import logging
from itertools import chain
from cPickle import dumps
import re
import bson
import tg
from pylons import g
from ming.base import Object
from ming.orm import mapper, session
from allura.lib import utils
from allura.lib import helpers as h
from allura.model.repo import CommitDoc, TreeDoc, TreesDoc, DiffInfoDoc
from allura.model.repo import LastCommitDoc, CommitRunDoc
from allura.model.repo import Commit
from allura.model.index import ArtifactReferenceDoc, ShortlinkDoc
log = logging.getLogger(__name__)
QSIZE=100
def refresh_repo(repo, all_commits=False, notify=True):
all_commit_ids = commit_ids = list(repo.all_commit_ids())
if not commit_ids:
# the repo is empty, no need to continue
return
new_commit_ids = unknown_commit_ids(commit_ids)
stats_log = h.log_action(log, 'commit')
for ci in new_commit_ids:
stats_log.info(
'',
meta=dict(
module='scm-%s' % repo.repo_id,
read='0'))
if not all_commits:
# Skip commits that are already in the DB
commit_ids = new_commit_ids
log.info('Refreshing %d commits on %s', len(commit_ids), repo.full_fs_path)
# Refresh commits
seen = set()
for i, oid in enumerate(commit_ids):
repo.refresh_commit_info(oid, seen, not all_commits)
if (i+1) % 100 == 0:
log.info('Refresh commit info %d: %s', (i+1), oid)
refresh_commit_repos(all_commit_ids, repo)
# Refresh child references
for i, oid in enumerate(commit_ids):
ci = CommitDoc.m.find(dict(_id=oid), validate=False).next()
refresh_children(ci)
if (i+1) % 100 == 0:
log.info('Refresh child info %d for parents of %s', (i+1), ci._id)
if repo.tool.lower() != 'svn':
# Refresh commit runs
commit_run_ids = commit_ids
# Check if the CommitRuns for the repo are in a good state by checking for
# a CommitRunDoc that contains the last known commit. If there isn't one,
# the CommitRuns for this repo are in a bad state - rebuild them entirely.
if commit_run_ids != all_commit_ids:
last_commit = last_known_commit_id(all_commit_ids, new_commit_ids)
log.info('Last known commit id: %s', last_commit)
if not CommitRunDoc.m.find(dict(commit_ids=last_commit)).count():
log.info('CommitRun incomplete, rebuilding with all commits')
commit_run_ids = all_commit_ids
log.info('Starting CommitRunBuilder for %s', repo.full_fs_path)
rb = CommitRunBuilder(commit_run_ids)
rb.run()
rb.cleanup()
log.info('Finished CommitRunBuilder for %s', repo.full_fs_path)
# Refresh trees
# Like diffs below, pre-computing trees for some SCMs is too expensive,
# so we skip it here, then do it on-demand later.
if repo._refresh_precompute:
cache = {}
for i, oid in enumerate(commit_ids):
ci = CommitDoc.m.find(dict(_id=oid), validate=False).next()
cache = refresh_commit_trees(ci, cache)
if (i+1) % 100 == 0:
log.info('Refresh commit trees %d: %s', (i+1), ci._id)
# Compute diffs
cache = {}
# Have to compute_diffs() for all commits to ensure that LastCommitDocs
# are set properly for forked repos. For some SCMs, compute_diffs()
# we don't want to pre-compute the diffs because that would be too
# expensive, so we skip them here and do them on-demand with caching.
if repo._refresh_precompute:
for i, oid in enumerate(reversed(all_commit_ids)):
ci = CommitDoc.m.find(dict(_id=oid), validate=False).next()
compute_diffs(repo._id, cache, ci)
if (i+1) % 100 == 0:
log.info('Compute diffs %d: %s', (i+1), ci._id)
log.info('Refresh complete for %s', repo.full_fs_path)
g.post_event(
'repo_refreshed',
commit_number=len(commit_ids),
new=bool(new_commit_ids))
# Send notifications
if notify:
send_notifications(repo, commit_ids)
def refresh_commit_trees(ci, cache):
'''Refresh the list of trees included withn a commit'''
if ci.tree_id is None: return cache
trees_doc = TreesDoc(dict(
_id=ci._id,
tree_ids = list(trees(ci.tree_id, cache))))
trees_doc.m.save(safe=False)
new_cache = dict(
(oid, cache[oid])
for oid in trees_doc.tree_ids)
return new_cache
def refresh_commit_repos(all_commit_ids, repo):
'''Refresh the list of repositories within which a set of commits are
contained'''
for oids in utils.chunked_iter(all_commit_ids, QSIZE):
for ci in CommitDoc.m.find(dict(
_id={'$in':list(oids)},
repo_ids={'$ne': repo._id})):
oid = ci._id
ci.repo_ids.append(repo._id)
index_id = 'allura.model.repo.Commit#' + oid
ref = ArtifactReferenceDoc(dict(
_id=index_id,
artifact_reference=dict(
cls=bson.Binary(dumps(Commit)),
project_id=repo.app.config.project_id,
app_config_id=repo.app.config._id,
artifact_id=oid),
references=[]))
link0 = ShortlinkDoc(dict(
_id=bson.ObjectId(),
ref_id=index_id,
project_id=repo.app.config.project_id,
app_config_id=repo.app.config._id,
link=repo.shorthand_for_commit(oid)[1:-1],
url=repo.url_for_commit(oid)))
# Always create a link for the full commit ID
link1 = ShortlinkDoc(dict(
_id=bson.ObjectId(),
ref_id=index_id,
project_id=repo.app.config.project_id,
app_config_id=repo.app.config._id,
link=oid,
url=repo.url_for_commit(oid)))
ci.m.save(safe=False, validate=False)
ref.m.save(safe=False, validate=False)
link0.m.save(safe=False, validate=False)
link1.m.save(safe=False, validate=False)
def refresh_children(ci):
'''Refresh the list of children of the given commit'''
CommitDoc.m.update_partial(
dict(_id={'$in': ci.parent_ids}),
{'$addToSet': dict(child_ids=ci._id)},
multi=True)
class CommitRunBuilder(object):
'''Class used to build up linear runs of single-parent commits'''
def __init__(self, commit_ids):
self.commit_ids = commit_ids
self.run_index = {} # by commit ID
self.runs = {} # by run ID
self.reasons = {} # reasons to stop merging runs
def run(self):
'''Build up the runs'''
for oids in utils.chunked_iter(self.commit_ids, QSIZE):
oids = list(oids)
for ci in CommitDoc.m.find(dict(_id={'$in':oids})):
if ci._id in self.run_index: continue
self.run_index[ci._id] = ci._id
self.runs[ci._id] = CommitRunDoc(dict(
_id=ci._id,
parent_commit_ids=ci.parent_ids,
commit_ids=[ci._id],
commit_times=[ci.authored['date']]))
self.merge_runs()
log.info('%d runs', len(self.runs))
for rid, run in sorted(self.runs.items()):
log.info('%32s: %r', self.reasons.get(rid, 'none'), run._id)
for run in self.runs.itervalues():
run.m.save()
return self.runs
def _all_runs(self):
'''Find all runs containing this builder's commit IDs'''
runs = {}
for oids in utils.chunked_iter(self.commit_ids, QSIZE):
oids = list(oids)
for run in CommitRunDoc.m.find(dict(commit_ids={'$in': oids})):
runs[run._id] = run
for run in CommitRunDoc.m.find(dict(parent_commit_ids={'$in': oids})):
runs[run._id] = run
seen_run_ids = set()
runs = runs.values()
while runs:
run = runs.pop()
if run._id in seen_run_ids: continue
seen_run_ids.add(run._id)
yield run
for run in CommitRunDoc.m.find(
dict(commit_ids={'$in':run.parent_commit_ids})):
runs.append(run)
def cleanup(self):
'''Delete non-maximal runs and merge any new runs with existing runs'''
runs = dict(
(run['commit_ids'][0], run)
for run in self._all_runs())
for rid, run in runs.items():
p_cis = run['parent_commit_ids']
if len(p_cis) != 1: continue
parent_run = runs.get(p_cis[0], None)
if parent_run is None: continue
run['commit_ids'] += parent_run['commit_ids']
run['commit_times'] += parent_run['commit_times']
run['parent_commit_ids'] = parent_run['parent_commit_ids']
run.m.save()
parent_run.m.delete()
del runs[p_cis[0]]
for run1 in runs.values():
# if run1 is a subset of another run, delete it
if CommitRunDoc.m.find(dict(commit_ids={'$all': run1.commit_ids},
_id={'$ne': run1._id})).count():
log.info('... delete %r (subset of another run)', run1)
run1.m.delete()
continue
for run2 in CommitRunDoc.m.find(dict(
commit_ids=run1.commit_ids[0])):
if run1._id == run2._id: continue
log.info('... delete %r (part of %r)', run2, run1)
run2.m.delete()
def merge_runs(self):
'''Find partial runs that may be merged and merge them'''
while True:
for run_id, run in self.runs.iteritems():
if len(run.parent_commit_ids) != 1:
self.reasons[run_id] = '%d parents' % len(run.parent_commit_ids)
continue
p_oid = run.parent_commit_ids[0]
p_run_id = self.run_index.get(p_oid)
if p_run_id is None:
self.reasons[run_id] = 'parent commit not found'
continue
p_run = self.runs.get(p_run_id)
if p_run is None:
self.reasons[run_id] = 'parent run not found'
continue
if p_run.commit_ids[0] != p_oid:
self.reasons[run_id] = 'parent does not start with parent commit'
continue
run.commit_ids += p_run.commit_ids
run.commit_times += p_run.commit_times
run.parent_commit_ids = p_run.parent_commit_ids
for oid in p_run.commit_ids:
self.run_index[oid] = run_id
break
else:
break
del self.runs[p_run_id]
def trees(id, cache):
'''Recursively generate the list of trees contained within a given tree ID'''
yield id
entries = cache.get(id, None)
if entries is None:
t = TreeDoc.m.get(_id=id)
entries = [ o.id for o in t.tree_ids ]
cache[id] = entries
for i in entries:
for x in trees(i, cache):
yield x
def unknown_commit_ids(all_commit_ids):
'''filter out all commit ids that have already been cached'''
result = []
for chunk in utils.chunked_iter(all_commit_ids, QSIZE):
chunk = list(chunk)
q = CommitDoc.m.find(dict(_id={'$in':chunk}))
known_commit_ids = set(ci._id for ci in q)
result += [ oid for oid in chunk if oid not in known_commit_ids ]
return result
def compute_diffs(repo_id, tree_cache, rhs_ci):
'''compute simple differences between a commit and its first parent'''
if rhs_ci.tree_id is None: return tree_cache
def _walk_tree(tree, tree_index):
for x in tree.blob_ids: yield x.id
for x in tree.other_ids: yield x.id
for x in tree.tree_ids:
yield x.id
for xx in _walk_tree(tree_index[x.id], tree_index):
yield xx
treedoc = TreesDoc.m.get(_id=rhs_ci._id)
# FIXME: There are cases of missing TreesDoc records in production
# that should be fixed, but this is a quick-and-dirty patch
# to at least staunch the bleeding. A "generate-if-missing"
# fix, and/or, even better, a cleanup / regen sweep plus
# audit to ensure there're no more bugs causing them to be
# missed should be done.
if not treedoc:
return tree_cache
rhs_tree_ids = treedoc.tree_ids
if rhs_ci.parent_ids:
lhs_ci = CommitDoc.m.get(_id=rhs_ci.parent_ids[0])
else:
lhs_ci = None
if lhs_ci is not None:
lhs_tree_ids = TreesDoc.m.get(_id=lhs_ci._id).tree_ids
else:
lhs_tree_ids = []
new_tree_ids = [
tid for tid in chain(lhs_tree_ids, rhs_tree_ids)
if tid not in tree_cache ]
tree_index = dict(
(t._id, t) for t in TreeDoc.m.find(dict(_id={'$in': new_tree_ids}),validate=False))
tree_index.update(tree_cache)
rhs_tree_ids_set = set(rhs_tree_ids)
tree_cache.clear()
tree_cache.update(
(id, t) for id,t in tree_index.iteritems() if id in rhs_tree_ids_set)
rhs_tree = tree_index[rhs_ci.tree_id]
if lhs_ci is None:
lhs_tree = Object(_id=None, tree_ids=[], blob_ids=[], other_ids=[])
else:
lhs_tree = tree_index[lhs_ci.tree_id]
differences = []
commit_info = get_commit_info(rhs_ci)
for name, lhs_id, rhs_id in _diff_trees(lhs_tree, rhs_tree, tree_index):
differences.append(
dict(name=name, lhs_id=lhs_id, rhs_id=rhs_id))
# Set last commit data
rhs_tree = tree_index[rhs_ci.tree_id]
refresh_last_commit(repo_id, '/', rhs_tree, lhs_tree, None, commit_info)
# Build the diffinfo
di = DiffInfoDoc(dict(
_id=rhs_ci._id,
differences=differences))
di.m.save()
return tree_cache
def send_notifications(repo, commit_ids):
'''Create appropriate notification and feed objects for a refresh'''
from allura.model import Feed, Notification
commit_msgs = []
base_url = tg.config.get('base_url', 'sourceforge.net')
for oids in utils.chunked_iter(commit_ids, QSIZE):
chunk = list(oids)
index = dict(
(doc._id, doc)
for doc in Commit.query.find(dict(_id={'$in':chunk})))
for oid in chunk:
ci = index[oid]
href = repo.url_for_commit(oid)
summary = _summarize(ci.message)
Feed.post(
repo, title='New commit',
description='%s<br><a href="%s">View Changes</a>' % (
summary, href),
author_link=ci.author_url,
author_name=ci.authored.name)
branches = repo.symbolics_for_commit(ci)[0]
commit_msgs.append('%s: %s by %s %s%s' % (
",".join(b for b in branches),
summary, ci.authored.name, base_url, ci.url()))
if commit_msgs:
if len(commit_msgs) > 1:
subject = '%d new commits to %s %s' % (
len(commit_msgs), repo.app.project.name, repo.app.config.options.mount_label)
text='\n\n'.join(commit_msgs)
else:
subject = '%s committed to %s %s: %s' % (
ci.authored.name,
repo.app.project.name,
repo.app.config.options.mount_label,
summary)
branches = repo.symbolics_for_commit(ci)[0]
text = "%s: %s %s%s" % (",".join(b for b in branches),
ci.message,
base_url, ci.url())
Notification.post(
artifact=repo,
topic='metadata',
subject=subject,
text=text)
def _summarize(message):
if not message: return ''
summary = []
for line in message.splitlines():
line = line.rstrip()
if line: summary.append(line)
else: break
return ' '.join(summary)
def _diff_trees(lhs, rhs, index, *path):
def _fq(name):
return '/'.join(reversed(
(name,) + path))
# Diff the trees
rhs_tree_ids = dict(
(o.name, o.id)
for o in rhs.tree_ids)
for o in lhs.tree_ids:
rhs_id = rhs_tree_ids.pop(o.name, None)
if rhs_id == o.id:
continue # no change
elif rhs_id is None:
yield (_fq(o.name), o.id, None)
else:
for difference in _diff_trees(
index[o.id], index[rhs_id], index,
o.name, *path):
yield difference
for name, id in rhs_tree_ids.items():
yield (_fq(name), None, id)
# DIff the blobs
rhs_blob_ids = dict(
(o.name, o.id)
for o in rhs.blob_ids)
for o in lhs.blob_ids:
rhs_id = rhs_blob_ids.pop(o.name, None)
if rhs_id == o.id:
continue # no change
elif rhs_id is None:
yield (_fq(o.name), o.id, None)
else:
yield (_fq(o.name), o.id, rhs_id)
for name, id in rhs_blob_ids.items():
yield (_fq(name), None, id)
def get_commit_info(commit):
if not isinstance(commit, Commit):
commit = mapper(Commit).create(commit, dict(instrument=False))
sess = session(commit)
if sess: sess.expunge(commit)
return dict(
id=commit._id,
author=commit.authored.name,
author_email=commit.authored.email,
date=commit.authored.date,
author_url=commit.author_url,
shortlink=commit.shorthand_id(),
summary=commit.summary
)
def refresh_last_commit(repo_id, path, tree, lhs_tree, parent_tree, commit_info):
'''Build the LastCommit info.
We only need to create LastCommit info for objects that are in the
RHS but not in the LHS, because only those objects are only ones
who have had anything changed in them. (If file x/y/z.txt changes,
then it's hash will change, which also forces the hash for tree x/y
to change, as well as the hash for tree x. So as long as an object's
hash isn't in the LHS, it means it's new or modified in this commit.)
In order to uniquely identify the tree or blob that a LastCommitDoc is
for, the tree or blob hash is not sufficient; we also need to know
either it's full path name, or it's parent tree and name. Because of
this, we have to walk down the commit tree.'''
if lhs_tree is not None and tree._id == lhs_tree._id:
# tree was not changed in this commit (nor was anything under it)
return
# map LHS entries for easy lookup
lhs_map = {}
if lhs_tree:
for lhs_child in chain(lhs_tree.tree_ids, lhs_tree.blob_ids, lhs_tree.other_ids):
lhs_map[lhs_child.name] = lhs_child.id
# update our children
for child in chain(tree.tree_ids, tree.blob_ids, tree.other_ids):
if child.id != lhs_map.get(child.name, None): # check if changed in this commit
lc = set_last_commit(repo_id, path, child.name, child.id, commit_info)
# (re)curse at our child trees
for child_tree in tree.tree_ids:
child_name = child_tree.name
child_tree = TreeDoc.m.get(_id=child_tree.id)
lhs_child = None
if child_name in lhs_map:
lhs_child = TreeDoc.m.get(_id=lhs_map[child_name])
refresh_last_commit(repo_id, path + child_name + '/', child_tree, lhs_child, tree, commit_info)
def set_last_commit(repo_id, path, name, oid, commit_info):
lc = LastCommitDoc(dict(
_id='%s:%s:%s' % (repo_id, path, name),
object_id=oid,
name=name,
commit_info=commit_info))
lc.m.save(safe=False, upsert=True)
return lc
def last_known_commit_id(all_commit_ids, new_commit_ids):
"""
Return the newest "known" (cached in mongo) commit id.
Params:
all_commit_ids: Every commit id from the repo on disk, sorted oldest to
newest.
new_commit_ids: Commit ids that are not yet cached in mongo, sorted
oldest to newest.
"""
if not all_commit_ids: return None
if not new_commit_ids: return all_commit_ids[-1]
return all_commit_ids[all_commit_ids.index(new_commit_ids[0]) - 1]