Fossil: Git to Fossil Translation Guide

Introduction

Fossil shares many similarities with Git. In many cases, the sub-commands are identical: fossil bisect does essentially the same thing as git bisect, for example.

This document covers the cases where there is no simple 1:1 mapping, usually because of intentional design differences in Fossil that prevent it from working exactly like Git. We choose to explain these differences rather than provide a simple “translation dictionary,” since to understand the conversion, you need to know why the difference exists.

We focus on practical command examples here, leaving discussions of the philosophical underpinnings that drive these command differences to another document. The case studies do get a bit philosophical, but it is with the aim of illustrating how these Fossil design differences cause Fossil to behave materially differently from Git in everyday operation.

We present this from the perspective of Git users moving to Fossil, but it is also possible to read this document as a Fossil user who speaks only pidgin Git, who may often have questions of the form, “Now how do I do X in Git again?”

This document’s authors are intimately familiar with Fossil, so it is difficult for us to anticipate the perspective of people who are intimately familiar with Git. If you have a lot of prior Git experience, we welcome your contributions and questions on the Fossil Forum.

While we do try to explain Fossil-specific terminology inline here as-needed, you may find it helpful to skim the Fossil glossary. It will give you another take on our definitions here, and it may help you to understand some of the other Fossil docs better.

Repositories and Checkouts Are Distinct

A repository and a check-out are distinct in Fossil, allowing them to be stored in separate directory trees, whereas the two are commingled by default with Git, with the repository stored in a .git subdirectory underneath your working directory. This difference shows up in several separate places when it comes to moving from Git to Fossil.

Checkout Workflows

A Fossil repository is a SQLite database storing the entire history of a project. It is not normally stored inside the working tree. A Fossil working tree — also called a check-out — is a directory that contains a snapshot of your project that you are currently working on, extracted for you from the repository database file by the fossil program.

There are ways to emulate the Fossil working style in Git, but because they’re not designed into the core concept of the tool, Git tutorials usually advocate a switch-in-place working mode instead, so that is how most users end up working with Git. Contrast Fossil’s check-out workflow document to see the practical differences.

There is one Git-specific detail we wish to add beyond what that document already covers. This command:

    git checkout some-branch

…is best given as:

    fossil update some-branch

…in Fossil. There is a fossil checkout command, but it has several differences that make it less broadly useful than fossil update in everyday operation, so we recommend that Git users moving to Fossil develop a habit of typing fossil up rather than fossil checkout. That said, one of those differences does match up with Git users’ expectations: fossil checkout doesn’t pull changes from the remote repository into the local clone as fossil update does. We think this is less broadly useful, but that’s the subject of the next section.

Update vs Pull

The closest equivalent to git pull is not fossil pull, but in fact fossil up.

This is because Fossil tends to follow the CVS command design: cvs up pulls changes from the central CVS repository and merges them into the local working directory, so that’s what fossil up does, too. (This design choice also tends to make Fossil feel comfortable to Subversion expatriates.)

The fossil pull command is simply the reverse of fossil push, so that fossil sync is functionally equivalent to:

    fossil push ; fossil pull

There is no implicit “and update the local working directory” step in Fossil’s push, pull, or sync commands, as there is with git pull.

Someone coming from the Git perspective may perceive that fossil up has two purposes:

Without the optional VERSION argument, it updates the working check-out to the tip of the current branch, as git pull does.
Given a VERSION argument, it updates to the named version. If that’s the name of a branch, it updates to the tip of that branch, as git checkout BRANCH does.

In fact, these are the same operation, so they’re the same command in Fossil. The first form simply allows the VERSION to be implicit: the tip of the current branch.

We think this is a more sensible command design than git pull vs git checkout. (…vs git checkout vs git checkout!)

Closing a Check-Out

The fossil close command dissociates a check-out directory from the Fossil repository database, nondestructively inverting fossil open. (Contrast its closest inverse, git worktree remove, which is destructive in Git!) This Fossil command does not remove the managed files, and unless you give the --force option, it won’t let you close the check-out with uncommitted changes to those managed files.

The close command also refuses to run without --force when you have certain other precious per-checkout data that Fossil stores in the .fslckout file at the root of a check-out directory. This is a SQLite database that keeps track of local state such as what version you have checked out, the contents of the stash for that working directory, the undo buffers, per-checkout settings, and so forth. The stash and undo buffers are considered precious uncommitted changes, so you have to force Fossil to discard these as part of closing the check-out.

Thus, .fslckout is not the same thing as .git!

In native Windows builds of Fossil — that is, excluding Cygwin and WSL builds, which follow POSIX conventions — this file is called _FOSSIL_ instead to get around the historical 3-character extension limit with certain legacy filesystems.

Closing a check-out directory is a rare operation. One use case is that you’re about to delete the directory, so you want Fossil to forget about it for the purposes of commands like fossil all. Even that isn’t necessary, because Fossil will detect that this has happened and forget the working directory for you.

Git Worktrees

There are at least three different ways to get Fossil-style multiple check-out directories with Git.

The old way is to simply symlink the .git directory between working trees:

    mkdir ../foo-branch
    ln -s ../actual-clone-dir/.git .
    git checkout foo-branch

The symlink trick has a number of problems, the largest being that symlinks weren’t available on Windows until Vista, and until the Windows 10 Creators Update was released in spring of 2017, you had to be an Administrator to use the feature besides. (Source) Git 2.5 solved this problem back when Windows XP was Microsoft’s current offering by adding the git-worktree command:

    git worktree add ../foo-branch foo-branch
    cd ../foo-branch

That is approximately equivalent to this in Fossil:

    mkdir ../foo-branch
    cd ../foo-branch
    fossil open /path/to/repo.fossil foo-branch

The Fossil alternative is wordier, but since this tends to be one-time setup, not something you do everyday, the overhead is insignificant. This author keeps a “scratch” check-out for cases where it’s inappropriate to reuse the “trunk” check-out, isolating all of my expedient switch-in-place actions to that one working directory. Since the other peer check-outs track long-lived branches, and that set rarely changes once a development machine is set up, I rarely pay the cost of these wordier commands.

That then leads us to the closest equivalent in Git to closing a Fossil check-out:

    git worktree remove .

Note, however, that unlike fossil close, once the Git command determines that there are no uncommitted changes, it blows away all of the checked-out files! Fossil’s alternative is shorter, easier to remember, and safer.

There’s another way to get Fossil-like separate worktrees in Git:

    git clone --separate-git-dir repo.git https://example.com/repo

This allows you to have your Git repository directory entirely separate from your working tree, with .git in the check-out directory being a file that points to ../repo.git, in this example.

Init in Place

To illustrate the differences that Fossil’s separation of repository from working directory creates in practice, consider this common Git “init in place” method for creating a new repository from an existing tree of files, perhaps because you are placing that project under version control for the first time:

    cd long-established-project
    git init
    git add *
    git commit -m "Initial commit of project."

The closest equivalent in Fossil is:

    cd long-established-project
    fossil init .fsl
    fossil open --force .fsl
    fossil add *
    fossil ci -m "Initial commit of project."

Note that unlike in Git, you can abbreviate the “commit” command in Fossil as “ci” for compatibility with CVS, Subversion, etc.

This creates a .fsl repo DB at the root of the project check-out to emulate the .git repo dir. We have to use the --force flag on opening the new repo because Fossil expects you to open a repo into an empty directory in order to avoid spamming the contents of a repo over an existing directory full of files. Here, we know the directory contains files that will soon belong in the repository, though, so we override this check. From then on, Fossil works like Git, for the purposes of this example.

We’ve drawn this example to create a tight parallel between Fossil and Git, not to commend this .fsl-at-project-root trick to you. A better choice would be ~/museum/home/long-established-project.fossil, if you’re following the directory scheme exemplified in the glossary. That said, it does emphasize an earlier point: Fossil doesn’t care where you put the repo DB file or what you name it.

Fossil’s Timeline Is The “Log”

Git users often need to use the git log command to dig linearly through commit histories due to its weak data model, giving O(n) performance.

Fossil parses a huge amount of information out of commits that allow it to produce its timeline CLI and its /timeline web view using indexed SQL lookups, which generally have the info you would have to manually extract from git log, produced much more quickly than Git can, all else being equal: operations over SQLite’s B-tree data structures generally run in O(log n) time, faster than O(n) for equal n when the constants are equal.

Yet the constants are not equal because Fossil reads from a single disk file rather than visit potentially many files in sequence as Git must, so the OS’s buffer cache can result in still better performance.

Unlike Git’s log, Fossil’s timeline shows info across branches by default, a feature for maintaining better situational awareness. Although the fossil timeline command has no way to show a single branch’s commits, you can restrict your view like this using the web UI equivalent by clicking the name of a branch on the /timeline or /brlist page. (Or manually, by adding the r= query parameter.) Note that even in this case, the Fossil timeline still shows other branches where they interact with the one you’ve referenced in this way; again, better situational awareness.

Emulating `git log`

If you truly need a backwards-in-time-only view of history in Fossil to emulate git log, this is as close as you can currently come:

    fossil timeline parents current

Again, though, this isn’t restricted to a single branch, as git log is.

Another useful rough equivalent is:

    git log --raw
    fossil time -v

This shows what changed in each version, though Fossil’s view is more a summary than a list of raw changes. To dig deeper into single commits, you can use Fossil’s info command or its /info view.

Inversely, you may more exactly emulate the default fossil timeline output with git log --name-status.

What Changed?

A related — though deprecated — command is git whatchanged, which gives results similar to git log --raw, so we cover it here.

Though there is no fossil whatchanged command, the same sort of information is available. For example, to pull the current changes from the remote repository and then inspect them before updating the local working directory, you might say this in Git:

    git fetch
    git whatchanged ..@{u}

…which you can approximate in Fossil as:

    fossil pull
    fossil up -n
    fossil diff --from tip

To invert the diff to show a more natural patch, the command needs to be a bit more complicated, since you can’t currently give --to without --from.

    fossil diff --from current --to tip

Rather than use the “dry run” form of the update command, you can say:

    fossil timeline after current

…or if you want to restrict the output to the current branch:

    fossil timeline descendants current

Symbolic Check-In Names

Note the use of human-readable symbolic version names in Fossil rather than Git’s cryptic notations.

For a more dramatic example of this, let us ask Git, “What changed since the beginning of last month?” being October 2020 as I write this:

    git log master@{2020-10-01}..HEAD

That’s rather obscure! Fossil answers the same question with a simpler command:

    fossil timeline after 2020-10-01

You may need to add -n 0 to bypass the default output limit of fossil timeline, 20 entries. Without that, this command reads almost like English.

Some Git users like to write commands like the above so:

    git log @{2020-10-01}..@

Is that better? “@” now means two different things: an at-time reference and a shortcut for HEAD!

If you are one of those that like short commands, Fossil’s method is less cryptic: it lets you shorten words in most cases up to the point that they become ambiguous. For example, you may abbreviate the last fossil command in the prior section:

    fossil tim d c

…beyond which the timeline command becomes ambiguous with ticket.

Some Fossil users employ shell aliases, symlinks, or scripts to shorten the command still further:

    alias f=fossil
    f tim d c

Granted, that’s rather obscure, but you you can also choose something intermediate like “f time desc curr”, which is reasonably clear.

Detached HEAD State

The SQL indexes in Fossil which we brought up above have a useful side benefit: you cannot have a detached HEAD state in Fossil, the source of untold pain and data loss in Git. It simply cannot be done in Fossil, because the indexes always let us find our way back into the hash tree.

Summary Line Convention in Commit Comments

The Git convention of a length-limited summary line at the start of commit comments is not enforced or obeyed by default in Fossil. However, there is a setting under Admin → Timeline → “Truncate comment at first blank line (Git-style)” to change this for /timeline displays. Alternately, you could enable the “Allow block-markup in timeline” setting under Admin → Timeline, then apply local skin customizations to put that first comment in in bold or whatever suits.

Because this isn’t a typical Fossil convention, you’re likely to find other odd differences between it and Git-based infrastructure. For instance, Vim doesn’t ship with syntax support for Fossil commit messages if you set EDITOR=vim in your shell environment, so you won’t get over-limit highlighting for first-line text beyond the 50th character and such, because it doesn’t recognize Fossil commit messages and apply similar rules as to Git commit messages.

There Is No Staging Area

Fossil omits the "Git index" or "staging area" concept. When you type "fossil commit" all changes in your check-out are committed, automatically. There is no need for the "-a" option as with Git.

If you only want to commit some of the changes, list the names of the files or directories you want to commit as arguments, like this:

    fossil commit src/feature.c doc/feature.md examples/feature

Note that the last element is a directory name, meaning “any changed file under the examples/feature directory.”

Although there are currently no commit splitting features in Fossil like git add -p, git commit -p, or git rebase -i, you can get the same effect by converting an uncommitted change set to a patch and then running it through Patchouli.

Rather than use fossil diff -i to produce such a patch, a safer and more idiomatic method would be:

    fossil stash save -m 'my big ball-o-hackage'
    fossil stash diff > my-changes.patch

That stores your changes in the stash, then lets you operate on a copy of that patch. Each time you re-run the second command, it will take the current state of the working directory into account to produce a potentially different patch, likely smaller because it leaves out patch hunks already applied.

In this way, the combination of working tree and stash replaces the need for Git’s index feature.

This also solves a philosophical problem with git commit -p: how can you test that a split commit doesn’t break anything if you do it as part of the commit action? Git’s lack of an autosync feature means you can commit locally and then rewrite history if the commit doesn’t work out, but we’d rather make changes only to the working directory, test the changes there, and only commit once we’re sure it’s right.

This also explains why we don’t have anything like git rebase -i to split an existing commit: in Fossil, commits are commitments, not something you’re allowed to go back and rewrite later.

If someone does contribute a commit splitting feature to Fossil, we’d expect it to be an interactive form of fossil stash apply, rather than follow Git’s ill-considered design leads.

Until then, there’s the third-party tool fnc and its interactive stash command.

Create Branches at Point of Need, Rather Than Ahead of Need

Fossil prefers that you create new branches as part of the first commit on that branch:

    fossil commit --branch my-branch

If that commit is successful, your local check-out directory is then switched to the tip of that branch, so subsequent commits don’t need the “--branch” option. You simply say fossil commit again to continue adding commits to the tip of that branch.

To switch back to the parent branch, say something like:

    fossil update trunk

(This is approximately equivalent to git checkout master.)

Fossil does also support the Git style, creating the branch ahead of need:

    fossil branch new my-branch
    fossil up my-branch
    ...work on first commit...
    fossil commit

This is more verbose, giving the same overall effect though the initial actions are inverted: create a new branch for the first commit, switch the check-out directory to that branch, and make that first commit. As above, subsequent commits are descendants of that initial branch commit. We think you’ll agree that creating a branch as part of the initial commit is simpler.

Fossil also allows you to move a check-in to a different branch after you commit it, using the "fossil amend" command. For example:

    fossil amend current --branch my-branch

This works by inserting a tag into the repository that causes the web UI to relabel commits from that point forward with the new name. Like Git, Fossil’s fundamental data structure is the interlinked DAG of commit hashes; branch names are supplemental data for making it easier for the humans to understand this DAG, so this command does not change the core history of the project, only annotate it for better display to the humans.

(The version string “current” is one of the special check-in names in Fossil. See that document for the many other names you can give to “amend”, or indeed to any other Fossil command documented to accept a VERSION or NAME string.)

Autosync

Fossil’s autosync feature, normally enabled, has no equivalent in Git. If you want Fossil to behave like Git, you can turn it off:

    fossil set autosync 0

However, it’s better to understand what the feature does and why it is enabled by default.

When autosync is enabled, Fossil automatically pushes your changes to the remote server whenever you "fossil commit", and it pulls all remote changes down to your local clone of the repository as part of a "fossil update". This provides most of the advantages of a centralized version control system while retaining the advantages of distributed version control:

Your work stays synced up with your coworkers’ efforts as long as your machine can connect to the remote repository. At need, you can go off-network and continue work atop the last version you synced with the remote.
It provides immediate off-machine backup of your commits. Unlike centralized version control, though, you can still work while disconnected; your changes will sync up with the remote once you get back online.
Because there is little distinction between the clones in the Fossil model — unlike in Git, where clones often quickly diverge from each other, quite possibly on purpose — the backup advantage applies in inverse as well: if the remote server falls over dead, one of those with a clone of that repository can stand it back up, and everyone can get back to work simply by re-pointing their local repo at the new remote. If the failed remote comes back later, it can sync with the new central version, then perhaps take over as the primary source of truth once again.

Sync Is All-or-Nothing

Fossil does not support the concept of syncing, pushing, or pulling individual branches. When you sync/push/pull in Fossil, it processes all artifacts in its hash tree: branches, tags, wiki articles, tickets, forum posts, technotes… This is not quite “everything,” full stop, but it’s close.

Furthermore, branch names sync automatically in Fossil, not just the content of those branches. That means this common Git command:

    git push origin master

…is simply this in Fossil:

    fossil push

Fossil doesn’t need to be told what to push or where to push it: it just keeps using the same remote server URL you gave it last until you tell it to do something different, and it pushes all branches, not just one named local branch.

The Main Branch Is Called "`trunk`"

In Fossil, the default name for the main branch is "trunk". The "trunk" branch in Fossil corresponds to the "master" branch in stock Git or to the “main” branch in GitHub.

Because the fossil git export command has to work with both stock Git and with GitHub, Fossil uses Git’s traditional default rather than GitHub’s new default: your Fossil repo’s “trunk” branch becomes “master” when mirroring to GitHub unless you give the --mainbranch option added in Fossil 2.14.

We do not know what happens on subsequent exports if you later rename this branch on the GitHub side.

Status Does Not Show Unmanaged Files

The "fossil status" command shows you what files in your check-out have been edited and scheduled for adding or removing at the next commit. But unlike "git status", the "fossil status" command does not warn you about unmanaged files in your local check-out. There is a separate "fossil extras" command for that.

There Is No Rebase

Fossil does not support rebase, on purpose.

This is a deliberate design decision that the Fossil community has thought about carefully and discussed many times, resulting in the linked document. If you are fond of rebase, you should read it carefully before expressing your views: it not only answers many of the common arguments in favor of rebase known at the time the document’s first draft was written, it’s been revised multiple times to address less common objections as well. Chances are not good that you are going to come up with a new objection that we haven’t already considered and addressed there.

There is only one sub-feature of git rebase that is philosophically compatible with Fossil yet which currently has no functional equivalent. We covered this and the workaround for its lack above.

Colorized Diffs

When you run git diff on an ANSI X3.64 capable terminal, it uses color to distinguish insertions, deletions, and replacements, but as of this writing, fossil diff produces traditional uncolored unified diff format output, suitable for producing a patch file.

Nevertheless, there are multiple ways to get colorized diff output from Fossil:

The most direct method is to delegate diff behavior back to Git:
```
  fossil set --global diff-command 'git diff --no-index'
```
The flag permits it to diff files that aren’t inside a Git repository.
Another method is to install colordiff — included in many package systems — then say:
```
  fossil set --global diff-command 'colordiff -wu'
```
Because this is unconditional, unlike git diff --color=auto, you will then have to remember to add the -i option to fossil diff commands when you want color disabled, such as when producing patch(1) files or piping diff output to another command that doesn’t understand ANSI escape sequences. There’s an example of this below.
Use the Fossil web UI to diff existing commits.
To diff the current working directory contents against some parent instead, Fossil 2.17 expanded the diff command so it can produce colorized HTML output and open it in the OS’s default web browser. For example, fossil diff -by will show side-by-side diffs.
Use the older fossil diff --tk option to do much the same using Tcl/Tk instead of a browser.

Viewed this way, Fossil doesn’t lack colorized diffs, it simply has one method where they aren’t colorized.

Showing Information About Commits

While there is no direct equivalent to Git’s “show” command, similar functionality is present in Fossil under other commands:

Show a Patch for a Commit

    git show -p COMMIT_ID

…gives much the same output as

    fossil diff --checkin COMMIT_ID

…only without the patch email header. Git comes out of the LKML world, where emailing a patch is a normal thing to do. Fossil is designed for cohesive teams where drive-by patches are rarer.

You can use any of Fossil’s special check-in names in place of the COMMIT_ID in this and later examples. Fossil docs usually say “VERSION” or “NAME” where this is allowed, since the version string or name might not refer to a commit ID, but instead to a forum post, a wiki document, etc. For instance, the following command answers the question “What did I just commit?”

    fossil diff --checkin tip

…or equivalently using a different symbolic commit name:

    fossil diff --from prev

Show a Specific Commit Message

    git show -s COMMIT_ID

…is

    fossil time -n 1 COMMIT_ID

…or with a shorter, more obvious command, though with more verbose output:

    fossil info COMMIT_ID

The fossil info command isn’t otherwise a good equivalent to git show; it just overlaps its functionality in some areas. Much of what’s missing is present in the corresponding /info web view, though.

Diff Statistics

Fossil’s closest internal equivalent to commands like git show --stat is:

    fossil diff -i --from 2020-04-01 --numstat

The --numstat output is a bit cryptic, so we recommend delegating this task to the widely-available diffstat tool, which gives a histogram in its default output mode rather than bare integers:

    fossil diff -i -v --from 2020-04-01 | diffstat

We gave the -i flag in both cases to force Fossil to use its internal diff implementation, bypassing your local diff-command setting. The --numstat option has no effect when you have an external diff command set, and some diff command alternatives like colordiff (covered above) produce output that confuses diffstat.

If you leave off the -v flag in the second example, the diffstat output won’t include info about any newly-added files.

Branch and Tag Names

Fossil has no special restrictions on the names of tags and branches, though you might want to keep Git's tag and branch name restrictions in mind if you plan on mirroring your Fossil repository to GitHub.

Fossil does not require tag and branch names to be unique. It is common, for example, to put a “release” tag on every release for a Fossil-hosted project. This does not create a conflict in Fossil, since Fossil resolves the ambiguity in a predictable way: the newest match wins. Therefore, “fossil up release” always gets you the current release in a project that uses this tagging convention.

The fossil git export command squashes repeated tags down to a single instance to avoid confusing Git, exporting only the newest tag, emulating Fossil’s own ambiguity resolution rule as best it can within Git’s limitations.

Cherry-Picking and Reverting Commits

Git’s separate "git cherry-pick" and “git revert” commands are options to the fossil merge command: --cherrypick and --backout, respectively. We view this as sensible, since these are both merge operations, and the two actions differ only in direction.

Unlike in Git, the Fossil file format remembers cherrypicks and backouts and can later show them as dashed lines on the graphical timeline.

File Moves and Renames Are Soft by Default

The "fossil mv" and "fossil rm" commands work like they do in CVS in that they schedule the changes for the next commit by default: they do not actually rename or delete the files in your check-out.

If you don’t like that default, you can change it globally:

    fossil setting --global mv-rm-files 1

Now these commands behave like in Git in any Fossil repository where this setting hasn’t been overridden locally.

If you want to keep Fossil’s soft mv/rm behavior most of the time, you can cast it away on a per-command basis:

    fossil mv --hard old-name new-name

Case Study 1: Checking Out a Version by Date

Let’s get into something a bit more complicated: a case study showing how the concepts lined out above cause Fossil to materially differ in day-to-day operation from Git.

Why would you want to check out a version of a project by date? Perhaps your customer gave you a vague bug report referencing only a date rather than a version. Or, you may be poking semi-randomly through history to find a “good” version to anchor the start point of a fossil bisect operation.

My search engine’s first result for “git checkout by date” is this highly-upvoted accepted Stack Overflow answer. The first command it gives is based on Git’s rev-parse feature:

    git checkout master@{2020-03-17}

There are a number of weaknesses in this command. From least to most critical:

It’s a bit cryptic. Leave off the refname or punctuation, and it means something else. You cannot simplify the cryptic incantation in the typical use case.
A date string in Git without a time will be interpreted as “at the local wall clock time on the given date,” so the command means something different from one second to the next! This can be a serious problem if there are multiple commits on that date, because the command will give different results depending on the time of day you run it.
It gives misleading output if there is no close match for the date in the local reflog. It starts out empty after a fresh clone, and while it does build up as you use that clone, Git automatically prunes the reflog to 90 days of history by default. This means the command above can give different results from one machine to the next, or even from one day to the next on the same clone.

The command won’t fail outright if the reflog can’t resolve the given date: it simply gives the closest commit it can come up with, even if it’s months or years out from your target! Sometimes it gives a warning about the reflog not going back far enough to give a useful result, and sometimes it doesn’t. If you’re on a fresh clone, you are likely to get the “tip” commit’s revision ID no matter what date value you give.

Git tries its best, but because it’s working from a purgeable and possibly-stale local cache, you cannot trust its results.

Consequently, we cannot recommend this command at all. It’s unreliable even in the best case.

That same Stack Overflow answer therefore goes on to recommend an entirely different command:

    git checkout $(git rev-list -n 1 --first-parent --before="2020-03-17" master)

We believe you get such answers to Git help requests in part because of its lack of an always-up-to-date index into its log and in part because of its “small tools loosely joined” design philosophy. This sort of command is therefore composed piece by piece:

◆ ◆ ◆

“Oh, I know, I’ll search the rev-list, which outputs commit IDs by parsing the log backwards from HEAD! Easy!”

    git rev-list --before=2020-03-17

“Blast! Forgot the commit ID!”

    git rev-list --before=2020-03-17 master

“Double blast! It just spammed my terminal with revision IDs! I need to limit it to the single closest match:

    git rev-list -n 1 --before=2020-03-17 master

“Okay, it gives me a single revision ID now, but is it what I’m after? Let’s take a look…”

    git show $(git rev-list -n 1 --before=2020-03-17 master)

“Oops, that’s giving me a merge commit, not what I want. Off to search the web… Okay, it says I need to give either the --first-parent or --no-merges flag to show only regular commits, not merge-commits. Let’s try the first one:”

    git show $(git rev-list -n 1 --first-parent --before=2020-03-17 master)

“Better. Let’s check it out:”

    git checkout $(git rev-list -n 1 --first-parent --before=2020-03-17 master)

“Success, I guess?”

◆ ◆ ◆

This vignette is meant to explain some of Git’s popularity: it rewards the sort of people who enjoy puzzles, many of whom are software developers and thus need a tool like Git. Too bad if you’re just a normal user.

And too bad if you’re a Windows user who doesn’t want to use Git Bash, since neither of the stock OS command shells have a command interpolation feature needed to run that horrid command.

This alternative command still has weakness #2 above: if you run the second git show command above on Git’s own repository, your results may vary because there were four non-merge commits to Git on the 17th of March, 2020.

You may be asking with an exasperated huff, “What is your point, man?” The point is that the equivalent in Fossil is simply:

    fossil up 2020-03-17

…which will always give the commit closest to midnight UTC on the 17th of March, 2020, no matter whether you do it on a fresh clone or a stale one. The answer won’t shift about from one clone to the next or from one local time of day to the next. We owe this reliability and stability to three Fossil design choices:

Parse timestamps from all commits on clone into a local commit index, then maintain that index through subsequent commits and syncs.
Use an indexed SQL ORDER BY query to match timestamps to commit IDs for a fast and consistent result.
Round incomplete timestamp strings up using rules consistent across computers and local time of day.

Case Study 2: Multiple "origin" Servers

Now let us consider a common use case at the time of this writing — during the COVID-19 pandemic — where you’re working from home a lot, going into the office one part-day a week only to do things that have to be done on-site at the office. Let us also say you have no remote access back into the work LAN, such as because your site IT is paranoid about security. You may still want off-machine backups of your commits while working from home, so you need the ability to quickly switch between the “home” and “work” remote repositories, with your laptop acting as a kind of sneakernet link between the big development server at the office and your family’s home NAS.

Git Method

We first need to clone the work repo down to our laptop, so we can work on it at home:

    git clone https://dev-server.example.com/repo
    cd repo
    git remote rename origin work

The last command is optional, strictly speaking. We could continue to use Git’s default name for the work repo’s origin — sensibly enough called “origin” — but it makes later commands harder to understand, so we rename it here. This will also make the parallel with Fossil easier to draw.

The first time we go home after this, we have to reverse-clone the work repo up to the NAS:

    ssh my-nas.local 'git init --bare /SHARES/dayjob/repo.git'
    git push --all ssh://my-nas.local//SHARES/dayjob/repo.git

Realize that this is carefully optimized down to these two long commands. In practice, we’d expect a user typing these commands by hand from memory to need to give four or more commands here instead. Packing the “git init” call into the “ssh” call is something more often done in scripts and documentation examples than done interactively, which then necessitates a third command before the push, “exit”. There’s also a good chance that you’ll forget the need for the --bare option here to avoid a fatal complaint from Git that the laptop can’t push into a non-empty repo. If you fall into this trap, among the many that Git lays for newbies, you have to nuke the incorrectly initted repo, search the web or Git man pages to find out about --bare, and try again.

Having navigated that little minefield, we can tell Git that there is a second origin, a “home” repo in addition to the named “work” repo we set up earlier:

    git remote add home ssh://my-nas.local//SHARES/dayjob/repo.git
    git config master.remote home

We don’t have to push or pull because the remote repo is a complete clone of the repo on the laptop at this point, so we can just get to work now, committing along the way to get our work safely off-machine and onto our home NAS, like so:

    git add
    git commit
    git push

We didn’t need to give a remote name on the push because we told it the new upstream is the home NAS earlier.

Now Friday comes along, and one of your office-mates needs a feature you’re working on. You agree to come into the office later that afternoon to sync up via the dev server:

    git push work master      # send your changes from home up
    git pull work master      # get your coworkers’ changes

Alternately, we could add “--set-upstream/-u work” to the first command if we were coming into work long enough to do several Git-based things, not just pop in and sync. That would allow the second to be just “git pull”, but the cost is that when returning home, you’d have to manually reset the upstream again.

This example also shows a consequence of that fact that Git doesn’t sync branch names: you have to keep repeating yourself like an obsequious supplicant: “Master, master.” Didn’t we invent computers to serve humans, rather than the other way around?

Fossil Method

Now we’re going to do the same thing using Fossil, with the commands arranged in blocks corresponding to those above for comparison.

We start the same way, cloning the work repo down to the laptop:

    fossil clone https://dev-server.example.com/repo
    cd repo
    fossil remote add work https://dev-server.example.com/repo

We’ve chosen the new “fossil clone URI” syntax added in Fossil 2.14 rather than separate clone and open commands to make the parallel with Git clearer. See above for more on that topic.

Our remote command is longer than the Git equivalent because Fossil currently has no short command to rename an existing remote. Worse, unlike with Git, we can’t just keep using the default remote name because Fossil uses that slot in its configuration database to store the current remote name, so on switching from work to home, the home URL will overwrite the work URL if we don’t give it an explicit name first.

Although the Fossil commands are longer, so far, keep it in perspective: they’re one-time setup costs, easily amortized to insignificance by the shorter day-to-day commands below.

On first beginning to work from home, we reverse-clone the Fossil repo up to the NAS:

    rsync repo.fossil my-nas.local:/SHARES/dayjob/

Now we’re beginning to see the advantage of Fossil’s simpler model, relative to the tricky “git init && git push” sequence above. Fossil’s alternative is almost impossible to get wrong: copy this to that. Done.

We’re relying on the rsync feature that creates up to one level of missing directory (here, dayjob/) on the remote. If you know in advance that the remote directory already exists, you could use a slightly shorter scp command instead. Even with the extra 2 characters in the rsync form, it’s much shorter because a Fossil repository is a single SQLite database file, not a tree containing a pile of assorted files. Because of this, it works reliably without any of the caveats inherent in using rsync to clone a Git repo.

Now we set up the second remote, which is again simpler in the Fossil case:

    fossil remote add home ssh://my-nas.local//SHARES/dayjob/repo.fossil
    fossil remote home

The first command is nearly identical to the Git version, but the second is considerably simpler. And to be fair, you won’t find the “git config” command above in all Git tutorials. The more common alternative we found with web searches is even longer: “git push --set-upstream home master”.

Where Fossil really wins is in the next step, making the initial commit from home:

    fossil ci

It’s one short command for Fossil instead of three for Git — or two if you abbreviate it as “git commit -a && git push” — because of Fossil’s autosync feature feature and deliberate omission of a staging feature.

The “Friday afternoon sync-up” case is simpler, too:

    fossil remote work
    fossil sync

Back at home, it’s simpler still: we may be able to do away with the second command, saying just “fossil remote home” because the sync will happen as part of the next commit, thanks once again to Fossil’s autosync feature. If the working branch now has commits from other developers after syncing with the central repository, though, you’ll want to say “fossil up” to avoid creating an inadvertent fork in the branch.

(Which is easy to fix if it occurs: fossil merge without arguments means “merge open tips on the current branch.”)

Fossil