Difference between revisions of "FPC git"
Line 465: | Line 465: | ||
<pre> | <pre> | ||
git checkout fixes_3_2 | git checkout fixes_3_2 | ||
+ | git pull | ||
git cherry-pick -x 12345678 # 12345678 is the hash of the commit to be cherry picked to fixes | git cherry-pick -x 12345678 # 12345678 is the hash of the commit to be cherry picked to fixes | ||
git push | git push |
Revision as of 11:05, 17 August 2021
│
English (en) │
русский (ru) │
Git setup
Git repository location
The Git repositories are hosted on Gitlab:
FPC compiler, RTL, packages and utils are located here:
The URL to clone the repository is:
git clone https://gitlab.com/freepascal.org/fpc/source.git
The Documentation Git repository is here:
The URL to clone the documentation repository is:
git clone https://gitlab.com/freepascal.org/fpc/documentation.git
The website Git repository is here:
The URL to clone the website repository is:
git clone https://gitlab.com/freepascal.org/fpc/website.git
Your identity in git
In order for your commits to be recognized, please configure your local (or better yet, global) git installation to use the address linked to your gitlab acccount.
git config --global user.email your.email.user@youremail.host
This will set your email address for all git repositories you use, and will allow the gitlab interface to correctly link commits to users in its HTML interface.
If you have multiple gitlab (or even github) accounts, you may prefer to set this email only locally:
git config --local user.email your.email.user@youremail.host
this command must be execute in a particular repository, and sets the email address for only that repository.
HTTP access and credentials
If you access the repository using HTTPS, by default, git will ask for the username/password. you can tell git to store them permanently, issue the following command:
git config --global credential.helper store
There are other possibilities for storing this info. More info can be found on
Using access tokens
You may want to use an access token and HTTPS access, it has some advantages. This is supported by most modern git clients:
and here you can see how to do this on gitlab:
The advantage of using an access token is that it can be given limited rights on gitlab, for example only accessing a repository. This also means you can easily enable 2FA for your gitlab account.
Git mirror on github
The Gitlab repositories are mirrored on Github. Cloning of the sources can be done with the following command
git clone https://github.com/fpc/FPCSource.git fpc
and for documentation:
git clone https://github.com/fpc/FPCDocumentation.git fpcdocs
In case you want to submit patches (pull requests) it is strongly recommended to use gitlab instead.
Branching model
In a first stage, the branching model as used in Subversion will be used: fixes done in trunk, and later cherry picked to the fixes branch.
At a later stage, the branching model may be changed into one of the many git branching schemes.
Windows Git clients
On macOS, Linux & BSD, the git command is installed with the system package manager (you must install XCode command line tools on macOS). On Windows, a separate git client must be installed. There are many available, but the following are popular ones:
- Git for windows: a command-line client with a bash shell, so you can copy & paste commands you find on internet:
- TortoiseGit integrates with the Windows explorer, much like TortoiseSVN does:
- https://tortoisegit.org/
- if you already have TortoisSVN installed, both can work side-by-side.
- Sourcetree is a free client for Windows and macOS:
- https://www.sourcetreeapp.com/
- It is made by the people from bitbucket.
- Smartgit runs on Windows, Linux and macOS:
- The VSCode and Atom editors have git support built in.
Concepts and SVN to GIT differences
A few "good to know" differences: FPC_git_concepts
Common SVN operations in Git
Check out
To check out a new copy of a repository is called 'cloning' in git.
git clone https://gitlab.com/freepascal.org/testconversion
As in subversion, you can give it another name:
git clone https://gitlab.com/freepascal.org/testconversion myconversion
This operation can take a while, the FPC source base is large.
Update
To update your source code with the changes in the remote repository, is called 'pulling' in git:
git pull
Note that in difference with subversion, git always updates the whole repository.
To keep a commit history which is as linear as possible and avoids merge commits, it is recommended to do a rebase when you pull.
git pull --rebase
You can make the use of rebase for every pull the default by executing:
git config pull.rebase true
Resolve
If you have conflicts in svn after updating or merging, you fix the conflicts and then tell svn everything is fine with the resolved command.
In git, conflicts generally occur when merging or when applying shelved/stashed changes.
When conflicts occur, then
- Changes to files that are not in conflict, will be staged
- Changes to files that do conflict will be applied without staging to the working copy (so that a plain diff will only show conflicting changes)
To resolve conflicts,
- If you want to commit the resolved version of the file, mark it as resolved by staging/adding it as well
- If you want to just get rid of the "both modified" status (= conflicted) without adding the file, use git reset file
Once all files have been staged or marked as not conflicted, you can commit (only staged files will be committed).
add (adding new files)
Subversion uses the add command to add a file to the versioning system. This is no different in git: If you want to add a file, you must add it first, like this:
git add newfile.pas
You must then (just as in subversion) still commit this newly added file.
rm (removing files)
Subversion uses the rm command to remove a file from the versioning system. This is no different in git:
git rm nolongerneededfile.pas
You must then (just as in subversion) still commit this removed file.
mv (renaming files)
Subversion uses the rm command to rename a file in the versioning system. This is no different in git:
git mv oldfile.pas newfile.pas
You must then (just as in subversion) still commit this change.
As in subversion, you can move one or more files to another directory:
git mv file1.pas file2.pas somesubdir
commit
If you have made some changes locally and wish to commit them, this is a 2-step process:
git add myfile.pas
This schedules the file to be committed (called staging in git, so the changes are now staged). You can add as many files as you want like this.
When you are ready to commit what you have staged, you can commit:
git commit -m '* Some nice comment telling what you did'
In fact, if you're in a hurry and know you want to commit a file, you can combine the 2 commands;
git commit -m '* Some nice comment telling what you did' myfile.pas
If you are really in a hurry, you can commit all changed files in one fell swoop:
git commit -m '* Some nice comment telling what you did' -a
But it is not really recommended to use this, as it will commit all changes in your local copy, not just the ones in the current working directory.
At this point, your changes have been saved locally, but have not yet been sent to the remote repository. To do that, you must additionally send the changes to the remote repository. This is called pushing:
git push
That will send all locally committed changes to the server. Obviously only changes that have not yet been sent previously are sent.
diff (show changes to working copy)
In svn, you can see the changes that have been made but have not yet been committed with diff. This is also correct in git:
git diff
This will show all changes in your working copy that are not yet staged (scheduled for committing).
You can view this for a single file as well:
git diff myfile.pas
This will show unstaged changes made to your working copy of myfile.pas.
If you have already staged one or more files (i.e. marked for inclusion in the next commit), the above will not show them in the generated diff. If you wish to see the staged changes (and only the staged changed), use the --cached command-line option:
git diff --cached
diff (show changes performed in a commit)
In svn, you can see the changes that a commit did through svn diff -c <revision>. In git, you use the show command instead:
git show -p <commit_hash_or_branch_or_tag>
If you omit the commit hash, the last commit to the current branch will be shown instead.
log (show commits in a branch)
In svn, you can see the commits in a branch with svn log. The same works in git:
git log
This will list the commit hashes (the string that git uses to identify commits, similar to revision numbers in svn) followed by the log messages. If you also want to see which files were changed, use
git log --stat
ls (listing branches)
To get a list of branches in Subversion, you use the 'ls' command with a server URL. By convention, all directories under '/branches' are the names of branches. Branches always exist both on the client(s) and server.
In git, branches are formalised as part of the version control system: commands deal specifically with branches rather than with paths that are treated as branches. Additionally, git differentiates between remote branches (branches that exist in the repository that you cloned, i.e., in this case on the gitlab server), and local branches (branches that you created on your machine after cloning). The repository that you cloned is referenced using the alias "origin" by default, and the names of remote branches are prepended by this alias.
The list of local branches can be obtained using:
git branch
By default, the only local branch you will have is main, which corresponds to svn trunk. This local branch is automatically created when performing the initial clone operation.
The list of remote branches can be obtained using:
git branch -r
By default, the remote branches are the branches that exist in the repository on the gitlab server, whose alias is "origin". The local main branch automatically tracks (~ is linked to) the remote origin/main branch. Because of this tracking, executing git pull resp. git push commands while you have your local main branch checked out will synchronise the two from resp. to gitlab. See switch on how to create more local branches and how to link them to remote branches.
copy (creating a branch)
To create a branch from the current working directory situation, you can create a branch like this:
git branch mybugfix
this will create a branch mybugfix from the currently checked out branch at the current commit. But it does not yet make this the active branch.
You need to check out the newly made branch first:
git checkout mybugfix
The two operations can also be combined with one command:
git checkout -b mybugfix
switch (checking out a branch)
To switch to an existing local branch mybugfix, you can use the checkout command:
git checkout mybugfix
this will check out branch mybugfix. If there are any uncommitted changes which would cause a conflict, git will refuse the checkout.
To switch to a remote branch `mybugfix2`, which does not exist yet locally, you can also use the checkout command:
git checkout mybugfix2
this will check out branch mybugfix2 and automatically set it up to track the remote branch. Here again, if there are any uncommitted changes which would cause a conflict, git will refuse the checkout.
Note that if the branch does not exist remotely, this will simply create a new branch locally. It is better to be explicit:
git checkout -b mybugfix2 --track origin/mybugfix2
If you created a branch locally which does not yet exist on the server, made some modifications, and you wish to push this branch to the server, you must tell this to git when pushing:
git push -u origin/mybugfix mybugfix
the -u origin/mybugfix tells git all it needs to know to connect the local with the remote branch.
revert
Revert the entire working tree
If you have made some changes locally and wish to undo them, you can do a hard reset in git:
git reset --hard
This will undo any changes you made in the entire checkout, including any staged changes.
Revert changes to specific files/directories
You cannot undo changes to individual files with git reset --hard. Instead, use
git checkout <pathspec>
pathspec can be any combination of files and directories. Note that if the files had staged changes, this command will restore the staged version rather than the last committed version.
merge (merging the changes in 2 branches)
To merge changes in 2 branches, you use the merge command. If we want to merge the changes in mybugfix to fixes then we do:
git checkout fixes git merge mybugfix
blame (check who made modifications)
To see who changes what line (and in what commit) svn offers you the 'blame' command. The same command exists in git.
git blame yourfile.pas
status (check status of files)
To see whether there are any changed files in your working repository, svn offers the 'status' command. The same command exists in git.
git status
will present you with all changed, staged and new files in the repository.
If you want only the status of files below a certain directory, you can specify the name of the directory. So to get the changes in the current working directory (or below) that would be:
git status .
will present you with all changed, staged and new files in the current directory. This is the default behaviour of svn.
shelve (temporarily undo changes)
Subversion has an (experimental) feature that allows to temporarily set aside changes to your working copy : shelve. This feature exists since a long time in git and is called stash:
git stash
will set aside any changes to the working copy, and restore the working copy to the state it would be in if you did a 'git pull' (or svn update).
To reapply the changes you set aside to the working copy, and to remove them from the list of stashes at the same time, you can execute the following command:
git stash pop
If any conflicts occur, you can
- resolve them, or
- use git reset --hard to revert all files again
In either case, the stash will not be touched when conflicts occurred while applying it. If you still with to remove it, you can do so with
git stash drop
info (get working dir and server info)
In subversion, you can get information about the current working directory and remote server configuration using svn info. Since git is a distributed system, no direct equivalent of the info command exists: there is no single server, and the revision number as known in subversion does not exist.
The following script attempts to display similar information as the svn info command:
#!/bin/bash # author: Duane Johnson # email: duane.johnson@gmail.com # date: 2008 Jun 12 # license: MIT # # Based on discussion at http://kerneltrap.org/mailarchive/git/2007/11/12/406496 pushd . >/dev/null # Find base of git directory while [ ! -d .git ] && [ ! `pwd` = "/" ]; do cd ..; done # Show various information about this git directory if [ -d .git ]; then echo "== Remote URL: `git remote -v`" echo "== Remote Branches: " git branch -r echo echo "== Local Branches:" git branch echo echo "== Configuration (.git/config)" cat .git/config echo echo "== Most Recent Commit" git --no-pager log -n1 echo echo "Type 'git log' for more commits, or 'git show' for full commit details." else echo "Not a git repository." fi popd >/dev/null
Note that this script will also work in the Windows environment if you have the windows git client installed, because it comes with a minimal unix environment.
file properties (EOL handling etc.)
Subversion allows you to set some properties on files. There are 'reserved' properties that svn itself uses, for example to handle EOL handling and file type. Because git operates on diffs and not on files, no similar concept exists.
However, git does allow you to set global or local attributes for classes of files. These are in the .gitattributes file. That file should be located in the root of your repository, and can contain some attributes. For example:
* text=auto *.txt text *.vcproj text eol=crlf *.sh text eol=lf *.jpg -text
This is a regular file which you can (and must) add with git add so it is saved for all users.
cherry picking
To copy one commit from branch to another, cherry picking can be used, checkout the branch where the commit shall be added and execute:
git cherry-pick <hash of commit>
If this operation causes a conflict, fix it and continue the operation with:
git cherry-pick --continue
To track which commits are cherry picked, use the option -x. This appends a small note to the commit message containing the hash of the cherry picked commit.
In case you want to abort the cherry picking e.g. because of too many conflicts, just run
git cherry-pick --abort
Managing the FPC fixes branch
So far, FPC uses a developement model which is sometimes called the "ghetto development model". Patches battle out in the main branch if they are good and if they are good, they are ported to the fixes branch from which releases are created. This development model is not naturally supported by git but can be implemented based on cherry-picking. To make tracking of cherry picks easier, git cherry-pick must be called with -x. Furthermore, only one commit at once may be cherry-pick.
Example:
git checkout fixes_3_2 git pull git cherry-pick -x 12345678 # 12345678 is the hash of the commit to be cherry picked to fixes git push
For conflict resolution, see section about cherry picking.
To make things a little bit more easy, to get more information in case of conflits and avoid to forget -x, use a script gcp like:
#!/usr/bin/env bash git show --name-only $1 git cherry-pick -x $1 || exit 255
To figure out which hashes are eliglibe (i.e. not yet cherry picked) for cherry picking, use a script like https://gitlab.com/freepascal.org/fpc/merging/-/blob/main/findeligible This script is comparable complicated as it takes also merges from svn times into account. A regularily updated list of eliglible commits can be found at: https://gitlab.com/freepascal.org/fpc/merging/-/blob/main/eligible.log?expanded=true&viewer=simple
Mass cherry picking
If you have a list of commits identified by their hashes in a file, they can be cherry picked using the script gcp from above like:
xargs -n 1 gcp < filewithhashes
Using the script from above is important because xargs only stops if the called program returns 255. So in case of a cherry pick conflict, the command aborts, one can fix the conflict, remove the already cherry picked commits from filewithwith and continue.
A file with svn revisions can be converted easily to a list of hashes by something like:
cat filewithrevs | xargs -n 1 git log --grep='trunk@{}' --format='%H' > filewithhashes