There are several terms-of-art in Fossil that have specific meanings which are either not immediately obvious to an outsider or which have technical associations that can lead someone to either use the terms incorrectly or to get the wrong idea from someone using those terms correctly. We hope to teach users how to properly “speak Fossil” with this glossary.
Each definition is followed by a bullet-point list of clarifying details. These are not part of the definition itself.
A collection of one or more computer files that serve some conceptually unified purpose, which purpose changes and evolves over time, with the history of that project being a valuable record.
We qualify the Fossil definition of this common word like this to set aside cases where a zip file or tarball would suffice. If you can pack your project up into such an archive once and be done with it, Fossil is overkill.
And yet that is often just the beginning, since there is often a need for something to be changed, so now you have “version 2” of the archive file. If you can foresee yourself creating versioned archive files for your project, then you probably should be using Fossil for it instead, then using Fossil’s zip or tarball command to automatically produce archives of the latest version rather than manually produce and track versions of the archive. The web version of these commands (
/tarball) are particularly useful for public distribution of the latest version of a project’s files.
Fossil was designed to host the SQLite software project, which is comprised of source code, makefiles, scripts, documentation files, and so forth. Fossil is also useful for many other purposes, such as a fiction book project where each chapter is held in a separate file and assembled into a finished whole deliverable.
We speak of projects being more than one file because even though Fossil can be made to track the history of a single file, it is far more often the case that when you get to something of a scale sufficient to be called a “project,” there is more than one version-tracked file involved, if not at the start, then certainly by the end of the project.
We used the example of a fiction book above, with one chapter per file. That implies scripts for combining those chapters into the finished book and converting that into PDF and ePub outputs, each of which benefit from being version-tracked.
You could instead use a Word DOCX file for the entire book project, with these implicit scripts replaced by Word menu commands. Fossil will happily track that single file’s evolution for you, though there are good reasons to not do that.
Let us say you choose to solve the primary problems brought up in that document by using a format like AsciiDoc instead. You could still use a single file for the entire book’s prose content, but even then you’re still likely to want separate files for a style sheet, a script to convert the HTML to PDF and ePub in a reliably repeatable fashion, cover artwork files, instructions to the printing house, and so forth.
Fossil requires that all the files for a project be collected into a single directory hierarchy, owned by a single user with full rights to modify those files. Fossil is not a good choice for managing a project that has files scattered hither and yon all over the file system, nor of collections of files with complicated ownership and access rights.
A good mental model for a Fossil project is a versioned zip file or tarball. If you cannot easily conceive of creating or extracting such an archive for your project, Fossil is probably not a good fit.
As a counterexample, a project made of an operating system installation’s configuration file set is not a good use of Fossil, because you’ll have all of your OS’s other files intermixed. Worse, Fossil doesn’t track OS permissions, so even if you were to try to use Fossil as a system deployment tool by archiving versions of the OS configuration files and then unpacking them on a new system, the extracted project files would have read/write access by the user who did the extraction, which probably isn’t want you were wanting.
Even with these problems aside, do you really want a
.fslckoutSQLite database at the root of your filesystem? Are you prepared for the consequences of saying
fossil clean --verilyon such a system? We believe Fossil is a poor choice for a whole-system configuration backup utility.
And as a counter-counterexample, a project made of your user’s Vim configuration is a much better use of Fossil, because it’s all held within
~/.vim, and your user has full rights to that subdirectory.
A Fossil repo is similar to an archive file in that it is a single file that stores compressed versions of one or more files. Files can be extracted from the repo, and new files can be added to the repo, but a Fossil repo has other capabilities above and beyond what simple archive formats can do.
Fossil does not care what you name your repository files, though we do suggest “
.fossil” as a standard extension. There is only one place in Fossil where that convention is required, being the
fossil server DIRECTORYcommand, since it serves up
DIRECTORY. If you don’t use that feature, you can name your repo files anything you like.
Cloned and synced repos redundantly store all available information about that project, so if any one repo is lost, all of the cloned historical content of the project as of the last sync is preserved in each surviving repo.
Each user generally clones the project repository down to each computer they use to participate in that project, and there is usually at least one central host for the project as well, so there is usually plenty of redundancy in any given Fossil-based project.
That said, a Fossil repository clone is a backup only in a limited sense, because some information can’t be cloned, some doesn’t sync by default, and some data is neither clonable nor syncable. We cover these limitations and the workarounds for them in a separate document, Backing Up a Remote Fossil Repository.
Rather than be a backup, a Fossil repository clone is a communication method for coordinating work on the project among multiple machines and users: local work done against one repository is communicated up to its parent repository on the next sync, and work done on other repositories that were previously synced up to that parent get pulled down into the local repo clone.
The Fossil philosophy is that all project information resides in each clone of the repository. In the ideal world, this would occur instantly and automatically, but in actual use, Fossil falls somewhat short of that mark. Some limitations are simply technological: a given clone may be temporarily out of communication with its parent repository, network delays exist, and so forth. Fossil is an AP mode system. (This is sometimes called “eventual consistency.”) Other cases come down to administrative necessity, as covered in the backup doc.
Fossil doesn’t require that you have redundant clones. Whether you do or not is a local decision based on usage needs, communication requirements, desire for backups, and so forth.
Fossil doesn’t care where the repositories are stored, but we recommend keeping them all in a single subdirectory such as "
~/fossils" or "
%USERPROFILE%\Fossils". A flat set of files suffices for simple purposes, but you may have use for something more complicated. This author uses a scheme like the following on mobile machines that shuttle between home and the office:
scale=0.8 box "~/museum/" fit move right 0.1 line right dotted move right 0.05 box invis "where one stores valuable fossils" ljust arrow down 50% from first box.s then right 50% box "work/" fit move right 0.1 line dotted move right 0.05 box invis "projects from $DAYJOB" ljust arrow down 50% from 2nd vertex of previous arrow then right 50% box "home/" fit move right 0.1 line dotted right until even with previous line.end move right 0.05 box invis "personal at-home projects" ljust arrow down 50% from 2nd vertex of previous arrow then right 50% box "other/" fit move right 0.1 line dotted right until even with previous line.end move right 0.05 box invis "clones of Fossil itself, SQLite, etc." ljust
We’ve listed the alternatives in decreasing preference order:
Version and revision are near-synonyms in common usage. Fossil’s code and documentation use both interchangeably because Fossil was created to manage the development of the SQLite project, which formerly used CVS, the Concurrent Versions System. CVS in turn started out as a front-end to RCS, the Revision Control System, but even though CVS uses “version” in its name, it numbers check-ins using a system derived from RCS’s scheme, which it calls “Revisions” in user-facing output. Fossil inherits this confusion honestly.
Hash refers to the SHA1 or SHA3-256 hash of the content of the checked-in data, uniquely identifying that version of the managed files. It is a strictly correct synonym, used more often in low-level contexts than the term “version.”
UUID is a deprecated term still found in many parts of the Fossil internals and (decreasingly) its documentation. The problem with using this as a synonym for a Fossil-managed version of the managed files is that there are standards defining the format of a “UUID,” none of which Fossil follows, not even the version 4 (random) format, the type of UUID closest in meaning and usage to a Fossil hash.1
You will find all of these synonyms used in the Fossil documentation. Some day we may settle on a single term, but it doesn’t seem likely.
A version of the project’s files that have been committed to the
repository; as such, it is sometimes called a “commit” instead. A
check-in is a snapshot of the project at an instant in time, as seen from
a single check-out’s perspective. It is sometimes styled
CHECKIN”, especially in command documentation where any
valid check-in name can be used.
There is a harmless conflation of terms here: any of the various synonyms for version may be used where “check-in” is more accurate, and vice versa, because there is a 1:1 relation between them. A check-in has a version, but a version suffices to uniquely look up a particular commit.2
Combining both sets of synonyms results in a list of terms that is confusing to new Fossil users, but it’s easy enough to internalize the concepts. Committing creates a commit. It may also be said to create a checked-in version of a particular revision of the project’s files, thus creating an immutable snapshot of the project’s state at the time of the commit. Fossil users find each of these different words for the same concept useful for expressive purposes among ourselves, but to Fossil proper, they all mean the same thing.
Check-ins are immutable.
Check-ins exist only inside the repository. Contrast a check-out.
Check-ins may have one or more names, but only the hash is globally unique, across all time; we call it the check-in’s canonical name. The other names are either imprecise, contextual, or change their meaning over time and across repositories.
Unlike a check-in, a check-out is mutable. It may start out as a version of a particular check-in extracted from the repository, but the user is then free to make modifications to the checked-out files. Once those changes are formally committed, they become a new immutable check-in, derived from its parent check-in.
You can switch from one check-in to another within a check-out directory by passing those names to the
Check-outs relate to repositories in a one-to-many fashion: it is common to have a single repo clone on a machine but to have it open in multiple working directories. Check-out directories are associated with the repos they were created from by settings stored in the check-out directory. This is in the
.fslckoutfile on POSIX type systems, but for historical compatibility reasons, it’s called
_FOSSIL_by native Windows builds of Fossil.
(Contrast the Cygwin and WSL Fossil binaries, which use POSIX file naming rules.)
In the same way that one cannot extract files from a zip archive without having a copy of that zip file, one cannot make check-outs without access to the repository file or a clone thereof.
Because a Fossil repository is a SQLite database file, the same rules for avoiding data corruption apply to it. In particular, it is nearly a hard requirement that the repository clone be on the same machine as the one where you make check-outs and the subsequent check-ins.
That said, the relative locations of the repo and the check-out within the local file system are arbitrary. The repository may be located inside the folder holding the check-out, but it certainly does not have to be, and it usually is not. As an example, the Fossil plugin for Visual Studio Code defaults to storing the repo clone within the project directory as a file called
.fsl, but this is because VSCode’s version control features assume it’s being used with Git, where the repository is the
.gitsubdirectory contents. With Fossil, different check-out workflows are preferred.
The simple rule for determining whether to use the wiki or embedded docs for any given document is whether the content is considered “evergreen,” as with a Wikipedia article.
While Fossil’s wiki feature does store the history of each document’s changes, Fossil always presents the current version of the document unless you manually go out of your way to dig back into the history. Then, having done so, links from that historical version of the wiki document take you to the current versions of the target documents, not to the version contemporaneous with the source document.
The consequence is that if you say something like…
$ fossil up 2020-04-01 $ fossil ui --page wcontent
…you will not see the list of wiki articles as of April Fool’s Day in 2020, but instead the list of current wiki article versions, the same as if you ran it from a check-out of the tip-of-trunk.
Contrast embedded docs, which are not only version-controlled as normal files are in Fossil, they participate in all the tagging, branching, and other versioning features. There are several consequences of this, such as that Fossil’s special check-in names work with embedded doc URLs:
If you visit an embedded doc as
/doc/release/file.mdand then click on a relative link from that document, you will remain on the release branch. This lets you see not only the release version of a software project but also the documentation as of that release.
If you visit
/doc/2020-04-01/file.md, you will not only pull up the version of
file.mdas of that date, relative links will take you to contemporaneous versions of those embedded docs as well.
If you say
fossil up 2020-04-01 && fossil uiand then visit
/doc/ckout/file.md, you’ll not only see the checked-out version of the file as of that date, relative links will show you other files within that checkout.
Fossil’s wiki presents a flat list of articles, while embedded docs are stored in the repository’s file hierarchy, a powerful organizational tool well-suited to complicated documentation.
Your repository’s Home page is a good candidate for the wiki, as is documentation meant for use only with the current version of the repository’s contents.
If you are at all uncertain whether to use the wiki or the embedded documentation feature, prefer the latter, since it is more powerful and, with the addition of the
/fileeditfeature in Fossil 2.12, it’s nearly as easy to use.
(This very file is embedded documentation: clone Fossil’s self-hosting repository and you will find it as
Some people — and indeed certain parts of Fossil’s own code — use the term “permissions” instead, but since operating system file permissions also play into this, we prefer the term “capabilities” (or “caps” for short) when talking about Fossil’s RBAC system to avoid a confusion here.
- ^ A pre-Fossil 2.0 style SHA1 hash is 160 bits, not the 128 bits many people expect for a proper UUID, and even if you truncate it to 128 bits to create a “good enough” version prefix, the 6 bits reserved in the UUID format for the variant code cannot make a correct declaration except by a random 1:64 chance. The SHA3-256 option allowed in Fossil 2.0 and higher doesn’t help with this confusion, making a Fossil version hash twice as large as a proper UUID. Alas, the term will never be fully obliterated from use since there are columns in the Fossil repository format that use the obsolete term; we cannot change this without breaking backwards compatibility.
You may sometimes see the term “snapshot” used as a synonym
for a check-in or the version number identifying said check-in. We
must warn against this usage because there is a potential confusion
stashcommand uses the term “snapshot,” as does the
undosystem to make a distinction with check-ins. Nevertheless, there is a conceptual overlap here between Fossil and systems that do use the term “snapshot,” the primary distinction being that Fossil will capture only changes to files you’ve added to the repository, not to everything in the check-out directory at the time of the snapshot. (Thus the
extrascommand.) Contrast a snapshot taken by a virtual machine system or a snapshotting file system, which captures changes to everything on the managed storage volume.