This question already has answers here:
Undo an accidental hg strip?
(3 answers)
Closed 3 years ago.
I have this Mercurial repository where I keep a counter:
$ echo 1 > count
$ hg add count
$ hg com -m 'incrementing to 1'
$ echo 2 > count
$ hg com -m 'incrementing to 2'
So far so good, but then I committed a mistake:
$ hg com -m 'incrementing to 3'
So I use hg strip to revert this last commit:
$ hg strip --keep -r -2
saved backup bundle to /home/adam/sandbox/count/.hg/strip-backup/bda856a578bf-ff2b025f-backup.hg
Unbeknownst to me, I have committed another, bigger mistake! I stripped the two topmost commits, and I just wanted to strip the topmost one:
$ hg log
changeset: 0:7b5533cf962a
tag: tip
user: Adam Victor Nazareth Brandizzi <brandizzi#gmail.com>
date: Wed May 15 08:00:27 2019 -0300
summary: incrementing to 1
How I get my commits back?
NOTE: this is a contrived example of a case were I needed to use hg strip. No need to wasting time pointing out there are alternatives etc.
As I've learned from this message in Mercurial's mailing list, the solution is quite straightforward. Note that when I ran hg strip --keep -r -2 I got this line in the output:
saved backup bundle to /home/adam/sandbox/count/.hg/strip-backup/bda856a578bf-ff2b025f-backup.hg
So, basically, Mercurial made a backup of the original history! Now I just need to pull from it:
$ hg pull /home/adam/sandbox/count/.hg/strip-backup/bda856a578bf-ff2b025f-backup.hg
pulling from /home/adam/sandbox/count/.hg/strip-backup/bda856a578bf-ff2b025f-backup.hg
searching for changes
adding changesets
adding manifests
adding file changes
added 2 changesets with 2 changes to 1 files
(run 'hg update' to get a working copy)
$ hg update
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
My file is back to the original state...
$ cat count
66
...as well as my history:
$ hg log
changeset: 2:fc4d1cc18a4b
tag: tip
user: Adam Victor Nazareth Brandizzi <brandizzi#gmail.com>
date: Wed May 15 08:00:51 2019 -0300
summary: incrementing to 3
changeset: 1:bda856a578bf
user: Adam Victor Nazareth Brandizzi <brandizzi#gmail.com>
date: Wed May 15 08:00:38 2019 -0300
summary: incrementing to 2
changeset: 0:7b5533cf962a
user: Adam Victor Nazareth Brandizzi <brandizzi#gmail.com>
date: Wed May 15 08:00:27 2019 -0300
summary: incrementing to 1
Now I can call the right strip command (hg strip --keep -r -1).
What if we do not have the backup file name anymore?
You may have lost the backup file name (it did happen to me!) but you can find it in $YOUR_REPO/.hg/strip-backup. If you have many backups, just list them sorted by time, descending:
$ ls -tl .hg/strip-backup/
total 4
-rw-rw-r-- 1 adam adam 754 May 15 08:10 bda856a578bf-ff2b025f-backup.hg
-rw-rw-r-- 1 adam adam 754 May 15 08:02 02ac79859dd0-0902f53b-backup.hg
The last one is probably the one you want.
Related
Example scenario:
revision #100 - user1 makes a commit on DEV branch in 2017-01-01
revision #101 - user2 cherry picked user1's commit to master branch (hg graft -r 100) in 2017-01-02
hg log:
changeset:101
user: user1
date: 2017-01-01
summary: message 100
changeset:100
user: user1
date: 2017-01-01
summary: message 100
But how would I know user2 pushed revision #101?
You cannot (besides what Lasse suggest in his comment).
Consider:
touch a; hg add a; hg commit -m 'Add a as myself'
touch b; hg add b; hg commit -m 'Add b as you' --user you#foo.com
hg update 0
hg branch stable
hg graft --user 3rd#bar.com --force 1
hg log --graph
# changeset: 2:3078a4888a8a
| branch: stable
| tag: tip
| parent: 0:f7fa0d60eaf2
| user: 3rd#bar.com
| date: Fri Apr 28 20:07:35 2017 +0200
| summary: Add b as you
|
| o changeset: 1:e12471b861d6
|/ user: you#foo.com
| date: Fri Apr 28 20:07:35 2017 +0200
| summary: Add b as you
|
o changeset: 0:f7fa0d60eaf2
user: This Is Me me#me.com
date: Fri Apr 28 20:06:42 2017 +0200
summary: Add a as myself
If you think about it, it is less shocking than what is seems (I am not saying it is not a problem, mind you).
Mercurial has an extension to perform PGP signatures which might help, but not after the fact and requires a certain setup.
To have detailed explanation, have a look at CommitSigningPlan, although I don't know the status.
To my knowledge, monotone was a distributed VCS system designed to solve this kind of problems, I am not sure whether it is still developed.
I would finish with a philosophical observation: if a project has to protect itself from this kind of problems (an authorised developer impersonating another), then maybe that project has other problems... Again, this doesn't mean that authenticating a commit is not important.
Folks, I need your help!
I tried to push my hg repository to bitbucket for the first time. It didn't work and I thought this was because the project is to huge. Therefore, I wanted to push only one single file.
Since I did "hg add *" before, I wanted to revoke this, and typed "hg remove *" and now it DELETED FILES ON MY COMPUTER. But not all of them, only some...
I need these files back, how do I do it? Please tell me they're not gone, please please please.
Thank you so much for your help!!
This is my command history:
518 hg commit commit_test.txt -m "first commit"
519 hg add commit_test.txt
520 hg commit -m "test"
521 ls -ll
522 ls -la
523 cd .hg/
524 ls
525 ls -la
526 pico hgrc
527 cd ..
528 hg status
529 hg add commit_test.txt~
530 hg add commit_test.txt
531 hg commit -m "blabla"
532 hg push
533 hg add commit_test.txt
534 hg commit -m "maaan"
535 hg status
536 hg remove *
537 hg status
538 hg commit -m "bla"
539 ls
540 hg add test.txt
541 hg commit -m "test"
542 hg push
543 hg revert
544 hg revert --all
545 history
546 hg add *
547 hg add -r *
548 hg help add
549 hg add -S *
550 cd gui_relabel/
551 hg add *
552 cd images/
553 hg add *
554 cd ..
555 hg revert all
556 hg revert *
557 history
Here is what hg log gives me:
$ hg log
changeset: 6:4726f671ae96
tag: tip
user: KG <...#gmail.com>
date: Wed Dec 04 12:21:30 2013 +0100
summary: test
changeset: 5:55b3158def38
user: KG <...#gmail.com>
date: Wed Dec 04 12:17:19 2013 +0100
summary: bla
changeset: 4:ae0dd836586d
user: KG <...#gmail.com>
date: Wed Dec 04 12:14:50 2013 +0100
summary: blabla
changeset: 3:0249fdc26fa7
user: KG <...#gmail.com>
date: Wed Dec 04 12:13:59 2013 +0100
summary: test
changeset: 2:40bdcf4d2104
user: KG <...#gmail.com>
date: Wed Dec 04 12:12:37 2013 +0100
summary: first commit test
changeset: 1:f9e20020ca1d
user: KG <...#gmail.com>
date: Sun Nov 10 14:54:46 2013 +0100
summary: first commit
changeset: 0:7a8edcee06ff
user: KG <...#gmail.com>
date: Mon Nov 04 20:36:41 2013 +0100
summary: blabla
Your files are not gone. Mercurial does not throw away data, that is one of the first rules of the system.
To explain what hg remove did to your files, let us look at the cases one by one. A file in your working copy can be in one out of a handful different states: it can be modified, added, removed, missing, unknown, ignored, or clean.
I've prepared a working copy that looks like this:
$ hg status --all
M a
A b
R c
! d
? e
I f
C g
The files a to g are in the seven states I mentioned above. We can now talk about what happens when you try to remove these files:
A modified file (a) is not removed:
$ hg remove a
not removing a: file is modified (use -f to force removal)
An added file (b) is also not removed:
$ hg remove b
not removing b: file has been marked for add (use forget to undo)
A removed file (c) is no longer present in the working copy, so nothing more happens.
A missing file (d) is no longer present in the working copy, so nothing more happens.
An untracked file (e) is not removed:
$ hg remove e
not removing e: file is untracked
An ignored file (f) is not removed (since it is not tracked):
$ hg remove f
not removing f: file is untracked
A clean file (g) is removed.
The only type of file that hg remove will actually remove from your working copy is a clean file. A clean file is already committed, so the content is safely stored in the repository. You can get it back with hg revert:
$ hg revert g
$ hg status --all
M a
A b
R c
! d
? e
I f
C g
The file g is back with the same content as it had before you removed it.
It looks like you could run
hg backout 5
You can read more about backout if you like
When using mercurial, when working on a large change, I will do several commits (as safe points, or share my progress with coworkers).
Whe I am done, I want to push my changes into the main repository. Before I do that, though, I will want to do a "hg pull -u && hg merge"to get changes that occured since my last pull, and review my changes.
Because I have several changesets, running hg diff on each of those is not a good solution.
I tried running hg diff -c"outgoing() and not merge()" to only show the diff for my outgoing changesets (without merge ones, as this tends to only add noise which I want to avoid).
However, it is still not ok for me, as mercurial seems to skip some information, and I do not understand on what basis. Please refer to the scenario below:
full scenario:
# clean working directory
rm -fr base clone
#prepare reference repo
mkdir base; cd base
hg init
touch 1.txt
echo " 1\n2" >>1.txt
hg add 1.txt
touch 2.txt
hg add 2.txt
hg commit -m"initial commit"
cd ..
hg clone base clone
cd base
#update ref repo.
echo "foo" >> 1.txt
hg commit -m"update file on remote repository"
#make changes locally
cd ../clone
echo "foo" >> 2.txt
hg commit -m "update initial file"
#now we are ready to push
hg pull -u && hg merge
hg commit -m"merge"
#actually...
sed -i -e 's/foo/bar/' 2.txt
hg commit -m"changed my mind"
hg out
changeset: 1:fce14ac089b2
date: Thu Nov 14 11:17:08 2013 +0100
summary: update initial file
changeset: 3:1967c50242c0
parent: 1:fce14ac089b2
parent: 2:2cdaf75afc6d
date: Thu Nov 14 11:17:10 2013 +0100
summary: merge
changeset: 4:120a0b8a0ede
tag: tip
date: Thu Nov 14 11:18:54 2013 +0100
summary: changed my mind
#try to display my changes
hg diff -c"outgoing() and not merge()"
--- a/2.txt Thu Nov 14 11:17:10 2013 +0100
+++ b/2.txt Thu Nov 14 11:18:54 2013 +0100
## -1,1 +1,1 ##
-foo
+bar
#only one change here!
#but
hg log -r"outgoing() and not merge()"
changeset: 1:fce14ac089b2
date: Thu Nov 14 11:17:08 2013 +0100
summary: update initial file
changeset: 4:120a0b8a0ede
tag: tip
date: Thu Nov 14 11:18:54 2013 +0100
summary: changed my mind
#is ok
So, could anyone please explain to me why mercurial skips changeset fce14ac089b2 entirely? What I expect is a consolidated output, i.e something like:
--- a/2.txt
+++ b/2.txt
## -1,1 +1,1 ##
+bar
Thanks a lot in advance.
I think the problem is a misunderstanding of what hg diff -c aka hg diff --change does. It takes a single revision id and shows you the change made by that one revision relative to its left parent.
You're calling:
hg diff -c"outgoing() and not merge()"
which yields multiple revisions (it's a revision set), but only one of them is being used. You're essentially seeing the output of:
hg diff --change 120a0b8a0ede
If what you're trying to do is see all of your outgoing changes lumped together in one diff you don't want --change/-c.
If you think about it the revset outgoing() could contain many changeset from many branches so smooshing them all into one diff isn't necessarily possible at all.
If the view you're trying to get is "how does the most recent head on branch default locally compare to the most recent head on default remotely" you need to know the node id of the most recent head on default, and then locally you'd just run:
hg diff node_id_of_most_recent_head_on_default_at_remote_site
This isn't guaranteed, as I expect that there's some way to have a change graph where the last public ancestor isn't the change-set you're after, but...
> hg diff -r "last(ancestors(.) and public())"
diff -r 5579be654db4 2.txt
--- a/2.txt Sat Nov 16 11:03:06 2013 +0000
+++ b/2.txt Sat Nov 16 11:29:34 2013 +0000
## -0,0 +1,1 ##
+bar
I want to reconstruct the entire history of a given file, across renames. I know that the follow flag to hg log will show me revisions in which the file, under it's previous name, was modified. But how can I see the previous name? A -v to hg log will show file names in each revision, but if there are enough names I won't be able to deduce which one it was.
Perform in cycle log by below example
hg log -r "adds(Seeker.txt)" --template "{file_copies}"
Seeker.txt (Искатель.txt)
Output (as Lasse wrote) is "NewName (Oldname)"
You can use the following two extra flags:
-v -C
I tested this in a repository, adding test1.txt and test2.txt, and then in the same commit I renamed them to test3.txt and test4.txt respectively, and this is what the log looked like if I asked for the log of test3.txt:
[D:\Temp\hg] :hg log test3.txt -v -C -f
changeset: 1:54dac6d79938
tag: tip
user: Lasse V. Karlsen <lasse#vkarlsen.no>
date: Mon Oct 31 08:10:36 2011 +0100
files: test1.txt test2.txt test3.txt test4.txt
copies: test3.txt (test1.txt) test4.txt (test2.txt)
description:
renamed
changeset: 0:89213dc6f36f
user: Lasse V. Karlsen <lasse#vkarlsen.no>
date: Mon Oct 31 08:10:25 2011 +0100
files: test1.txt test2.txt
description:
initial
Granted, it isn't super-easy to spot the files, but you can see it.
Let's say I have this:
hg init
touch a
hg add a
hg commit -m a
touch b
hg add b
hg commit -m b
hg up -r 0
touch c
hg add c
hg commit -m c
Now I will have multiple heads, because of the last commit. If, for example I want to keep the last head, the one created by commit c ( effectively discarding b, and all other commits made after the first ) , how could I do it? I played a little with mq's strip command, and I'm able to achieve what I want, but I'd like to know if there's another way.
Yes, there is another way. From your example above, the output of hg glog looks a bit like this:
# changeset: 2:925573c7103c
| tag: tip
| parent: 0:4fe26dfe856d
| user: Joel B Fant
| date: Thu Jul 28 23:20:45 2011 -0400
| summary: c
|
| o changeset: 1:9dc928176506
|/ user: Joel B Fant
| date: Thu Jul 28 23:20:24 2011 -0400
| summary: b
|
o changeset: 0:4fe26dfe856d
user: Joel B Fant
date: Thu Jul 28 23:20:12 2011 -0400
summary: a
If you clone test but specify a revision, it will only clone that revision and its ancestors. So if your repo's directory is TwoHeadsMightNotBeBetter, you can go to its parent directory and issue:
hg clone TwoHeadsMightNotBeBetter OneHeadIsFine -r 925573c7103c
I specified changeset id in that, but you could also use -r 2 instead, or even -r tip since it is currently the tip of that repo. Now when you go into the new clone and do hg glog, you have only one head:
# changeset: 1:925573c7103c
| tag: tip
| user: Joel B Fant
| date: Thu Jul 28 23:20:45 2011 -0400
| summary: c
|
o changeset: 0:4fe26dfe856d
user: Joel B Fant
date: Thu Jul 28 23:20:12 2011 -0400
summary: a
If you had 3 heads and wanted to keep 2, it would be a little different. You'd have to clone one of them with -r and the also specify -r on hg pull to pull specific branches.
You can also perform a dummy merge of the changeset containing b (changeset 1 in this case):
$ hg --config ui.merge=internal:fail merge 1
resolving manifests
getting b
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
(branch merge, don't forget to commit)
$ hg revert --all --no-backup --rev .
removing b
$ hg status
R b
$ hg commit -m "Dummy merge -- eliminate b"
committed changeset 3:c163151f19df
$ ls
a c
Note that in this case you haven't rewritten any history or created any new repos -- particularly important if b had already been pushed to a centralized repo and pulled by others.