Our Perforce server experienced some kind of database corruption a few years ago.
While the file data and revision history are mostly intact,
some metadata for several changesets got lost.
For example, inspecting certain changelists produces errors.
"""
$ p4 describe -s 12345
Date 2019/02/26 16:46:17:
Operation: user-describe
Operation 'user-describe' failed.
Change 12345 description missing!
"""
While some metadata (like changeset descriptions) is obviously lost,
most of it can be reconstructed via other commands:
* `p4 changes -l -t //...@12345,12345` --
to obtain date+time, author, beginning of changeset description;
* `p4 files -a //...@12345,12345` --
to obtain file revisions, file types, file actions;
* `p4 diff2 -u //...@12344 //...@12345` --
to get a unified diff of text files in a changeset;
* `p4 print -o binary.blob@12345 //depot/binary.blob@12345` --
to get a revision of a binary file.
It might be possible to teach git-p4 to fallback to other methods if `p4 describe` fails,
but it's probably too special-cased (really depends on kind and scale of DB corruption),
so some manual intervention is perhaps acceptable.
So, with some manual work, it's possible to reconstruct `p4 -G describe ...` output manually.
In our case, once git-p4 passes `p4 describe` stage,
it can proceed further just fine.
Thus, it's tempting to feed resurrected metadata to git-p4 when a normal `p4 describe` would fail.
This functionality may be useful to cache changelist information,
or to make some changes to changelist info before feeding it to git-p4.
A new config parameter is introduced to tell git-p4
to load certain changelist descriptions from files instead of from a server.
For simplicity, it's one marshalled file per changelist.
```
git config --add git-p4.changelistDescriptionFile 12345.marshal
git config --add git-p4.changelistDescriptionFile 12346.marshal
```
The following trivial script may be used to produce marshalled `p4 -G describe`-compatible output.
"""
#!/usr/bin/env python
import marshal
import time
# recovered commits of interest
changes = [
{
'change': '12345',
'status': 'submitted',
'code': 'stat',
'user': 'username1',
'time': str(int(time.mktime(time.strptime('2019/02/28 16:00:30', '%Y/%m/%d %H:%M:%S')))),
'client': 'username1_hostname1',
'desc': 'A bug is fixed.\nDetails are below:<lost>\n',
'depotFile0': '//depot/branch1/foo.sh',
'action0': 'edit',
'rev0': '28',
'type0': 'xtext',
'depotFile1': '//depot/branch1/bar.py',
'action1': 'edit',
'rev1': '43',
'type1': 'text',
'depotFile2': '//depot/branch1/baz.doc',
'action2': 'edit',
'rev2': '8',
'type2': 'binary',
'depotFile3': '//depot/branch1/qqq.c',
'action3': 'edit',
'rev3': '6',
'type3': 'ktext',
},
]
for change in changes:
marshal.dump(change, open('{0}.marshal'.format(change['change']), 'wb'))
"""
Or, the following script may be used to produce marshalled `p4 -G describe`-compatible output
for our particular database corruption.
"""
#!/usr/bin/env python
import itertools
import marshal
import subprocess
import tempfile
def p4_unmarshal(fileobj):
result = []
while True:
try:
result += [marshal.load(fileobj)]
except EOFError:
break
return result
def p4_describe_fallback(cl):
with tempfile.TemporaryFile() as p4_changes_output:
with tempfile.TemporaryFile() as p4_files_output:
subprocess.check_call(['p4', '-G', 'changes', '-l', '-t', '//...@{0},{0}'.format(cl)], stdout=p4_changes_output)
subprocess.check_call(['p4', '-G', 'files', '-a', '//...@{0},{0}'.format(cl)], stdout=p4_files_output)
p4_changes_output.seek(0)
p4_files_output.seek(0)
p4_changes_unmarshalled = p4_unmarshal(p4_changes_output)
p4_files_unmarshalled = p4_unmarshal(p4_files_output)
described_cl = p4_changes_unmarshalled[0] # there is usually only one entry
described_cl['desc'] += '<lost>\n'
assert described_cl['change'] == str(cl)
for (file_info, i) in itertools.izip(p4_files_unmarshalled, itertools.count()):
for f in ('depotFile', 'action', 'rev', 'type'):
described_cl['{}{}'.format(f, i)] = file_info[f]
assert file_info['change'] == described_cl['change']
return described_cl
cls_wanted = ( 12345, 12346 )
for cl in cls_wanted:
with open('{0}.marshal'.format(cl), 'wb') as f:
cl_info = p4_describe_fallback(cl)
marshal.dump(cl_info, f)
"""
Signed-off-by: Andrey Mazo <amazo@xxxxxxxxxxxxxx>
---
Documentation/git-p4.txt | 27 +++++++++++++++++++++++++++
git-p4.py | 24 +++++++++++++++++++++++-
2 files changed, 50 insertions(+), 1 deletion(-)
diff --git a/Documentation/git-p4.txt b/Documentation/git-p4.txt
index ceabab8b86..f751ae729f 100644
--- a/Documentation/git-p4.txt
+++ b/Documentation/git-p4.txt
@@ -654,10 +654,37 @@ git config --add git-p4.mapUser "p4user = First Last <mail@xxxxxxxxxxx>"
-------------
+
A mapping will override any user information from P4. Mappings for
multiple P4 user can be defined.
+git-p4.changelistDescriptionFile::
+ This config variable points 'git p4' to a file,
+ containing a serialized (marshalled) changelist description.
+ 'git p4' loads a description from such a file
+ instead of asking Perforce server about it.
+ The file format is the same as produced by 'p4 -G describe -s <changelist>'.
+ This option can be specified multiple times
+ to feed multiple changelist descriptions to 'git p4'.
+ The path is relative to git work tree,
+ file names or extensions don't matter.
+ This example loads 2 changelist descriptions:
++
+-------------
+git config --add git-p4.changelistDescriptionFile cl-12345.marshal
+git config --add git-p4.changelistDescriptionFile cl-12347.marshal
+-------------
++
+Under some circumstances (for example, Perforce database corruption)
+this option is useful to supply changelist description to 'git p4' bypassing 'p4'.
+Also, it can be used for caching of changelist descriptions
+to reduce load on the Perforce server in case of successive imports
+(say, when splitting the depot into multiple Git repositories)
+or for overriding some changelist information.
+This config variable is generally not needed after the initial import,
+so it can be removed from the config file
+together with corresponding description files after the import.
+
Submit variables
~~~~~~~~~~~~~~~~
git-p4.detectRenames::
Detect renames. See linkgit:git-diff[1]. This can be true,
false, or a score as expected by 'git diff -M'.
diff --git a/git-p4.py b/git-p4.py
index 98b2b7bbca..e65df92d75 100755
--- a/git-p4.py
+++ b/git-p4.py
@@ -2613,10 +2613,12 @@ def __init__(self):
self.initialParents = {}
self.tz = "%+03d%02d" % (- time.timezone / 3600, ((- time.timezone % 3600) / 60))
self.labels = {}
+ self.loadedChangelistDescriptions = {}
+
# Force a checkpoint in fast-import and wait for it to finish
def checkpoint(self):
self.gitStream.write("checkpoint\n\n")
self.gitStream.write("progress checkpoint\n\n")
out = self.gitOutput.readline()
@@ -3319,10 +3321,25 @@ def getBranchMappingFromGitBranches(self):
branch = "main"
else:
branch = branch[len(self.projectName):]
self.knownBranches[branch] = branch
+ def loadChangelistDescFromFile(self):
+ changelistDescriptionFiles = gitConfigList("git-p4.changelistDescriptionFile")
+ for clMarshalledDescFile in changelistDescriptionFiles:
+ if not clMarshalledDescFile:
+ continue
+
+ try:
+ with open(clMarshalledDescFile, 'rb') as clFileObj:
+ clDesc = marshal.load(clFileObj)
+ if not ("status" in clDesc and "user" in clDesc and "time" in clDesc and "change" in clDesc):
+ die("Changelist description read from {0} doesn't have required fields".format(clMarshalledDescFile))
+ self.loadedChangelistDescriptions[int(clDesc["change"])] = clDesc
+ except (IOError, TypeError, ValueError, EOFError) as e:
+ die("Can't read changelist description from {0}: {1}".format(clMarshalledDescFile, str(e)))
+
def updateOptionDict(self, d):
option_keys = {}
if self.keepRepoPath:
option_keys['keepRepoPath'] = 1
@@ -3420,11 +3437,14 @@ def searchParent(self, parent, branch, target):
return None
def importChanges(self, changes, origin_revision=0):
cnt = 1
for change in changes:
- description = p4_describe(change)
+ if change in self.loadedChangelistDescriptions:
+ description = self.loadedChangelistDescriptions[change]
+ else:
+ description = p4_describe(change)
self.updateOptionDict(description)
if not self.silent:
sys.stdout.write("\rImporting revision %s (%s%%)" % (change, cnt * 100 / len(changes)))
sys.stdout.flush()
@@ -3706,10 +3726,12 @@ def run(self, args):
bad_changesfile = True
break
if bad_changesfile:
die("Option --changesfile is incompatible with revision specifiers")
+ self.loadChangelistDescFromFile()
+
newPaths = []
for p in self.depotPaths:
if p.find("@") != -1:
atIdx = p.index("@")
self.changeRange = p[atIdx:]
--
2.19.2
