recent_info
+    cache.write_cache_to_disk()
+    after = results["cursor"]["after"]
+    print("after: {0!r}".format(after))
+    print("most_recent_info: {0}".format(
+        datetime.fromtimestamp(most_recent_info)))
+    while (after is not None
+           and datetime.fromtimestamp(oldest_info) > oldest_info_to_fetch):
+        need_more_older_data = \
+            (cache.oldest_info is None or
+             datetime.fromtimestamp(cache.oldest_info) > oldest_info_to_fetch)
+        print(("need_more_older_data={0} cache.oldest_info={1} " +
+               "oldest_info_to_fetch={2}").format(
+                   need_more_older_data,
+                   datetime.fromtimestamp(cache.oldest_info)
+                   if cache.oldest_info is not None else None,
+                   oldest_info_to_fetch))
+        need_more_newer_data = \
+            (cache.most_recent_info is None or
+             cache.most_recent_info < most_recent_info)
+        print(("need_more_newer_data={0} cache.most_recent_info={1} " +
+               "most_recent_info={2}")
+              .format(need_more_newer_data, cache.most_recent_info,
+                      most_recent_info))
+        if not need_more_older_data and not need_more_newer_data:
+            break
+        results = q(order=order, after=after, limit=LIMIT)
+        most_recent_info, oldest_info = record_results(cache, results, phab)
+        after = results["cursor"]["after"]
+        print("after: {0!r}".format(after))
+        print("most_recent_info: {0}".format(
+            datetime.fromtimestamp(most_recent_info)))
+        cache.write_cache_to_disk()
+    cache.most_recent_info = most_recent_info_overall
+    if after is None:
+        # We did fetch all records. Mark the cache to contain all info since
+        # the start of time.
+        oldest_info = 0
+    cache.oldest_info = oldest_info
+    cache.write_cache_to_disk()
+
+
+def record_reviews(cache, reviews, phab):
+    most_recent_info = None
+    oldest_info = None
+    for reviewInfo in reviews["data"]:
+        if reviewInfo["type"] != "DREV":
+            continue
+        id = reviewInfo["id"]
+        # phid = reviewInfo["phid"]
+        dateModified = int(reviewInfo["fields"]["dateModified"])
+        dateCreated = int(reviewInfo["fields"]["dateCreated"])
+        title = reviewInfo["fields"]["title"]
+        author = reviewInfo["fields"]["authorPHID"]
+        phabReview = cache.get(id)
+        if "dateModified" not in phabReview.__dict__ or \
+           dateModified > phabReview.dateModified:
+            diff_results = phab.differential.querydiffs(revisionIDs=[id])
+            diff_ids = sorted(diff_results.keys())
+            phabDiffs = []
+            for diff_id in diff_ids:
+                diffInfo = diff_results[diff_id]
+                d = PhabDiff(diff_id)
+                d.update(diffInfo)
+                phabDiffs.append(d)
+            phabReview.update(title, dateCreated, dateModified, author)
+            phabReview.setPhabDiffs(phabDiffs)
+            print("Updated D{0} modified on {1} ({2} diffs)".format(
+                id, datetime.fromtimestamp(dateModified), len(phabDiffs)))
+
+        if most_recent_info is None:
+            most_recent_info = dateModified
+        elif most_recent_info < dateModified:
+            most_recent_info = dateModified
+
+        if oldest_info is None:
+            oldest_info = dateModified
+        elif oldest_info > dateModified:
+            oldest_info = dateModified
+    return most_recent_info, oldest_info
+
+
+def record_users(cache, users, phab):
+    most_recent_info = None
+    oldest_info = None
+    for info in users["data"]:
+        if info["type"] != "USER":
+            continue
+        id = info["id"]
+        phid = info["phid"]
+        dateModified = int(info["fields"]["dateModified"])
+        # dateCreated = int(info["fields"]["dateCreated"])
+        realName = info["fields"]["realName"]
+        phabUser = cache.get(id)
+        phabUser.update(phid, realName)
+        if most_recent_info is None:
+            most_recent_info = dateModified
+        elif most_recent_info < dateModified:
+            most_recent_info = dateModified
+        if oldest_info is None:
+            oldest_info = dateModified
+        elif oldest_info > dateModified:
+            oldest_info = dateModified
+    return most_recent_info, oldest_info
+
+
+PHABCACHESINFO = ((reviews_cache, ("differential", "revision", "search"),
+                   "updated", record_reviews, 5, 7),
+                  (users_cache, ("user", "search"), "newest", record_users,
+                   100, 1000))
+
+
+def load_cache():
+    for cache, phab_query, order, record_results, _, _ in PHABCACHESINFO:
+        cache.populate_cache_from_disk()
+        print("Loaded {0} nr entries: {1}".format(
+            cache.get_name(), len(cache.get_ids_in_cache())))
+        print("Loaded {0} has most recent info: {1}".format(
+            cache.get_name(),
+            datetime.fromtimestamp(cache.most_recent_info)
+            if cache.most_recent_info is not None else None))
+
+
+def update_cache(phab):
+    load_cache()
+    for cache, phab_query, order, record_results, max_nr_entries_per_fetch, \
+            max_nr_days_to_cache in PHABCACHESINFO:
+        update_cached_info(phab, cache, phab_query, order, record_results,
+                           max_nr_entries_per_fetch, max_nr_days_to_cache)
+        ids_in_cache = cache.get_ids_in_cache()
+        print("{0} objects in {1}".format(len(ids_in_cache), cache.get_name()))
+        cache.write_cache_to_disk()
+
+
+def get_most_recent_reviews(days):
+    newest_reviews = sorted(
+        reviews_cache.get_objects(), key=lambda r: -r.dateModified)
+    if len(newest_reviews) == 0:
+        return newest_reviews
+    most_recent_review_time = \
+        datetime.fromtimestamp(newest_reviews[0].dateModified)
+    cut_off_date = most_recent_review_time - timedelta(days=days)
+    result = []
+    for review in newest_reviews:
+        if datetime.fromtimestamp(review.dateModified) < cut_off_date:
+            return result
+        result.append(review)
+    return result
+
+
+# All of the above code is about fetching data from Phabricator and caching it
+# on local disk. The below code contains the actual "business logic" for this
+# script.
+
+_userphid2realname = None
+
+
+def get_real_name_from_author(user_phid):
+    global _userphid2realname
+    if _userphid2realname is None:
+        _userphid2realname = {}
+        for user in users_cache.get_objects():
+            _userphid2realname[user.phid] = user.realName
+    return _userphid2realname.get(user_phid, "unknown")
+
+
+def print_most_recent_reviews(phab, days, filter_reviewers):
+    msgs = []
+
+    def add_msg(msg):
+        msgs.append(msg)
+        print(msg)
+
+    newest_reviews = get_most_recent_reviews(days)
+    add_msg("These are the reviews that look interesting to be reviewed. " +
+            "The report below has 2 sections. The first " +
+            "section is organized per review; the second section is organized "
+            + "per potential reviewer.\n")
+    oldest_review = newest_reviews[-1] if len(newest_reviews) > 0 else None
+    oldest_datetime = \
+        datetime.fromtimestamp(oldest_review.dateModified) \
+        if oldest_review else None
+    add_msg(("The report below is based on analyzing the reviews that got " +
+             "touched in the past {0} days (since {1}). " +
+             "The script found {2} such reviews.\n").format(
+                 days, oldest_datetime, len(newest_reviews)))
+    reviewer2reviews_and_scores = {}
+    for i, review in enumerate(newest_reviews):
+        matched_reviewers = find_reviewers_for_review(review)
+        matched_reviewers = filter_reviewers(matched_reviewers)
+        if len(matched_reviewers) == 0:
+            continue
+        add_msg(("{0:>3}. https://reviews.llvm.org/D{1} by {2}\n     {3}\n" +
+                 "     Last updated on {4}").format(
+                     i, review.id,
+                     get_real_name_from_author(review.author), review.title,
+                     datetime.fromtimestamp(review.dateModified)))
+        for reviewer, scores in matched_reviewers:
+            add_msg("    potential reviewer {0}, score {1}".format(
+                reviewer,
+                "(" + "/".join(["{0:.1f}%".format(s) for s in scores]) + ")"))
+            if reviewer not in reviewer2reviews_and_scores:
+                reviewer2reviews_and_scores[reviewer] = []
+            reviewer2reviews_and_scores[reviewer].append((review, scores))
+
+    # Print out a summary per reviewer.
+    for reviewer in sorted(reviewer2reviews_and_scores.keys()):
+        reviews_and_scores = reviewer2reviews_and_scores[reviewer]
+        reviews_and_scores.sort(key=lambda rs: rs[1], reverse=True)
+        add_msg("\n\nSUMMARY FOR {0} (found {1} reviews):".format(
+            reviewer, len(reviews_and_scores)))
+        for review, scores in reviews_and_scores:
+            add_msg("[{0}] https://reviews.llvm.org/D{1} '{2}' by {3}".format(
+                "/".join(["{0:.1f}%".format(s) for s in scores]), review.id,
+                review.title, get_real_name_from_author(review.author)))
+    return "\n".join(msgs)
+
+
+def get_git_cmd_output(cmd):
+    output = None
+    try:
+        logging.debug(cmd)
+        output = subprocess.check_output(
+            cmd, shell=True, stderr=subprocess.STDOUT)
+    except subprocess.CalledProcessError as e:
+        logging.debug(str(e))
+    if output is None:
+        return None
+    return output.decode("utf-8", errors='ignore')
+
+
+reAuthorMail = re.compile("^author-mail <([^>]*)>.*$")
+
+
+def parse_blame_output_line_porcelain(blame_output):
+    email2nr_occurences = {}
+    if blame_output is None:
+        return email2nr_occurences
+    for line in blame_output.split('\n'):
+        m = reAuthorMail.match(line)
+        if m:
+            author_email_address = m.group(1)
+            if author_email_address not in email2nr_occurences:
+                email2nr_occurences[author_email_address] = 1
+            else:
+                email2nr_occurences[author_email_address] += 1
+    return email2nr_occurences
+
+
+def find_reviewers_for_diff_heuristic(diff):
+    # Heuristic 1: assume good reviewers are the ones that touched the same
+    # lines before as this patch is touching.
+    # Heuristic 2: assume good reviewers are the ones that touched the same
+    # files before as this patch is touching.
+    reviewers2nr_lines_touched = {}
+    reviewers2nr_files_touched = {}
+    # Assume last revision before diff was modified is the revision the diff
+    # applies to.
+    git_repo = "git_repos/llvm"
+    cmd = 'git -C {0} rev-list -n 1 --before="{1}" master'.format(
+        git_repo,
+        datetime.fromtimestamp(
+            diff.dateModified).strftime("%Y-%m-%d %H:%M:%s"))
+    base_revision = get_git_cmd_output(cmd).strip()
+    logging.debug("Base revision={0}".format(base_revision))
+    for change in diff.changes:
+        path = change.oldPath
+        # Compute heuristic 1: look at context of patch lines.
+        for hunk in change.hunks:
+            for start_line, end_line in hunk.actual_lines_changed_offset:
+                # Collect git blame results for authors in those ranges.
+                cmd = ("git -C {0} blame --encoding=utf-8 --date iso -f -e " +
+                       "-w --line-porcelain -L {1},{2} {3} -- {4}").format(
+                           git_repo, start_line, end_line, base_revision, path)
+                blame_output = get_git_cmd_output(cmd)
+                for reviewer, nr_occurences in \
+                        parse_blame_output_line_porcelain(blame_output).items():
+                    if reviewer not in reviewers2nr_lines_touched:
+                        reviewers2nr_lines_touched[reviewer] = 0
+                    reviewers2nr_lines_touched[reviewer] += nr_occurences
+        # Compute heuristic 2: don't look at context, just at files touched.
+        # Collect git blame results for authors in those ranges.
+        cmd = ("git -C {0} blame --encoding=utf-8 --date iso -f -e -w " +
+               "--line-porcelain {1} -- {2}").format(git_repo, base_revision,
+                                                     path)
+        blame_output = get_git_cmd_output(cmd)
+        for reviewer, nr_occurences in parse_blame_output_line_porcelain(
+                blame_output).items():
+            if reviewer not in reviewers2nr_files_touched:
+                reviewers2nr_files_touched[reviewer] = 0
+            reviewers2nr_files_touched[reviewer] += 1
+
+    # Compute "match scores"
+    total_nr_lines = sum(reviewers2nr_lines_touched.values())
+    total_nr_files = len(diff.changes)
+    reviewers_matchscores = \
+        [(reviewer,
+          (reviewers2nr_lines_touched.get(reviewer, 0)*100.0/total_nr_lines
+           if total_nr_lines != 0 else 0,
+           reviewers2nr_files_touched[reviewer]*100.0/total_nr_files
+           if total_nr_files != 0 else 0))
+         for reviewer, nr_lines
+         in reviewers2nr_files_touched.items()]
+    reviewers_matchscores.sort(key=lambda i: i[1], reverse=True)
+    return reviewers_matchscores
+
+
+def find_reviewers_for_review(review):
+    # Process the newest diff first.
+    diffs = sorted(
+        review.phabDiffs, key=lambda d: d.dateModified, reverse=True)
+    if len(diffs) == 0:
+        return
+    diff = diffs[0]
+    matched_reviewers = find_reviewers_for_diff_heuristic(diff)
+    # Show progress, as this is a slow operation:
+    sys.stdout.write('.')
+    sys.stdout.flush()
+    logging.debug("matched_reviewers: {0}".format(matched_reviewers))
+    return matched_reviewers
+
+
+def update_git_repos():
+    git_repos_directory = "git_repos"
+    for name, url in GIT_REPO_METADATA:
+        dirname = os.path.join(git_repos_directory, name)
+        if not os.path.exists(dirname):
+            cmd = "git clone {0} {1}".format(url, dirname)
+            output = get_git_cmd_output(cmd)
+        cmd = "git -C {0} pull --rebase".format(dirname)
+        output = get_git_cmd_output(cmd)
+
+
+def send_emails(email_addresses, msg):
+    s = smtplib.SMTP()
+    s.connect()
+    for email_address in email_addresses:
+        email_msg = email.mime.multipart.MIMEMultipart()
+        email_msg['From'] = ''
+        email_msg['To'] = email_address
+        email_msg['Subject'] = 'LLVM patches you may be able to review.'
+        email_msg.attach(email.mime.text.MIMEText(msg, 'plain'))
+        # python 3.x: s.send_message(email_msg)
+        s.sendmail(email_msg['From'], email_msg['To'], msg)
+    s.quit()
+
+
+def filter_reviewers_to_report_for(people_to_look_for):
+    # The below is just an example filter, to only report potential reviews
+    # to do for the people that will receive the report email.
+    return lambda potential_reviewers: [r for r in potential_reviewers
+                                        if r[0] in people_to_look_for]
+
+
+def main():
+    parser = argparse.ArgumentParser(
+        description='Match open reviews to potential reviewers.')
+    parser.add_argument(
+        '--no-update-cache',
+        dest='update_cache',
+        action='store_false',
+        default=True,
+        help='Do not update cached Phabricator objects')
+    parser.add_argument(
+        'email_addresses',
+        nargs='*',
+        help="The email addresses (as known by LLVM git) of " +
+        "the people to look for reviews for.")
+    parser.add_argument('--verbose', '-v', action='count')
+
+    args = parser.parse_args()
+
+    if args.verbose >= 1:
+        logging.basicConfig(level=logging.DEBUG)
+
+    people_to_look_for = [e.decode('utf-8') for e in args.email_addresses]
+
+    phab = init_phab_connection()
+
+    if args.update_cache:
+        update_cache(phab)
+
+    load_cache()
+    update_git_repos()
+    msg = print_most_recent_reviews(
+        phab,
+        days=1,
+        filter_reviewers=filter_reviewers_to_report_for(people_to_look_for))
+    send_emails(people_to_look_for, msg)
+
+
+if __name__ == "__main__":
+    main()