Changelog for 66¶

In 0.9.2.0¶

Adaptation to oblibs 0.4.1.0

Overview¶

This release makes 66 more predictable and easier to debug. Commands that refuse leave nothing half done, a service can no longer hang its own stop forever, oneshots get the same care as long-running services, and the upgrade works on a running machine.

Debugging. The verbosity you set for the boot, VERBOSITY in init.conf or on the kernel command line, now reaches every 66 tool: the scandir, the supervisors, the internal helpers and the services themselves. When you debug a boot at a high verbosity, you get their trace as well, not only the one of the 66 command. A VERBOSITY exported in your shell does the same for the 66 commands you type.

Stopping a service. A service that does not stop is now killed after 10 seconds, the default chosen when 66 is built, unless its frontend sets its own [Stop] Timeout; [Stop] Timeout = 0 keeps waiting without end. A shutdown or a 66 stop can no longer wait forever on one stuck service.

Oneshots. A oneshot that runs past its timeout is stopped together with every process it started, instead of leaving them behind. 66 status tells you whether its script failed, was killed by a signal or could not be run at all. And a reactor can now act when a oneshot finishes, the same way it already could for a long-running service.

A broken service no longer stops everything. A tree, the boot and the shutdown now start or stop every service that is fine, and only leave aside a broken service, one whose files cannot be read or that sits in a dependency loop, together with the services that depend on it; the tree is then reported as failed, and the broken service is named with the command that repairs it. Before, the tree stopped at the first refusal and the services after it were never touched, so one bad service could leave a whole boot half started.

What a command does. Every command behaves the same way in every situation. A command that refuses one of the services you named changes nothing at all, instead of acting on some of them. -P acts on the services you named and nothing else, not even their logger. A command that needs a helper of the scandir that is still starting now waits a moment instead of failing.

Running 66 commands at the same time. Two 66 commands run at the same time no longer write over each other: they take turns, and the one that waits says so. A service can therefore run 66 itself, for instance Execute = ( 66 start serviceA ), without blocking the command that started it nor breaking the commands you type meanwhile.

Names. A name now means one thing: either the service that bears it, or the service that provides it with Provide, never both, and only one provider answers to a name at a time. 66 remove on a name only a provider answers to tells you which service to remove, instead of removing the provider.

Frontend files. 66 now reads and runs the Execute line of your frontends itself: execl-cmdline is no longer needed around a command, a syntax error stops the start with a message naming the service, and a variable left empty in [Environment] no longer breaks the start.

Gone. A few options that never worked are removed: -l of 66, 66-hpr and 66-shutdownd, clone= of 66 tree, and the LIVE= line of init.conf. The place where 66 keeps its runtime files, /run/66 by default, is now chosen when 66 is built. If you never used these, nothing changes for you.

The upgrade. It is automatic and works without a reboot. The first 66 command after the upgrade converts the files of 66, moves your aliases, rewrites the dependencies and the stop delay of each service from its frontend, and restarts the helpers whose behaviour changed, so a running machine keeps working, see What happens by itself. Check What you must do first.

What you must do¶

Remove the LIVE= line from /etc/66/init.conf if you edited that file. A key that is not in the table is never read, so a leftover LIVE= is ignored in silence: if you had pointed it somewhere other than the compiled default, the machine now boots on the compiled default without telling you. Drop -l from any script of yours that passes it to 66, 66-hpr or 66-shutdownd, and drop clone= from any call to 66 tree create or 66 tree admin, since all four now stop on an unknown option.

Then check these points against your frontends and your scripts:

66 remove -P is gone. Drop the -P from your scripts: 66 remove refuses it as an unknown option.
A service name ending with -log is refused. The suffix is reserved to the loggers 66 creates. Rename such a frontend, and every Depends, RequiredBy, OptsDepends, Provide or Conflict that names it. A service already parsed under such a name is removed by the migration, see below.
A service or tree name starting with a dot is refused.
A service that must never be killed on its way down needs [Stop] Timeout = 0 written in its frontend. Without the key, the way down is killed after the general stop delay, 10000 milliseconds by default.
An Emit naming an event with a dot is refused by the parse: the dotted names belong to the events 66 raises itself. Rename the event, and the On that listens to it.
A CopyFrom path that does not exist is refused by the parse.
A script that relied on -P taking the logger along must name the logger itself: -P now acts on the named services only.
66 fdholder start|stop|restart is gone. Replace it in your scripts by 66 internal start|stop|restart fdholderd: 66 fdholder refuses these subcommands as unknown.
66 fdholder store|retrieve|list lost their -T option. Pass the timeout to 66 itself: 66 -T 5000 fdholder list instead of 66 fdholder list -T 5000.

New things to explore¶

66 internal (internal):

Each scandir runs three helpers of its own: fdholderd, which keeps the pipes between your services and their loggers, oneshotd, which runs your oneshots, and eventd, which runs your reactors. 66 internal start, stop and restart act on one of them, for instance 66 internal restart eventd when the reactors no longer answer. With -o, root acts on the helpers of the scandir of another user. The command waits until the helper is in the state you asked for, for the time given by 66 -T. 66 internal status shows the state of the three helpers, or of the one you name, with the same line as 66 status.
66 reconfigure -c (66 reconfigure):

Repairs a service that 66 can no longer read. It removes what the service left on disk, as 66 remove does, its log directory and its configuration included, parses its frontend again and puts it back in the state it was in. Commands that meet a broken service name this repair for you.
66 status with several services (66 status):

66 status sshd nginx postgresql shows the three of them, in the order 66 starts them.
Reactors on a oneshot (event system):

A reactor can now act when a oneshot finishes, the same way it already could for a long-running service: From = ( mount-fstab ) with On = ( done ) runs once the mounts are done, On = ( failed ) or On = ( exited:3 ) when they fail. A oneshot sends the end of its script only: done, down or failed, with how its script ended, never starting nor stopping.
How a oneshot ended (66 status):

66 status and 66 runstate now tell how the script of a oneshot ended: signaled with the signal that killed it, exec-failed with the reason it could not be run at all, besides exited with its exit code. A failed start says so as well, with unable to spawn its script: <reason>.
Handing a name over to a provider (Provide):

A provider can take over the name of a disabled service. 66 disable -P dbus then 66 enable dbus-launch hand dbus over to dbus-launch, the services that depend on dbus staying in place; 66 enable dbus takes the name back. While the service bearing the name is enabled, its provider is refused.
Services that run 66:

Two 66 commands run at the same time take turns instead of writing over each other, so a service or a reactor can run 66 itself. A oneshot can start the services it prepared, Execute = ( 66 start serviceA ), and a reactor can remove a service, and itself, once that service goes down:
```
[Start]
Execute = ( 66 remove sddm sddm-remove )

[Event]
EventType = service
From = ( sddm )
On = ( down )
Do = start
```

Interface changes¶

The global -l, --live option is gone:

It claimed to move the supervision directory of a service, and it did not. A service command reads its scandir from the resolve file, not from the option, so 66 -l /somewhere start foo acted on the scandir the service was parsed with. The only commands the option really reached were the ones that write the live directory, and 66 was its own only caller, passing it to the children it spawns at boot.
66-hpr -l and 66-shutdownd -l are gone:

Both stored the path and then overwrote it with the compiled default one line later, so a value passed there had never had any effect.
The LIVE= key of init.conf is gone:

It fed the global -l, and nothing else. The skeleton no longer ships it. A LIVE= on the kernel command line is ignored as well.
66 tree -o clone= is gone:

It copied the resolve file of a tree under a new name. Use 66 tree create with the options you want instead.
66 remove -P is gone (66 remove):

A removal always takes the services that depend on the removed one with it. A neighbour that names the removed service in its RequiredBy keeps its declaration, and the name becomes absent.
-P never takes the logger:

Under -P, start, stop, free, signal, reload, restart, enable and disable act on the named services only, their logger included in what is left aside. A named logger is acted upon alone under -P.
66 state no longer shows toparse, isparsed and torestart:

They told nothing: isparsed was always true, a service not parsed has no state file, and toparse and torestart were never set to true. 66 state -f refuses them as unknown fields.
66 fdholder store|retrieve|list take the timeout of 66 -T:

Their own -T, --timeout option is gone; like every command, they wait for the time given by 66 -T, without limit by default. The wait for a fdholderd that does not listen yet stays bounded to 3 seconds.

Behaviour changes¶

A refused command changes nothing. Every named frontend is checked before anything is written, enabled, started or stopped; a command naming several services refuses all of them when one of them is refused. 66 reconfigure checks the frontends before it stops anything.
RequiredBy is a declaration of the dependent's Depends. A service named there depends on the declaring one exactly as if its own frontend said so. A service named there and never parsed is not parsed by the declaration.
Invalid services are named, not tripped over. A service whose resolve file or addon does not read, or that is part of or depends on a dependency cycle, is refused with 100 by a command whose action touches it, with a message that names its repair: 66 reconfigure -c or 66 parse -f. A command that does not touch it acts and warns at verbosity 2. 66 tree start, 66 tree stop, 66 tree free, the boot and the shutdown leave such a service out, with what depends on it, and start or stop the others; the tree then counts as failed. A service whose frontend is gone is not invalid: it is refused only by 66 parse -f, 66 reconfigure and 66 status naming it, the others act from its resolve file, and 66 remove takes it out.
66 parse -f of a supervised service whose frontend changes its type exits 100 instead of 111, and its message names 66 reconfigure as the other way out.
Enabling a provider while a rival answers to the same name exits 100 instead of 111. A script that tested the code must be adjusted.
The [Start] Timeout of a Notify service is enforced by its supervisor, whatever brings it up: a service not ready in time is brought down, stays down and supervised, and its result reads timeout-start. Without Notify, nothing is timed.
The way down waits for [Stop] Timeout, never for [Start] Timeout, then SIGKILL. A stop that ends with that kill exits with 111. The logger gets the general stop delay, and the finish script is bounded by the same delay, where it got five seconds when the frontend wrote no [Stop] Timeout.
An event means the order acted. An order that changes nothing sends no event: a start of a classic that runs, a stop of a service already down, a signal to a service that is down. 66 reload sends no signal event.
The script of a oneshot is timed by 66-oneshotd. Past its [Start] Timeout, the script and every process it forked get SIGTERM, then SIGKILL once its [Stop] Timeout has passed; past its [Stop] Timeout, its [Stop] script and its children get SIGKILL. The oneshot is down and its result reads timeout-start or timeout-stop; with [Stop] Timeout = 0, no SIGKILL follows the SIGTERM. The script used to be killed alone with SIGKILL, its children left running, and the oneshot read failed.
66 runstate shows an armed oneshot reactor as done or down, the record its script left, as it shows a classic reactor as down. 66 status still shows it waiting, with its last run; a reactor that never ran reads waiting (never run).
66 runstate shows the current state: stopping for a service brought down by 66 stop, finishing for one that died by itself, up for a classic that runs after 66 signal -o, restarting during a crash loop.
A failed start says how the service ended, at verbosity 2: exited with code <n>, killed by signal <n>, not ready within its start timeout, unable to spawn its script: <reason> for a oneshot, and killed at its stop timeout for a stop.
66 env import refuses a value starting with an exclamation mark, as 66 env set does.
Every 66 program follows the VERBOSITY of the boot. 66 boot passes the VERBOSITY of init.conf or of the kernel command line on in the environment, and every 66 program, 66 itself, the scandir and the services included, reads it when it starts; an explicit -v takes precedence. A VERBOSITY exported in your shell now sets the verbosity of the 66 commands you type. The scandir of the boot no longer runs at verbosity 0, and a service started without VERBOSITY in its environment runs 66-execute at verbosity 1 instead of 2. The value is read as a decimal number: 8 and 9 used to send the boot to sulogin.
The Execute of an auto build is run by 66-execute itself. The substituted text is split into words with the execline grammar and the first word is executed, without going through execlineb, and no run.user or finish.user is written in the live directory any more. An empty variable written alone leaves no word, so the execl-cmdline -s { ... } that frontends wrapped around their command to drop empty variables is no longer needed. A syntax error stops the start with syntax error in run_user of: <service>, unmatched { in ... or unmatched } in ..., see Substitution in Execute.
A conflicting service exits 100 instead of 111, at 66 enable (enabled rival) as at 66 start (running rival). The check now covers every service the command brings up, its dependencies included, and 66 restart, 66 reload and 66 tree start make it too: a tree holding a service in conflict with a running one starts nothing and counts as failed.
66 remove and 66 enable refuse a name only a provider answers to. 66 remove network used to remove connmand, and 66 enable network to enable connmand again when it was enabled. Both now exit 100:
```
name: network is provided by: connmand -- '66 remove connmand' instead
name: network is provided by: connmand and no service has this name -- '66 enable connmand' instead
```

66 log tells an unknown name from a service never parsed, with the messages of every other command, where it answered unknown service: <name> in both cases:

service: <name> is unknown: no frontend and no service answers to the name -- is its provider enabled?
service: <name> is not parsed -- parse it with 66 parse <name>

The scandir starts its internal services first, and commands wait for them. At its first scan, 66-scandir starts fdholderd, oneshotd, eventd and, on a boot scandir, the scandir-log catch-all logger, alone. The other services and the readiness notification of 66 scandir start -d wait until each of them is up and ready, or for 3 seconds at most, see internal services. Meanwhile, a command that needs one of them, the start of a oneshot, 66 fdholder, 66 emit, 66 env or the shutdown, waits up to 3 seconds for it to listen, and 66 start, stop, signal, reload and the tree commands wait for the supervisor of a service that is not there yet, for the time of 66 -T, 3 seconds by default. They used to fail at once, the last ones with supervisor not listening. Nothing is waited for when the scandir does not run.
66 boot waits for the readiness of the scandir with CATCHLOG=1 too. The stage 2 starts the trees once the internal services of the scandir are ready, whatever the value of CATCHLOG.
A stopped internal service comes back with its scandir. 66 internal stop stops fdholderd, oneshotd or eventd until the scandir is started again, which always brings its internal services back up. 66 fdholder stop kept the helper down even after a restart of the scandir.
The 66 scandir subcommands no longer act on fdholderd and oneshotd on their own. zombies, abort, nuke and annihilate leave them running, reconfigure and check no longer start them; stop and quit bring them down with the other services.
A restart of 66-eventd runs no reactor. It arms again the reactors it finds and waits for their next event. A oneshot reactor that had already run used to run once more, without any event.
66 start of a stopped oneshot reactor runs it if its event already happened. The source is already in the state the reactor waits for, so the reactor runs at once, as at its first start. It used to wait for the next event of its source.

A reactor whose From names a module, or a member of a module from outside that module, is refused by the parse with 100. It used to be accepted:

service: <name> watches in its From: <module>, a module -- a module sends no event
service: <name> watches in its From: <module>:<member>, a member of the module: <module> -- a service outside a module watches none of its members

A value between quotes opens and closes them on the line of its key. Description = no quotes, Description = "no end and a value written on the next line are refused with invalid format of key: Description. They used to be accepted: the first with an empty description, the keys after it being lost, the two others with whatever text was found.
Text after the closing parenthesis of a value, on its line, is refused with invalid format of key: Execute, as Execute = ( true ) # note already was: Execute = ( true )#Timeout = 3333, Execute = ( true )Timeout = 3333, or a # note after the ) that ends a script written on several lines. They used to be accepted, the text ignored or, for a key glued to the parenthesis, read.
A service is taken out of the supervision once its supervisor has exited. 66 free, 66 remove, 66 stop -u, 66 disable -S, 66 tree free, 66 reconfigure and the shutdown wait for the 66-supervise of each service to exit before removing its live directory, for the time of 66 -T, 3 seconds by default. A supervisor that does not exit in time stops the command with 111:
```
supervisor of: <name> did not exit
```

The live directory used to be removed at once, under the supervisor. - The parse no longer reports a frontend without Version at the default verbosity: key Version at section [Main] was not set -- define it to: <version> is written at the trace level only.

What happens by itself¶

The migration to 0.9.2.0 runs under the lock of the system: commands started together migrate the system once.

The dependencies are written again from the frontends. What 66 deduced and wrote in the dependencies of a service before is dropped. The frontends are parsed as 66 parse does: the configure script of a module runs.
The stop delay of each service is written again from its frontend: the general stop delay when [Stop] Timeout is not written, 0 when it is written 0.
A service named with the suffix -log that is no logger is removed from the disk; its frontend stays. The migration acts on no supervision: such a service that runs keeps running until the next reboot. The migration names it with the services that depend on it, which name an absent service until their frontend is changed:
```
service: <name> ends with the suffix -log reserved to the loggers -- removed by the migration, parse its frontend again under another name; required by: <list>
```
The runas of a logger, and the runas of the io of its service, left implicit by the frontend, are written again as the user of the loggers chosen at the build of 66, whatever their value; a runas the frontend names explicitly is kept. The log directory and its content are left as they are, with the owner the previous release gave them.
The state file of each service is written again without toparse, isparsed and torestart, the copy under the live directory included.
The run and finish scripts of each service are written again, without the VERBOSITY import of the previous release: 66-execute takes its verbosity from the environment.
The daemons whose behaviour changed are started again from this release, in a scandir that runs: 66-shutdownd, so the next shutdown is driven by it, and 66-oneshotd and 66-eventd, which the first command after the upgrade waits for. A scandir that does not run starts the daemons of this release when it starts. fdholderd keeps running, with the descriptors it holds. A shutdown planned before the upgrade, with 66 reboot +<minutes> or 66 poweroff <hh:mm>, is dropped: plan it again once the upgrade is done.
The aliases move into .provide/. An alias whose name a service already bears does not survive: the service keeps the name, the alias loses its claim, and a warning names the service concerned. Both frontends stay. To hand the name over to the provider, 66 disable -P the service then 66 enable the provider.
```
service: <name> has a name a provider claims -- the service keeps it, remove one of them to hand it over
```
A oneshot already started gets its event directory, the one its reactors listen to, so the first shutdown after an upgrade made without a reboot reports the end of each oneshot.

Bug fixes¶

A forced shutdown no longer writes past the end of its argument array:

66 poweroff -f, 66 reboot -f and 66 halt -f build the command line of 66-hpr in an array sized one element short, and the terminating null landed outside it.
66 fdholder applies the options of its subcommands:

--fd, --expire and --delete were accepted and ignored: 66 fdholder store --fd 4 <id> stored the descriptor 0, and a value that was not a number passed without an error.
66 configure --import refuses a single version:

66 configure --import 0.9.1.2 foo, without a destination, did nothing and exited 0. It now exits 100: an import names a source and a destination.
An empty variable no longer erases an Execute script:

A reference to an empty variable in an auto Execute dropped the whole text of the script, and the start failed. key= in [Environment] now gives the variable an empty value, see Substitution in Execute.
66 boot and 66 tree start do nothing when no tree is enabled:

They stopped with unable to bad parameter: Invalid argument and exited 111.
66 snapshot remove "" no longer destroys every snapshot:

An empty name designated the snapshot directory itself. create, remove and restore now refuse an empty name with 100.
66 tree admin -o depends=none forgets the tree in its former dependencies:

The former dependency kept the tree in its requiredby, and a later 66 tree disable of it disabled the tree as well.
66 tree admin -o requiredby= writes both sides:

requiredby=<tree> recorded nothing, and requiredby=none did not remove the tree from the trees that required it.
66 parse -f checks the dependency names:

A forced parse skipped the check, and wrote a resolve file naming a dependency no frontend provides. It is now refused, as the first parse is.
A name too long for its buffer is refused instead of overflowing it:

A From naming a service longer than the maximum, a StdOut path longer than the maximum and a long snapshot name overran a fixed buffer. The parse now refuses the first two, and the snapshot commands size the path to the name.
A module reactor runs its members on every event:

A Do = start module reactor ran its members on the first event only and did nothing on the next ones, where the event system said it runs once per event.
The parse reads a key only at the start of a line:

A capital word in a script was taken for a key. Execute = ( echo HelloWorld ) warned unknown key: HelloWorld), Execute = ( echo TimeoutStart=7 ) was refused, and a script line Timeout=7777 set the start timeout of the service. These scripts are now read as they are written, without a warning.
A [ inside a value or a comment no longer opens a section:

Description = "its [Stop] takes 2s" was cut to its, and the comment # answer [yes|no] warned invalid section name: yes|no.
A commented key is ignored:

# Type = classic and # Version = 1.2.3 were read as keys.
A misnamed section no longer gives its keys to the previous one:

After [Stp], a typo for [Stop], the Execute = ( echo stop ) that followed replaced the script of [Start]. The keys of the misnamed section are now ignored, with invalid section name: Stp.

For packagers¶

The build dependency is now oblibs >= 0.4.1.0.
New build option default-stop-timeout: the stop delay, in milliseconds, given to a service whose frontend writes no [Stop] Timeout. 10000 by default, 0 for without end.
livedir must be an absolute path. meson setup refuses a relative one and trims a trailing slash. Nothing changes it at runtime any more: the LIVE= key is gone from the init.conf skeleton.
lib66 breaks its API and its ABI. Rebuild everything that links against it, 66-tools included, and ship lib66 with the 66 programs of the same release: a 66-oneshotd and a lib66 of different releases do not understand each other.
The installed lib66 carries an absolute RUNPATH, <prefix>/<libdir>, where it was relative.

In 0.9.1.2¶

Overview¶

A follow-up to 0.9.1.1, which moved the temporary of a resolve write into /run/66/tmp and laid that directory down together with the scandir. That holds on a machine booted by 66, and nowhere else: where 66 has never booted, an installer chroot, a container or a package build, the directory does not exist and every command that writes a resolve stops on it. Only 0.9.1.1 is affected.

The migration is automatic and stamps the resolve files at the new version.

What you must do¶

Nothing.

Bug fixes¶

A command that writes a resolve no longer needs a booted machine:

The temporary directory is created on demand, at the first write that finds it missing. An account that owns nothing under the live directory, which is the case of every regular account since /run/66 belongs to root, writes its temporaries in a tmp directory of its own base instead. A 66 enable run as a user inside a chroot therefore no longer depends on a root command having passed there first.

For packagers¶

The test suite no longer needs /run/66/tmp to exist. A build that runs meson test in a clean environment, on a machine where 66 never booted, passes under any account.

In 0.9.1.1¶

Overview¶

A single fix, for a machine whose bootloader hands the kernel a root filesystem mounted read-only: the scandir was not created at all, and the boot ended on a sulogin prompt. Releases 0.9.0.0 and 0.9.1.0 are affected, 0.8.x is not.

The migration is automatic. It stamps the resolve files at the new version and completes the live directory, so a system already running does not have to be rebooted to get the fix.

What you must do¶

Nothing.

Bug fixes¶

The scandir is created again when / is still read-only at boot:

Writing a resolve file builds the database in a temporary file before putting it in place, and that temporary was created under /tmp. At the moment the scandir is created, /tmp is still a plain directory of the root filesystem: the service that mounts a tmpfs over it, like the one that remounts / read-write, lives in the very scandir being created. A root mounted read-only left nowhere to write, and the boot stopped on the first service internal to the scandir, which since 0.9.0.0 carries a minimal resolve file read locally by its supervisor.

The temporary is now built in /run/66/tmp, a directory of the live tree created together with the scandir, on the tmpfs 66 mounts itself and which is therefore always writable. Nothing in your declarations changes.

In 0.9.1.0¶

Overview¶

A maintenance release: one new frontend key, and a long list of fixes on what a command carries with it, on what it reports when it does not get there, and on the event system.

The migration is automatic. It stamps the resolve files at the new version, the addons included, and repairs a list whose count was kept while its content was not. Your declarations are read as they are, nothing of yours has to be rewritten.

Two observable changes are worth a look before you meet them, both in the next section: an exit code that no longer claims success, and a few warnings that now obey the verbosity you asked for. The new key defaults to what your reactors already did.

What you must do¶

Nothing. Two things are worth knowing before you meet them.

A command that does not reach its target now fails. A deadline set with -T that expires, and a transaction interrupted by a signal, used to return 0 while the services were still in transition. They return 111 now, with their own message. A script that read the exit code of such a command was told it had worked; it will now be told the truth.
A few warnings are quieter. They were printed whatever the verbosity asked for, and now follow their level like every other warning. Add -v2 to see them again: the deprecation notices of the moved frontend keys, the refusal to enable a tree of the boot group, and the report of an aborted transaction.

Read the Propagate entry below if you run a reactor whose action restarts a service that much of the system depends on, which is the case the key was made for.

New things to explore¶

[Event] Propagate (66-frontend):

Whether a reactor's Do command walks the dependency chain. It defaults to true, because Do = X means what 66 X means: a reactor is a trigger, not a variant of the command, so your existing reactors are untouched.

Set it to false and the action carries -P. This is what makes a narrow reaction usable: an inotify reactor reloading postgresql when pg_hba.conf changes would otherwise drag every service that depends on the database, and their cascades, with it, SIGHUP-ing the whole application stack for one edited access control line.

Which way the chain is walked depends on the verb, and so does what false costs you. stop, restart, reload, reconfigure and free reach the services that depend on the reactor. start instead reaches what the reactor depends on, and turning propagation off there starts it without bringing its dependencies up, which is rarely what you want.

Bug fixes¶

A module whose member fails no longer takes the session down with it (63f404a):

Starting a module whose member could not start reached a kill on a field the manager never wrote, which held the -1 of its initialiser: the signal went to every process the caller was allowed to signal. As root, at boot, that is the whole system. The field was a leftover of the days when the manager forked, and is gone.
A command that never reaches its target now says so (dfe8366, ef980be):

Three ways of not finishing all returned success. A service that failed its transition, a deadline given with -T that expired, and a SIGTERM or SIGINT that cut the transaction short. All three now leave with 111, each with its own message.
66 start no longer walks the graph backwards (b17e5dd):

A start declares readiness as its target, which the sender did not recognise, so it carried the services that depend on the selection instead of those it depends on. Enabled but uninitialised services were dragged into the batch, which took the whole batch down with them, and at boot the tree never came up.
Nothing released what was waiting on a failure (1a26569, bd0aaa0):

When a service or a tree failed, whatever waited on it stayed in the manager's count, which then never reached zero: the command hung. Waiters are now released and told why. A module, which is computed in place and never enters that count, is no longer subtracted from it either, which took the count below zero and hung the command just the same.
66 signal brings up what the service needs (4267af6, 9e5e698):

-u, -U and -o ask the service to come up, so they now carry its dependencies, as they did before the target replaced the signal string. The direction was also read from the first operation alone, which made -uO and -Ou two different commands.
66 restart of a service that was never started explains itself (9f6f115):

It guarded on the number of services collected rather than on what the graph kept, and an unsupervised service is collected but never becomes a vertex. The restart then handed an empty selection to the start, which complained about a missing argument. It now says the selection is not supervised and suggests starting it.
A tree that ran out of time is signalled (d8899b2, 20c931b):

The timeout tested a pointer that is never null, so the child running the transition was never signalled and the timeout event never raised. An empty tree and a tree whose command could not be launched also reported the same thing, so a failed fork was announced as a success.
The migration no longer carries a count without its list (3dbe731, 52e9f33, 0594314, 6053065):

A count kept beside an offset that means unset makes every reader walk an empty string, which is fatal. The offset now decides, at the point where the resolve files are read, and the migration neither carries such a pair over nor leaves an addon behind at the version it was found at.
66 reconfigure gives every service back the state it was found in (d4f92e7):

A service the user had stopped came back started; so did a service that had never been supervised and was only pulled in as a dependency. The command now takes a snapshot of each service before tearing anything down, and restores exactly that: what was running is started again, what was merely supervised is supervised again and left down, and what was not supervised is left alone.
A reactor that was running is running again after a reconfigure (ae91b3b3):

It used to fall back to armed-and-idle. Whether a service must merely be able to serve or must actually be up is now carried per service, so the reactor's own state decides.
66 reload and 66 signal no longer behave like a stop (bb55e971):

Both inferred their intent from the signal character they were handed, and anything that was not u fell in the stop branch. A reload dropped a down file in the scandir, marking a running service as one not to bring back up, used the stop timeout, and disarmed an event source it was merely signalling.
A Do = reconfigure reactor no longer disarms itself (e388b444):

Its own action unsupervises the service in passing, and the disarm that reached the daemon was taken for a request to disappear: the reactor never reacted again.
requiredby of a leaf service is no longer empty (ddb6118):

A service declaring no dependency of its own owns no addon on disk, and the first string written to it landed at the offset that means unset. The count was right, the list read empty, on every leaf of a dependency tree.
A module member is found by its bare name (52dde235, 8e907276):

Inside a module, a service naming a sibling, in Depends or in the From/On of an [Event] section, is now resolved against the module namespace as well.

In 0.9.0.0¶

Adaptation to oblibs 0.4.0.0

Overview¶

This is a Major release. It completes the direction announced at 0.7.0.0: 66 is now an independent service manager. It no longer builds against skalibs, s6 or execline — the scanner, the supervisor, the logger, the fd holder and the privilege dropper are native programs of the suite. This release also brings the event system promised on the roadmap, and rewrites the boot and shutdown procedures in C.

The ecosystem migration is automatic: the first 66 command run after the upgrade converts your system and takes a snapshot beforehand. You do not have to convert anything by hand.

What you do have to look at is your own files: a few frontend keys moved or disappeared, the init.conf skeleton lost several keys, the rc.* skeleton scripts are gone, and the boot no longer needs to be told which tree to start — which may make one of your own boot services redundant.

This page is organised around that: what you must do, then what happens by itself, then the rest.

What you must do¶

A short checklist. Everything here is detailed further down. The first four are silent — nothing errors out, the behaviour simply changes under you.

If your init.conf sets CONTAINER=1, the key is gone and is now ignored: your container boots as bare metal. Use 66 boot -c instead.
If a frontend uses [Logger] Destination, the key is gone and is now ignored: the service silently logs to the default directory. Replace it with [Execute] StdOut = 66log:/path.
If you customized rc.init, rc.init.container, rc.shutdown or rc.shutdown.final, those files are gone and are no longer read: your changes have no effect. See Your boot.
If one of your boot services starts the enabled trees — anything calling 66 tree start on trees other than the boot tree — 66 now does that natively at boot. Remove it, or it runs twice. See How 66 boots now.
Update your frontends to silence the deprecation warnings: keys that moved section, Build, TimeoutStart/TimeoutStop, s6log. All of them still work and warn — this one is not urgent, but the old spellings will be removed in a future release.
Reboot. The migration rewrites your on-disk state; a running 0.8.2.x scandir keeps its old picture until then.
Packagers: see For packagers — the build dependencies and several meson options changed.

Your frontend files¶

Keys recentered on their section¶

Several [Main] keys that actually describe a transition or the process execution have moved to the section of their domain. The former names TimeoutStart / TimeoutStop merge into a single Timeout key, since the section now disambiguates the start/stop meaning.

Key	Was	Now declare in	Rename
`Notify`	`[Main]`	`[Start]`	—
`DownSignal`	`[Main]`	`[Stop]`	—
`StdIn` / `StdOut` / `StdErr`	`[Main]`	`[Execute]`	—
`TimeoutStart`	`[Main]`	`[Start]`	→ `Timeout`
`TimeoutStop`	`[Main]`	`[Stop]`	→ `Timeout`
`TimeoutStart`	`[Logger]`	`[Logger]`	→ `Timeout`

Nothing breaks. The old placements and the old names are still accepted: they warn at parse time and will be removed in a future release. If a key is declared both at its deprecated location and at its canonical section, the canonical section wins.

For example, this still parses and warns:

[Main]
Type = classic
TimeoutStart = 5000
Notify = 3
StdOut = 66log

and becomes:

[Main]
Type = classic

[Start]
Timeout = 5000
Notify = 3

[Execute]
StdOut = 66log

One thing that is now refused: swapping a [Start] key into [Stop] or the reverse. TimeoutStart or Notify in [Stop], and TimeoutStop or DownSignal in [Start], are a parse error. Their deprecated [Main] spelling is still accepted (as above); only the Start/Stop mismatch is fatal.

Keys removed¶

An unknown key is only ignored with a warning — it is not a parse error. So a frontend still using these keeps working, minus the behaviour you asked for. Check your frontends.

Key	What now
`[Logger] Destination`	Gone. Your custom log directory is silently ignored and the service logs to the default. Use `[Execute] StdOut = 66log:/path`.
`[Logger] TimeoutStop`	Gone. A logger has a single run transition; use `[Logger] Timeout`.

Destination was deprecated-but-honored at 0.8.2.0 — it was automatically converted for you. That conversion is over.

`Build` is deprecated and ignored — the shebang decides¶

Build is still parsed in [Start], [Stop] and [Logger], but it does nothing and warns. The build type is now auto-detected: a script is custom if and only if the first non-blank line of Execute starts with a shebang (#!); otherwise it is built as an execline script.

Two consequences if you relied on Build:

Build = custom with an Execute carrying no shebang used to be a hard error. It is now silently built as execline.
Build = auto (or no Build at all) with a shebang in Execute used to be built as execline. It is now auto-detected as custom.

In both cases, declaring the shebang you actually want is the fix.

`s6log` becomes `66log`¶

The s6log value of StdIn / StdOut / StdErr is renamed 66log, since the logger is now 66-log. s6log is still accepted, converted at parse time with a warning, keeping your destination. Frontend files are not rewritten on disk — edit them yourself to silence the warning.

`MaxDeath` becomes a crash budget, and `MaxDeathInterval` is new¶

MaxDeath no longer means the number of death events the supervisor remembers. Together with the new MaxDeathInterval key it now expresses a crash budget: how many times a service may die within a time window before 66 gives up, declares it failed and stops restarting it. Both are now declared in [Start] — still accepted, deprecated, in [Main].

	`0.8.2.2`	`0.9.0.0`
`MaxDeath`	death events remembered	deaths allowed inside a `MaxDeathInterval` window
accepted range	`0`–`4096`	`0`–`16`
default	`5`	`5`
`MaxDeath = 0`	—	infinite restart, never failed
`MaxDeathInterval`	—	new, milliseconds, default `30000`

If a frontend declares MaxDeath above 16, it no longer parses. Everything else keeps its meaning: the default budget is 5 deaths in 30 seconds.

Only an intrinsic run-then-die consumes the budget: a commanded stop never counts, and a failed exec is reported as exec-failed with a progressive backoff without consuming the budget. An exhausted budget gives state failed with result crash-limit; a new 66 start resets the counter.

The logger default timestamp is honored again¶

A frontend that omits Timestamp in [Logger] used to produce untimestamped log lines, ignoring the compile-time default. It now produces ISO timestamps. Timestamp = none still yields no timestamp; tai and iso were always honored — only the unset case was broken. Expect timestamps to appear in logs where you had none, if you never set the key.

`execl-envfile` rejects newline and NUL in a value¶

A value decoding to a newline (\n, \012, \x0a) or a NUL (\0, \000, \x00) is now a syntax error. The environment is held as a NUL-separated list and re-emitted as newline-separated text; neither byte survives the round trip, so the parse now fails loudly instead of corrupting the value silently. Note this diverges from execline's envfile, which permits a newline in a value.

Your boot¶

How 66 boots now¶

This is the most visible behaviour change of the release, and the one most likely to affect services you wrote yourself.

Before, the boot exec'd the rc.init skeleton script, which started one tree, the one named by init.conf's TREE key, boot by default. Starting the other trees, the enabled ones, was somebody else's job: typically a service inside the boot tree that called 66 tree start on them once the boot tree was up.

Now, 66 boot does the whole thing itself, in memory, with no script and no intermediate 66 process. Stage 2 brings the system up in two waves:

the boot trees: every tree belonging to the boot group (or the single tree named by TREE, if you set it — see below);
then every enabled tree.

So if you have a service whose job is to start the enabled trees, it is now redundant and does the work a second time. Look for anything in your boot tree that calls 66 tree start on trees other than the boot tree, and remove it. This is worth checking whatever your setup: the service is likely to be provided by your distribution rather than written by you.

Two more consequences of the boot running in-process:

You no longer name the tree to boot. Put your trees in the boot group (66 tree admin -o groups=boot <tree>, or at creation) and 66 brings them all up, whatever they are called. No tree in the group is a no-op, not an error.
66 status now reports by boot for services started at boot. Before, rc.init exec'd a fresh 66, which lost the context, and boot-started services were recorded as by user.

A tree in the boot group cannot be enabled, so the two waves never overlap.

`TREE` is still there and now optional¶

TREE is not removed. It is still read from init.conf and from the kernel command line. What changed is that the shipped skeleton no longer sets it, so it now defaults to empty, and empty means group mode.

`TREE`	Boot starts
unset (the new default)	every tree in the `boot` group, then every enabled tree
`TREE=boot` (or any name)	that one tree, then every enabled tree

An existing init.conf that still says TREE=boot keeps working exactly as before. Nothing forces you to move. But it pins the boot to that single tree by name and opts you out of the group mode, so leaving it set is a choice, not an oversight. Clear the key to opt in.

Both forms start the enabled trees in the second wave, that part is unconditional.

`init.conf`¶

Key	Change
`VERBOSITY`	default `1` → `2`
`TREE`	removed from the skeleton, still honored, now empty by default
`CONTAINER`	removed — use `66 boot -c`
`RCINIT_CONTAINER`	removed
`RCINIT`	removed
`RCSHUTDOWN`	removed
`RCSHUTDOWNFINAL`	removed
`RESCAN`	removed (it was never wired to anything)
`LIVE`, `PATH`, `UMASK`, `CATCHLOG`	unchanged

No key was added or renamed.

A stale init.conf still listing the removed keys is not an error — the keys are ignored, without a warning. That is fine for five of them, but CONTAINER=1 is a trap: it used to select the container init, it now does nothing at all, so the system silently boots as bare metal. Pass 66 boot -c instead.

Every key is now optional, and an absent key keeps its built-in default. Previously PATH, TREE and LIVE were exported to stage 2 from buffers that stayed empty when the key was absent.

The default verbosity moves to 2 so that the warnings the internal daemons emit, for instance an event reactor stopped by the anti-loop backstop, are visible without reconfiguring anything.

The `rc.*` skeleton files are gone¶

Removed	Replaced by
`rc.init`	`66 boot` stage 2, in memory
`rc.init.container`	the same stage 2; a container is a branch selected by `66 boot -c`
`rc.shutdown`	`66-shutdownd` stage 3, in memory
`rc.shutdown.final`	the final stage, now compiled C

If you customized any of these, your changes are silently ignored. The files are no longer read, and nothing warns you.

In practice rc.init and rc.shutdown only ever ran 66 tree start|stop ${TREE}, so there is nothing to carry over. The two that carried real hooks:

rc.init.container held the "append the command to launch inside your container here" block and a settable HALTCODE. There is no drop-in replacement: a container is now a fully supervised system, so that command has to be re-expressed as a 66 service. See the new container guide.
rc.shutdown.final was the documented last-resort hook for actions after every filesystem is unmounted. The final stage is now compiled C, so it has no replacement. The upstream skeleton shipped empty and told admins to leave it empty, so most systems are unaffected. But if you used it, there is nowhere to put that logic today. Please report your use case.

Containers¶

Container mode is now selected only by 66 boot -c / --container. CONTAINER, RCINIT_CONTAINER, the CONTAINER_HALTCMD environment channel and the rc.init.container skeleton are all gone.

A container now follows the same path as bare metal, which is the real change: a container is a fully supervised system, not a wrapper around a single command. That is why a customized rc.init.container has no mechanical migration.

The exit-code protocol is unchanged in spirit: pid 1 exits 0 by default, 111 if the boot failed; 66 halt makes pid 1 exit with the EXITCODE found in the halt file; 66 poweroff and 66 reboot report SIGINT and SIGHUP.

A container built from a bare image now sets its own state up instead of aborting: initialising the version marker is best-effort, and boot skips a tree resolution it never used. Three container-only bugs are also fixed, each of which meant a container never stopped: the generated final-stage script was written empty, that script killed itself before it could stop the scanner, and the boot logger echoed every line back into its own fifo and looped forever.

The new container guide covers the whole thing, with build paths for the Obarun base image, Alpine and Debian. The dockerfiles live in contributions/docker/.

The `66-shutdown` program is gone¶

Its job — planned shutdown time, /etc/66/shutdown.allow, the wall message, cancelling — is folded into the 66 halt, 66 poweroff and 66 reboot verbs, which gain -c/--cancel. If you called 66-shutdown directly, call the verb instead.

66-shutdownd moves from bindir to libexecdir — it is an internal daemon, not a user command. Anything hardcoding /usr/bin/66-shutdownd breaks.

Other boot fixes¶

An enabled empty tree hung the boot forever.
CATCHLOG=0 never booted — the readiness fd was wrong, so the boot never got past stage 2.
A boot could silently drop TREE on some machines and not others — the rebuilt environment was not NUL-terminated, so a key lookup could miss.
66 reboot -f crashed — the shutdown verbs no longer route through the system sanitation.
Boot output is no longer interleaved out of order.

What happens by itself¶

You have nothing to convert by hand, that is the point of this section. But automatic is not nothing to do: once the migration has run you reboot, and that is the whole of your part. The rest is here so you know what 66 is doing on your behalf, and the one place it can stop.

The migration is automatic. 66 detects the version change on the first command you run after the upgrade and converts the ecosystem. A snapshot named system@<your current version> (for example [email protected]) is taken beforehand, so the previous state is always recoverable.

Under the hood, 66 converts every service resolve file to a new internal layout, rebuilds your logger run scripts to use 66-log instead of s6-log preserving the backup count, timestamp, max size and destination (the notification flag takes the 66-log default), points the scandir at fdholderd instead of fdholder, and gives oneshot and module services a scandir entry that the new scanner will not try to supervise.

Supported source versions are 0.8.0.0 through 0.9.0.0; anything older is a hard error, not a silent fallback. From 0.8.2.2 it is a single step.

Three things worth knowing:

A reboot is expected. The migration rewrites your on-disk state, but a running 0.8.2.x scandir keeps its old in-memory picture until you reboot.
Each base migrates on its own, as whoever runs 66 first. Root's /var/lib/66 migrates on root's first command; each user's ~/.66 migrates on that user's first command. There is no "migrate everything" command.
The scandir conversion is best-effort, and only a genuine disk error interrupts it. If a resolve file cannot be read or written, 66 skips that one service and warns rather than aborting the whole migration; the warning tells you to force the next reboot with 66 reboot -f, which brings /run back up clean. On a healthy disk none of this fires. Every record the shutdown needs is written, and the reboot is the ordinary one above.

Migrating a 0.8.0.0–0.8.1.1 resolve file now also preserves the logger timeout instead of dropping it, and fixes the logger directory ownership by honoring the service's own logger RunAs instead of forcing the default runner.

New things to explore¶

Nothing here is required. Your services keep working without touching any of it.

The event system¶

A service can now react to something happening elsewhere on the system by running a 66 command on itself. It is entirely opt-in: a service that declares nothing behaves exactly as before. Two roles:

A reactor is an ordinary classic, oneshot or module service with an [Event] section. When its trigger fires, it runs a 66 command on itself (Do) and/or raises a named event (Emit).
A source is a service of the new Type = event. It runs no process and has no [Start] section; 66 start and 66 stop on a source arm and disarm it.

A reactor triggers on a service's state or result, on a signal, on a filesystem event, on a cron expression, on a timer, or on a user event raised with the new 66 emit command. For instance, restarting a service when its configuration file changes on disk:

[Main]
Type = oneshot
Description = "A simple proof of event concept"

[Start]
Execute = ( echo "I restart myself when /etc/my-daemon is edited" )

[Event]
EventType = inotify
From = ( my-config-watcher )
Do = restart

with the source it depends on:

[Main]
Type = event
EventType = inotify
Watch = /etc/my-daemon
On = ( IN_CLOSE_WRITE )

One 66-eventd daemon serves the whole scandir; it is created for you at 66 scandir create time. A reactor that feeds itself is caught by an anti-loop backstop: more than 10 fires within a sliding 10-second window disarms it, and a fresh 66 start is needed to bring it back.

A Do = start reactor of type oneshot or module, held down until its trigger fires, reports the new waiting state.

The full grammar, the vocabulary of each EventType, and the runtime behaviour are on the event system page — see also 66-eventd and 66 emit, and the [Event] section of the frontend reference.

Per-session variables now reach your services¶

A scandir is started long before a session exists — a user one at the first login, the system one at boot — and its environment is frozen at that point. A variable that only a session knows (DISPLAY, WAYLAND_DISPLAY, XAUTHORITY) could not reach the services it supervises at all. The new 66 env command is the way in:

66 env import DISPLAY XAUTHORITY

import copies the values from the environment of the caller, verbatim. Nothing is computed, guessed or checked — a variable that is not set is skipped with a warning. set, unset and list complete the command.

What is published lands in /run/66/environment/<uid>/, one file per variable, and 66-execute merges that directory into the environment of every service of the scandir at each start, no frontend has to declare anything. The merge happens last and verbatim, so a published value overrides the [Environment] section, the service configuration and the ImportFile files, and a ${...} inside it stays literal.

Each publication raises env.<variable>, each withdrawal unenv.<variable>, so a service can wait for the value it needs rather than be ordered around:

[Main]
Type = classic
Description = "Clipit daemon"

[Start]
Execute = ( clipit )

[Event]
EventType = user
On = ( env.DISPLAY )
Do = start

The directory is created and destroyed with the scandir, so nothing published survives a 66 scandir remove or a reboot. Until you reboot, the scandir carried over from 0.8.2.x has no such directory yet, and 66 env says so rather than publishing into a store no service reads. At most 20 variables can be published at a time; the twenty-first is refused with an explicit error.

New commands¶

Command	What it does
`66 log`	read the logs of a service, of the system, or of everything interleaved — with `-f` to follow, `-g` to grep, `-s`/`-u` to bound a time range
`66 emit`	raise a user event by name
`66 env`	publish per-session environment variables to every service of the scandir
`66 runstate`	dump a service's runtime record, the low-level counterpart to `66 state`
`66 fdholder`	manage the scandir's fd holder daemon
`66 suspend`	suspend the system to RAM
`66 hibernate`	hibernate the system to disk

No command was removed.

Long options everywhere¶

Until now every 66 program accepted short options only. The whole suite now accepts long options too — --help, --verbosity, --live, --tree, --timeout, --color, and per-command forms such as --field, --no-name, --follow, --graph. Both --opt=value and the glued -tfoo form work. Short options are unchanged, so your scripts keep working.

66 signal also gains --stop-group, --cont-group and --kill-group to signal a service's whole process group.

`66 status` tells you more¶

Every service type now reports the same way; there is no s6-svstat fork any more, and oneshot, module and event services get the same treatment as classic ones:

Status : enabled, up (pid 731) since 9h 40min by user

Durations are now human-readable (9h 40min, 40min, 45s) instead of a raw second count, and there is no separate ready figure. A notable last result is shown in parentheses after the pid ((exited 1), (signaled 9), (crash)) and nothing on a clean success, as above.

The new part is by <who>. who caused the last transition: user, boot, event, shutdown, or self when the service acted on its own. unsupervised disappears as a status word, waiting appears, and a stopped service in the graph view (-g) now prints down instead of a bare 0.

66 state displays 7 lines instead of 8: the isup flag is gone, since whether a service is up is now read from the runtime record.

Fields are selectable¶

66 status and 66 tree status gain -f/--field to choose the fields to display; -o is deprecated — it still works and warns. The same -f/--field and -n/--no-name pair is now on 66 state, 66 resolve, 66 tree resolve and 66 runstate, which previously took -h only.

Options that no longer exist¶

Gone	Use instead
`66 scandir reload`	`66 scandir reconfigure`
`66 scandir create -s\\|--skeleton`	nothing — boot reads `init.conf` itself
`66 tree start -f`	nothing — `-f`/`--fork` is for `tree stop` and `tree free`
`-o rename` on `66 tree create\\|admin`	nothing — it parsed but never renamed anything, and now errors instead of doing nothing silently

66 tree remove|enable|disable|current and the 66 scandir signal subcommands now accept -h only. They used to advertise administrative options they never applied.

Under the hood¶

You do not have to do anything about this section. It is here because a few of these choices surface in places you may touch.

66 builds against oblibs only. skalibs, s6 and execline are gone from the build entirely. execline remains a runtime dependency, and only that: the generated service scripts still chainload it.

The programs 66 used to borrow are now its own: s6-svscan → 66-scandir, s6-supervise → 66-supervise, s6-log → 66-log, s6-svc → 66-svctl, the s6-fdholder suite → 66-fdholderd, s6-setuidgid → execl-runas.

Oneshot services no longer go through s6-sudo. They are handled by 66-oneshotd, a daemon written for that purpose, not an s6-sudo equivalent. The 66-oneshot and 66-fdholder-filler helpers are gone, absorbed by their daemons.

Where that surfaces for you:

Your loggers now run 66-log. The migration rebuilds their run scripts. Build = custom in [Logger] remains the escape hatch if you want s6-log and its full feature set.
execl-runas is not a drop-in s6-setuidgid: it never introduces UID/GID/GIDLIST in the environment, and the user:group colon syntax is not supported — the account is always a single user name.
The scandir control directory is renamed .s6-svscan → .66-scandir. Anything of yours reaching into it directly must follow.
66-scandir is not s6-svscan: it exposes only -t and -d. The per-scandir limits are compile-time options, and 66 scandir create -L <log_user> is unaffected.

The internal resolve file is also split into an autonomous core plus addons, which lets 66 answer most runtime questions without loading everything. This is invisible to you (the migration rewrites it) but it is a breaking change for external users of lib66, such as 66-tools: resolve_read_g / write_g / check_g / remove_g are renamed to resolve_read / write / check / remove, and the raw primitives take an _at suffix.

Bug fixes¶

Beyond the boot, container and migration fixes above, and all of them reachable from ordinary use:

Parsing any service that has a logger smashed the stack, the run-script buffer was ~18 bytes too small. The same path also used a pointer that a later allocation could invalidate.
A frontend could crash the parser: a […] run of 20 characters or more inside script context (a shell array index, a glob) overflowed a fixed buffer; a frontend whose very first byte is a key read far outside the buffer; a duplicated section overran the section table (now reported as too many sections in frontend file).
Every key value parsed overflowed its allocation by one byte, as did the fallback Description of a service that declares none, and the snapshot name built during an upgrade.
A service name shorter than 6 bytes (svc, ntpd) caused an out-of-bounds read at validation, and every non-instanced name read one byte before its buffer.
66 enable / 66 disable used a dangling tree name, and 66 configure -V read freed memory.
66-shutdownd wrote out of bounds on every shutdown or reboot while generating its final-stage script.
Memory leaks: 66 tree status leaked per tree displayed, 66 remove leaked on every removal of a classic service with a logger, a malformed frontend leaked per parse error, and 66 -h leaked.
Nice was never honored: a service declaring any Nice value ran at nice 19, while a service declaring none correctly ran at 0.
CapsBound and CapsAmbient together killed the service with exit 111 and CAP_SETPCAP is not in the bounding set of the process. Each key worked alone.
The container scandir crash handler could not find 66-nuke on any install where the execline prefix differs from bindir.
A missing User key in [Main] no longer warns, it is a defaulted value, not a problem.
66 halt --help no longer dies with unable to set the tree name on a tree-less system, and 66 free -P works again.
The CLI -T/--timeout override is now a 64-bit millisecond value; it used to be capped at ~49.7 days.

For packagers¶

The only build dependency is oblibs >= 0.4.0.0. skalibs, s6 and execline are no longer needed to build. execline >= 2.9.6.1 is still needed at runtime for the generated scripts.
Meson is the only build system. configure, Makefile, package/ and the build helper scripts are removed. The old system is no longer deprecated — it is gone.
Options renamed: s6-log-user → 66-log-user, s6-log-timestamp → 66-log-timestamp, s6-log-notification → 66-log-notification.
Option removed: sysdeps-dir. Cross-compilation uses with-include-dir and with-dynamiclib-dir.
Options added: execline-bindir and execline-extbindir (default /usr/bin) — the path prefixes for execline binaries in generated scripts, now separable from bindir; and doc-only, which builds the documentation alone, skipping the C sources and the oblibs lookup.
Install layout changed: 66-shutdownd moves to libexecdir; the new 66-scandir, 66-supervise, 66-log and execl-runas go to bindir; 66-eventd, 66-fdholderd, 66-oneshotd and 66-svctl to libexecdir. 66-shutdown, 66-oneshot and 66-fdholder-filler no longer exist.
Documentation builds with meson targets instead of configure-time scripts, so a documentation failure now fails the build. lowdown is still required when with-doc=true. mkdocs is not needed — meson only generates the mkdocs sources; the site build runs in CI. meson compile -C build doc builds the documentation alone.
The test suite grew from 4 to 36 files and runs under address and undefined sanitizers in CI, on merge requests as well as master. Enable it with -D test=true and run meson test -C builddir.
Copyright headers are normalized to a single year per file — relevant if you extract copyright years automatically.

Notes¶

New onboarding guides are available: getting started, a cheatsheet, dependencies, logging, troubleshooting, coming from systemd, OpenRC or runit and running 66 inside a container.
For detailed documentation on new keys and features, refer to the documentation.

In 0.8.2.2¶

Overview¶

This is a bug fix release and convenient build against the latest skarnet software version.

Behavior Changes¶

Allow to reconfigure an unsupervised service (e855a0d):

Previously, reconfiguring a service required it to be supervised. This change allows the reconfigure command to work on services that are not currently supervised — only supervised services will be stopped during the reconfiguration process.

Bug Fixes¶

ChangeDirectory, CapsBound and CapsAmbient sanitization (497eda3):

Fixed a bug where the ChangeDirectory, CapsBound and CapsAmbient entries from the [Execute] section were not correctly sanitized during the resolve phase, potentially leading to missing or corrupted values.
Meson build default option typos (b096fa1):

Fixed typos in meson.build default options where s6_log-user, s6_log-timestamp and s6_log-notification were using underscores instead of hyphens, causing build failures.

In 0.8.2.1¶

Overview¶

This is an hot fix release and only concern a build make by the Meson build system. Compilation with the previous build system is not affected.

Bug Fixes¶

66-shutdownd installation (d75acc6):

Fixed the path installation of the 66-shutdownd program to /usr/bin instead of /usr/libexec.

In 0.8.2.0¶

Adaptation to oblibs 0.3.4.0

Overview¶

This document outlines the changes, new features, and bug fixes introduced in the 66 service management system from version 0.8.1.1 to 0.8.2.0. It is intended for users and administrators upgrading their systems to version 0.8.2.0. Please read these notes carefully to understand the changes.

New Features¶

Introduction of the [Execute] Section (30647e8):

A new [Execute] section has been added to configure tasks executed just before calling exec for a service’s start and stop processes. This section enhances service configuration flexibility.

New Keys in [Execute] Section:
- ChangeDirectory (a6f4bfa): Sets the working directory via chdir().
- Nice (de8aefc): Configures CPU scheduling priority via setpriority().
- UMask (b0e24b4): Sets the file creation mask via umask().
- BlockPrivileges (edfa67d): Restricts privileges via prctl().
- LimitXXX (de8bccd): A family of keys (e.g., LimitNICE, LimitNOFILE, LimitRTPRIO) to configure resource limits via setrlimit().
- CapsBound and CapsAmbient (9e9fd64): Sets bounding set and ambient capabilities without requiring libcap dependencies.
New Conflict Key in [Main] Section (731f708):

A new Conflict key allows the declaration of conflicting services, improving service dependency management.

Non-Mandatory Keys in Service Files (d3fbcee):

The Description, User, and Version keys are no longer mandatory.

Defaults are:

Description: Set to " service".
User: Set to the name of the process owner.
Version: Set to the installed version of 66.
Tree Creation with Enable Option (dceeb88):

Trees can now be enabled at creation time, streamlining tree administration.

Meson Build System Adoption (1f68ff1):

The project has transitioned to the Meson build system, offering improved cross-platform support and build efficiency. The traditional configure and make system remains functional during the transition. Users are encouraged to test the Meson system and provide feedback. See INSTALL_MESON.md for details.

Bug Fixes¶

Logger Destination Ownership (Critical) (6079cf7):

Fixed a significant bug introduced in version 0.8.0.0 that incorrectly set the logger destination owner to root for 66log type loggers and removed the runas field during migration. This release fix the ownership of the logger destination during the migration phase but only for service logger of type StdOut=66log.

Service Crash Handling (c035a94):

Ensured a fatal error is triggered when a service crashes, improving reliability.

Compiler Warnings (c440c4d, cd9a0c2):
Removed GCC/Clang warning messages.
Fixed a compiler warning message.
Migration Process Fixes (96d4b49, 6a3aae7, 9e8d6c9):
Ignored unknown logger directories during migration.
Fixed handling of Master resolve file during migration.
File Descriptor and Resource Handling (e028da2, fe7f14e):
Fixed reading and writing of uint64_t values.
Fixed log file display with status command for StdOut=file:/path/to/file redirection type.
Memory Leak Prevention (7a36e5a):

Addressed a memory leak issue.

Miscellaneous Fixes (b8a963b, 2d96003, b138e98, 6ab8a36, a04c574):
Fixed option name and removed invalid key destination in examples.
Fixed headers and reorganized display order fields.
Fixed typo and ensured Linux-specific compatibility.
Removed useless header file.

Enhancements¶

CI/CD Enhancements (1afbec5, 3677099):
Switched CI/CD pipeline to use the Meson build system.
New Boolean Syntax (67de347):

A new boolean syntax has been introduced, documented for the BlockPrivileges key.

Version Comparison API (044d4ff):

Adapted to the new version_compare API interface and added tests for the function.

Migration Notes¶

Migration Process for 0.8.2.0 (9e8d6c9):

A dedicated migration process has been provided to ensure a smooth transition to version 0.8.2.0. Please, consider the introduced bug below. The migration try its best to fix this issue by itself but it may impact your migration process as the migration process only handle with service logger of type StdOut=66log. If after a restart of a service logger you get a down process, ensure that your loggers directories have the correct permissions. The error should be visible at the /run/66/log/0/current uncaught-logs file.

Logger Destination Ownership (6079cf7):

If you are upgrading from version 0.8.0.0 or 0.8.1.1, verify the ownership of logger destinations for 66log type loggers, as the bug fix may affect existing configurations and upgrade process.

Meson Build System (1f68ff1):

Users are encouraged to switch to the Meson build system and to consider the old build system as deprecated. The old configure and make system remains supported during the transition. Refer to INSTALL_MESON.md for setup instructions.

Notes¶

For detailed documentation on new keys and features, refer to the documentation.

In 0.8.1.1¶

Bug Fixes¶

Adapt correctly reconfigure command to the new graph API and respect -P option.

In 0.8.1.0¶

Adaptation to oblibs 0.3.3.0

Overview¶

This release of the 66 project includes several bug fixes, new features, and improvements to enhance the functionality and maintainability of the service management system. Below is a summary of the key changes introduced in this release.

New Features¶

Service Alias Support (a36a4fad): Added a new Provide key to allow defining aliases for service names, improving service management flexibility.
Environment File Import (4426ab70): Introduced a new ImporFile field in the [Environment] section to support importing environment files.
Simplified Execute Field Writing (099872e0): Enhanced the Execute field to permit spaces, tabs, carriage returns, and newlines (removed during processing) for custom builds, improving readability and usability.
Sanitize Resolve Function (c91ff167): Added a new sanitize_resolve function to ensure the validity of resolve files for trees and services for the targeted version.
Tree Graph Build System (afb18e68): Implemented a new tree_graph_build_system function to improve system graph generation.
Split Info Walk Functions (a058225c): Separated info_walk into service_info_walk and tree_info_walk for better modularity and clarity.
Enum and Macro Enhancements (ddc04da2): Split enum definitions and introduced macros to automate enum, string table, and struct generation. This reduces error-prone code, improves readability, and simplifies maintenance. Added an enum test suite.

Bug Fixes¶

Dependency Cleanup (e2b2c539): Fixed unnecessary entries from the dependency field of the requiredby dependency for the service to remove.
Empty Field Values (b4be03f2): Fixed handling of empty values for fields like contents, dependencies, requiredby, optional depends, and regex fields for directories, files, and infiles, allowing modules to set empty strings (e.g., when services are commented out).
Frontend Field in Modules (a9e7ce23): Corrected the setting of the frontend field for services inside modules.
Frontend Dependencies Preservation (7bc3cab2): Ensured that dependencies declared in the frontend file for a module are always retained.
Tree Name Respect (2b306df2): Fixed an issue where the previous tree name was not respected for services in a selection when the -t option was not used.
Scandir Symlink (df68b262): Added support to remake the service scandir symlink, ensuring its presence.

Improvements¶

Migration Support:
- Continued support for 0.7.2.1 version migration (1cb94c73).
- Provided migration processes for version 0.8.1.0 (d2288e73) and hiercopy field changes (958ee60c).
- Ensured snapshots are created in all migration cases (0270ebda).
Code Maintenance:
- Renamed hiercopy to copyfrom internally (3411e29f) for consistency.
- Adapted to a new enum interface (cf8359c2) and graph algorithm (33287ebd) with new structures, functions, and macros.
- Made the set_info function public (958a0964) for broader usability.
- Removed redundant graph_remove_deps function (10c83ba7) and moved hash functionality to oblibs (7c491133).
- Cleaned up header files (a83db505) and removed unnecessary parameters (8ea7f830) for cleaner code.
Documentation:
- Fixed typos and improved documentation (5f5c8ee9, 89ef88d8).
- Updated the project roadmap (fc97c4e3) to reflect current plans.
Miscellaneous
- Minor Changes (4767cdc4, 89ef88d8): Addressed typos and minor code adjustments with no significant user-facing impact.

Notes¶

This release ensures that users upgrading from previous versions will have an operational system without manual intervention. Migration processes for hiercopy internal field changes and version 0.8.1.0 are handled automatically.
For detailed information on new fields and functions, refer to the updated documentation.

In 0.8.0.2¶

Adaptation to oblibs 0.3.2.1

Bug Fixes¶

Different documentation typo and bugs fix.
Restart command: explicitly call stop and start command separately to avoid notification readiness issues. Retrieve the exact same services states for interdependences services after a restart command.
Frontend parsing: handle all possible cases on '()' format.
reconfigure command: respect -p option.
66-execute: respect exclamation mark to avoid polluting the environment variable.
status command: avoid to crash reading status of service just after using a 66 remove -P command.
scandir command: fix stack overflow at create command.

In 0.8.0.1¶

Bug Fixes¶

Always update state of fdholder and oneshotd daemon even in case of crash.
Avoid badly formatted environment variables coming from EFI.
Clean the algorithm of the shutdown procedure to avoid killing fdholder and oneshotd before the very end of the process.
Fix connection refused error at fdholder connection.
Prevent extra characters during the parse process of the init.conf.

In 0.8.0.0¶

Adaptation to oblibs 0.3.1.0

Synthesized Release Notes¶

Nearing the End of Breaking Changes: This release marks a major step towards the end of planned breaking changes for version 66. Version 0.7.0.0 introduced stability in the directory hierarchy of the 66 ecosystem, and version 0.8.0.0 solidified the stability of frontend file keys. While the implementation of events may happen in version 0.9.0.0, this can potentially occur without breaking existing 66 components(see roadmap).
Introduction of Snapshot Command: A new snapshot command has been added, offering a comprehensive backup solution for the 66 ecosystem. This feature guarantees that you can restore the exact state of the ecosystem before any changes. It also enables the replication of ecosystems across different hosts to maintain consistent states and behaviors. Importantly, using snapshots does not alter the state of running services.
Transparent Upgrade Process: The release includes a new upgrade process that is fully transparent to the user. Whether a migration involves changes to the resolve file or other components, 66 automatically detects version upgrades and makes the necessary adjustments, ensuring continuity with the new version. This process does not affect the state of running services, and an automatic snapshot is created during the migration to facilitate easy recovery if needed.

New features¶

New keys for Input/Output redirection (StdIn, StdOut, StdErr) have been added to control the behavior of standard input, output, and error, respectively.
A new 66 snapshot command allows users to create, remove, list, and restore snapshots, providing enhanced backup and reliability management.
An automatic migration process is now available, ensuring seamless upgrades. This process is completely transparent to the user and is triggered by any 66 command following an upgrade of the 66 program.

Frontend files¶

Key Name Formatting Changes: The @ prefix in key names has been removed, and key names now start with a capital letter. Additionally, the - character has been removed from composite names and also replaced by a capital letter. 66 will automatically migrate the frontend files present on your system, but this will only apply to active services(i.e., those listed when running 66 tree status with the field contents). You may need to manually translate your own frontend files.

To help with this, use the script below by running: ./migration_service.sh /path/to/my/<FrontendFile>

#!/bin/sh

service="${1}"

sed -e "s:\[main\]:\[Main\]:g" \
    -e "s:@type:Type:g" \
    -e "s:@description:Description:g" \
    -e "s:@version:Version:g" \
    -e "s:@depends:Depends:g" \
    -e "s:@requiredby:RequiredBy:g" \
    -e "s:@optsdepends:OptsDepends:g" \
    -e "s:@options:Options:g" \
    -e "s:@flags:Flags:g" \
    -e "s:@notify:Notify:g" \
    -e "s:@user:User:g" \
    -e "s:@timeout-kill:TimeoutStart:g" \
    -e "s:@timeout-up:TimeoutStart:g" \
    -e "s:@timeout-finish:TimeoutStop:g" \
    -e "s:@timeout-down:TimeoutStop:g" \
    -e "s:@maxdeath:MaxDeath:g" \
    -e "s:@down-signal:DownSignal:g" \
    -e "s:@hiercopy:CopyFrom:g" \
    -e "s:@intree:InTree:g" \
    -e "s:\[start\]:\[Start\]:g" \
    -e "s:@build:Build:g" \
    -e "s:@runas:RunAs:g" \
    -e "s:@execute:Execute:g" \
    -e "s:\[stop\]:\[Stop\]:g" \
    -e "s:\[logger\]:\[Logger\]:g" \
    -e "s:@destination:Destination:g" \
    -e "s:@backup:Backup:g" \
    -e "s:@maxsize:MaxSize:g" \
    -e "s:@timestamp:Timestamp:g" \
    -e "s:\[environment\]:\[Environment\]:g" \
    -e "s:\[regex\]:\[Regex\]:g" \
    -e "s:@configure:Configure:g" \
    -e "s:@directories:Directories:g" \
    -e "s:@files:Files:g" \
    -e "s:@infiles:InFiles:g" \
    -i ${service}

See Rosetta Stone for the list of keyword name changes.

Behavior enhancements¶

The 66 status command now provides an overview of all system services, organized by tree, when no specific service is specified.
The Identifier feature has been expanded: in addition to @I, 66 now recognizes seven new identifiers to facilitate the creation of more generic frontend files.
Module configuration: Two more news variables are exported at execution time of the configure script called MOD_ENVIRONMENT_ADMDIR=/etc/66/environment and MOD_ENVIRONMENT_USERDIR=.66/environment.
The 66 remove command is designed to always succeed, ensuring that no service becomes impossible to remove.

Deprecated and Obsolete keywords¶

The deprecated key @shebang has been completely removed and is no longer recognized by the parser.

The Destination key (previously @destination) in the [Logger] section is now deprecated and replaced by new StdIn, StdOut, and StdErr keywords. For compatibility, the parser will automatically handle the conversion.

Removal of Deprecated Options:

For the disable command: -F, -R options are removed.
For the parse command: -F, -C,-c,-m` options are removed.
For the stop command: -X, -K options are removed.

Bug Fixes¶

Command line-defined timeouts now take precedence.
Do not create finish script if section [Stop] doesn't exist.
Proper handling of errno during signal reception.
Fixed the resolution of the source frontend file in a module's resolve file for a service.
Prevented crashes when encountering an empty field in a seed file.
Corrected parsing errors when an unknown key is found at the end of the frontend file.
The 66 tree status command now only displays services associated with a specific tree.
Fixed behavior of the -l option at execl-envfile program.
Always deal with logger at disable time.

In 0.7.2.1¶

Bugs fix:
- Respect the VERBOSITY during the whole boot process.
- Fix importation of the kernel command line environment variables.
- Fix order of precedence for the build of the environment variable used during the whole boot process.

In 0.7.2.0¶

Adapt to oblibs 0.3.0.1`
Bugs fix:
- Logger destination directory for a oneshot or logger destination directory for a service inside a module are now properly deleted at remove subcommand invocation.
- Allow empty modules.
- Respect the associated tree of a service at enable process time if it were already enabled.
Behavior changes:
- The -s and -c options to the configure subcommand are now mutually exclusive.
- Service inside module cannot be reconfigured alone. The whole module must be reconfigured instead.
- To be consistent between subcommand, the scandir reload subcommand was renamed to scandir reconfigure. The old subcommand invocation still work.
- The configuration directory used to build a module is not kept anymore after a parse process. A temporary is used instead.
New features:
- The contains of the /etc/66/environment or .66/environment for root and regular user respectively is imported by default at scandir start command invocation. Although this can be changed at compile time by passing the --with-sysadmin-environment=DIR, --with-user-environment=DIR for root and regular user respectively
Enhancement:
- Avoid to crash at unsupervise process with a corrupted service. This allows to avoid to be stuck at remove command invocation for a corrupted service or a corrupted state service within module.
- Avoid to crash at stop process if the service is already unsupervised.
- Earn rapidity at unsupervise process by sending only one signal for all services to the fdholder service.
- Use temporary directory for the creation of a the module configuration directories.
- Complete review of the struct enum ecosystem and passing from an O(n*m) algorithm to O(1). This also serves as preparation for future releases that will include new sections and key fields.

In 0.7.1.1¶

Bugs fix:
- corrected environment computation and string cleanup from the frontend key.
- fixed the value of the intree key.
- ensured sufficient space for renaming interdependencies.
- fixed the use of parse_list functions, even in the case of multiple lines.

In 0.7.1.0¶

Adapt to oblibs 0.3.0.0
Bugs fix:
- make exclusive -s and -c options at configure command.
- respect -s option for edition at configure command.
- provide wall command documentation. This command was added from 0.7.0.0 tag but the documentation wasn't.
- fix closing string at parse time.
- typo fix.
Complete review of the execl-envfile program. Adapting it to the new lexer and environ familly functions make it faster and easier to debug and use less HEAP memory.
Remove deprecated env key of the @options field.

In 0.7.0.2¶

Hot fix:
- Always build the graph with the service associated to a module.

In 0.7.0.1¶

Bug fix:
- Fix buffer overflow at tree administration
- Typo fix

In 0.7.0.0¶

Adaptation to skalibs 2.14.1.0
Adaptation to execline 2.9.4.0
Adaptation to s6 2.12.0.3
Adaptation to oblibs 0.2.1.0

This release marks a significant rewrite of 66, introducing a new UI and serving as a comprehensive service supervision suite and independent service manager.

Primarily, expect no compatibility with previous versions due to:

The removal of s6-rc support. 66 is now a complete independent service management based on s6 for init and service supervision.
A complete overhaul of the folder structure for storage and runtime directories, simplifying it considerably.
An overhaul of tree behavior. Trees now function as services, and a complete tree dependency graph has been implemented.
Services can now depend on each other regardless of whether the service is declared on the same tree or the declaration order of the tree. For instance, if service Sb depends on service Sa and Sa is within TreeB while service Sb is within TreeA, and TreeB depends on TreeA, launching TreeA will start Sa even if TreeB isn't started first. When TreeB is executed, Sb will find Sa already started and commence directly.

For UI changes, frontend file convertion and clean of the 66 architecture, a Rosetta stone is available.

Frontend Files¶

New fields have been added:¶

@requiredby: Specifies which service is required-by another service. For example, if service Sa declares service Sb as required-by Sa's dependency, Sb won't start until Sa does. This allows building complex graph structures without modifying every frontend file of each service.
@earlier: Declares any service as an earlier one, starting as soon as the scandir is running, similar to tty12. This field is mandatory for services intended to start earlier.
@intree: Specifies the tree to use at enable/start time.

The following field has been removed:

@extdepends: No longer necessary as services can depend on any service regardless of the tree used.

Frontend files for regular account must be now localized at /usr/share/66/service/user, /etc/66/service/user or ${HOME}/.66/service.

Behavioral changes:¶

@options:
- pipeline: This option was removed. It was only present for s6-rc.
- env: This option was removed. The simple declaration of the [environment] section is sufficient to activate the options.
@shebang: Deprecated but kept for compatibility reasons. Declare your shebang directly within the @execute field. Refers to frontend documentation for futhers information.
@build: Not mandatory anymore, as it will be declared 'auto' by default.
@addservices: This options was removed from the [regex] section.

The classic type now accepts the fields @depends and @requiredby. The classic type replaces the longrun type.

Logger destinations for oneshot type services can now be declared on a tmpfs directory, particularly useful during boot time.

Services can be started without being enabled first. In this case, the service won't start on the next reboot.

The bundle and longrun types have been removed, replaced by classic, oneshot, and module types. For compatibility reasons, if your old frontend file declares the service as a longrun type, the parser will convert it to a classic type automatically. No automatic conversion is made for services of type bundle.

[Environment] Section¶

This section now allows reusing the same variable from the actual environment. For instance:

socket_name=!/run/dbus/system_bus_socket
cmd_args=!--system --address=unix:path=${socket_name}
PATH=/usr/local/bin:${PATH}

The order of key-value pair declaration doesn't matter:

cmd_args=!--system --address=unix:path=${socket_name}
socket_name=!/run/dbus/system_bus_socket

Variable name must be between ${} to get it value. For instance, $var is not replaced by its value.

Double-quote within value must be escaped with backslash \. Refers to the updated documentation of execl-envfile for futhers information.

Module Changes¶

The module directory structure has been completely redesigned for better intuitiveness and comprehensiveness. Expect no compatibility with the previous version; a rewrite is required if you use module on your system.

A module cannot contain another module; instead, you can declare it as a dependency via @depends or @requiredby. These can also be specified through the configure module script.

Refer to the specific module page for furthers information.

Trees¶

Trees now react as services regarding graph dependencies. You can declare a tree depending on or required-by others.

A default named global tree is provided. Services without their localization defined or users not specifying a tree to use will be defined within that tree.

A seed file can be provided for automatic tree configuration at creation time. For example, it defines depends/requiredby dependencies of the tree.

If a service declares a non-existing tree, the tree will be created automatically with a default configuration, but without any depends/requiredby dependencies. To configure the tree with specific requirements at creation time, provide and install seed file with your service.

Configure Script¶

Removed flags: - --with-system-module=DIR - --with-sysadmin-module=DIR - --with-user-module=DIR

Added flags: - --with-default-tree-name=NAME - --max-path-size=BYTES - --max-service-size=BYTES - --max-tree-name-size=BYTES - --with-system-seed=DIR - --with-sysadmin-seed=DIR - --with-user-seed=DIR

The slashpackage convention was removed.

Skeleton files¶

The skeleton files shudown, reboot, poweroff and halt are removed and replaced by 66 reboot, 66 poweroff, 66 halt command respectively.

Code Changes¶

The code has been largely rewritten and simplified, offering more features with approximately the same number of code lines. Additionally, the code now uses less HEAP memory, although this optimization is ongoing.

The parser was completely rewritten and heavily optimized, significantly reducing the time to parse a service(by three times).

The start process was rewritten due to the removal of s6-rc. Oblibs now provide general functions to build any Acyclic graph.

The code for the module part was revamped and greatly simplified.

In 0.6.2.0¶

WARNING: 66-update is no longer compatible with 66 version under 0.5.0.0.
Adapt to skalibs 2.11.0.0
Adapt to execline 2.8.1.0
Adapt to s6 2.11.0.0
Adapt to s6-rc 0.5.2.3
Adapt to oblibs 0.1.4.0

In 0.6.1.3¶

Adapt to skalibs 2.10.0.3.
Adapt to execline 2.8.0.1.
Adapt to s6 2.10.0.3.
Adapt to s6-rc 0.5.2.2.
Bugs Fix:
- 66-boot: fix call of 66-scandir -c option.
- configure script: fix installation of skel/init.conf file.
Configure script: remove the slashpackage convention.

In 0.6.1.2¶

Bugs fix:
- Avoid segmentation fault if 66-init is used without arguments. Thanks Timothy Murphy.
- Fix the build of man pages.
- Documentation fix.
66-env:
- Bug fix:
  - Create the user configuration file at -r option if it doesn't exist yet.
  - Fix -c option.
66-inservice
- Do not output warn message from upstream configuration file at contents display.
- Bug fix:
  - Respect the number passed by -p options.

In 0.6.1.1¶

Bugs fix:
- remove the exclamation mark from variable at module parse time before passing it to the script configure

In 0.6.1.0¶

Bugs fix:
- Avoid to crash if no tree is enabled yet.
- Documentation fix.
- Handle correctly the return value of s6_rc_servicedir_manage.
- Fix 66-all command at s6-svscan control file of a regular user scandir creation.
- Handle correctly a same(or commented) key=value pair on multiple service configuration file.
execl-envfile:
- new option:
  - -v: allows verbosity changes.
- Bug fix:
  - Handle correctly a same(or commented) key=value pair on multiple service configuration file.
66-env:
- new option:
  - -e editor: Allows to choose the editor to use for the edition of the configuration file.

In 0.6.0.1¶

Bugs fix:
- Respect variables value from kernel command line.
- Fix release number at 66-upgrade.html file.
- Fix string lenght at regex replacement time.

In 0.6.0.0¶

Adapt to skalibs 2.10.0.0
Adapt to execline 2.7.0.0
Adapt to s6 2.10.0.0
Adapt to s6-rc 0.5.2.1
Adapt to oblibs 0.1.2.0
Bugs fix:
- 66-tee -c: check if backup is empty.
- parse_module: avoid infinite loop at dependencies resolution on sub-module.
- 66-disable: do not crash if the service is not enabled at state check.
- execl-envfile: fix parse process for a configuration file.
66-disable:
- new options:
  - -R: removes configuration files and logger directory of the service. So, no components of the service is kept.
66-env:
- -r can be passed multiple time
66-tree:
- -U option is now deprecated and passed to 66-all.
66-all:
- new argument:
  - unsupervise: unsupervise all services of a tree. This argument replace the 66-tree -U command.
66-inservice:
- It displays the absolute path of the file at Environment file field.
- The -c option is now no longer available.
66-update:
- the -c option is now no longer available.
66-env:
- the -d option is now no longer available.
66-enable:
- the -c|m|C|i is now deprecated. The configuration file is handled automatically (see Service configuration file for further information.)
- -I: new options to avoid the copy of a modified configuration file.
66-parser:
- deprecated the -c|m|C to follow changes about the 66-enable tool.
66-scandir: the interface change.
- The owner argument is now an option(-o) and the arguments are create|remove.
- The up option is passed to 66-scanctl.
- The -e option is passed to 66-scanctl.
- New options:
  - -B: specifies to create a scandir for a container. In this case, a /run/66/scandir/container/halt file is created(see 66-boot for further information).
  - -c: do not set the catch-all logger.
skeleton file: (see 66-boot for further information).
- init.conf:
  - New variables:
    - CONTAINER: specifies to boot inside a container.
    - RCINIT_CONTAINER: absolute path to the init file used to boot inside a container.
    - CATCHLOG: create or not the catch-all logger.
  - Variable removed:
    - ISHELL.
- rc.init.container: file used in case of boot inside a container.
- ishell: not longer available.
66-scanctl: the interface change.
- arguments signal are now: start|stop|reload|nuke|zombies.
- New options: this two options are only valid for a start signal.
  - -d: file descriptor to use for readiness notification.
  - -t: perform a scan every rescan milliseconds.
66-boot:
- New option:
  - -z: to use color
66-shutdownd:
- swicth -B and -C (which is renamed to -c) to be consistent with the 66-scandir tool.
frontend:
- nosetsid value at @flags key is deprecated and have no effects if it defined.
- new format for the @options key:
  - log: the logger is automatically created even if the value log is not set. If you don't want a logger at all prefix the value log with an exclamation mark as it: !log.
  - env: if the environment section is defined, the env value is not mandatory. Also, the old behavior is always valid: if the env value is set, the [envrionment] must be set.
html documentation: documentation is now versionned.
New tool:
- 66-nuke: this tool is a strict copy of the [s6-linux-init-nuke] (https://skarnet.org/software/s6-linux-init/s6-linux-init-nuke.html) tool.

In 0.5.1.0¶

Bugs fix:
- Avoid to parse twice a service coming from a service of type module.
66-disable: new options: - -F: forces the service to be disabled even if it's already marked disabled. See the 66-disable documentation page for further information.

In 0.5.0.1¶

Bugs fix:
- Handle old service format where the version directory doesn't exist. This is last release handling this case.
- 66-tree: Remove the live tree state directory in any case even if the directory is empty.

In 0.5.0.0¶

This is a Major release, you need to update your trees with 66-update tool. If you skip from a version earlier than 0.4.0.1, the 66-update will not work. In this case, you need to reconstruct your trees manually. Downgrading to a previous version will not work either, due to the new format of the resolve inner files.

Adapt to oblibs v0.1.0.0
Pass the writing of the resolve files to a CDB format. From that point onward the 66-update will no longer be mandatory, even after major version release.
Adapt 66-update to the new CDB format where applicable.
Bugs fix:
- Write the dependencies of the contents file for a module in the proper order as to avoid multiple repeated names.
- 66-update:
  - Get the correct exit status at tree contents process.
  - Fix segmentation fault when a crash occur at enable time.
- Compilation configure script improvements and bugs fix.
- Fix the location of the modules directory service at enable time. The place of the module frontend file determines the place of the result process for the service module directory.
- Fix the creation of the logger directory when field @build is not set.
- Fix -r signal and -R signal behavior at 66-start tool.
The rc.init skeleton file does not launch the ISHELL script anymore during a crash at stage2. It's the responsability of the sysadmin to deal with this error at his convenience.
@destination field in section [logger] is no longer mandatory at the use of @build with value custom.
66-env:
- -e is now the default option.
- -L displays now all key=value pair from all files found at the configuration directory. new options:
  - -c: changes the current symlink to the specified version.
  - -V: displays available version.
  - -i: import an extra configuration file from one version to another.
  - -s: handle a specific version for command -L|V|e|r.
All key=value pairs from init.conf skeleton file can be now passed to the kernel command line. Also, variables from init.conf are now passed to the rc.init skeleton file as arguments.
@version field is now mandatory.
The version symlink of the configuration file points now to the configuration directory instead of the configuration file. This allows overwriting a same key=value with the writing of an extra configuration file instead of changing the upstream file.
66-in{resolve,state}: field Real logger name name is renamed Real_logger_name.
66-enable:
- Allow the use of a different version of a configuration file than the frontend service file, if any of -c/m/C options are used. new option:
  - -i: import extra configuration files from a previous version.

In 0.4.0.1¶

Hot fix: @build is no longer mandatory even for [stop] section.

In 0.4.0.0¶

This is a Major release, you need to update your trees with 66-update tool.

new extra-tools:
- execl-envfile: this tool come from 66-tools software and was incorporated inside the 66 software.
  - It parses now all files found at a directory by alphabetical order.
Documentation is now written in markdown format. Lowdown software it necessary to build in html and manpages format.
Bugs fix:
- 66-update: respect the tree start order.
- 66-tree: fix behavior when we have only one tree enabled.
- 66-inservice: do not crash in case of empty log file.
- Fix reverse dependencies search for a service with type module.
- Respect timestamp format given at compile time for the uncaught-logs.
- Accept empty environment file.
Configuration service files: These files is now automatically versioned in function of the field @version declared on the frontend service file e.g. /etc/66/conf/<service>/version/<file>. A symlink called version is created or updated at /etc/66/conf/<service>/version. This symlink point to the configuration file currently in use. Also, you can now write and save your own configuration file for a service inside the configuration service directory. All files found on that directory will be parsed and exported to the environment of the service at start time. The parse process is made on alphabetical order.
@hiercopy field accept now relative path.
@build field is no longer mandatory. If it not set, auto is the default.
@version field must be in the form digit.digit.digit e.g. 0.10.2.
module service type:
- General code improvement and evolve.
- Add @addservices field at [regex] section.
- Disabling a service contained inside a module is not possible. This break entirely the module operation.
- Sub module directories is no longer mandatory. 66-enable will create it if it doesn't exist.
- 66-enable export some variable environment at configure script runtime-see modules-service documentation page.
- Fix cyclic dependency: a module cannot call it itself.
- Redesign of the inner directories structure: instantiated service service must be define at service@/ sub-directory where other type go to service/ sub-directory. .configure/ is now named configure.
- All configuration files for each service contained inside the module is written inside the versioned directory of the module e.g. /etc/66/conf/<module>/<version>/<service>/<service_version>/file. This allows to have multiple module using a same service with a specific configuration for each service.
66-inresolve: add field Real_exec_run,Real_exec_finish for the service and logger associated to display the exact contain of the run/up and finish/down files respectively.
66-enable: The absolute path of the frontend service can also be set but must contain the primary path of the path define at compile time by --with-system-service or --with-sysadm-service or --with-user-service e.g. /etc/66/service/lamp/httpd.
66-env:
- General code improvement.
- Follow the change about the versioned configuration service file.

In 0.3.0.3¶

Hot fix: Avoids overwriting the current file

In 0.3.0.2¶

Fix check and installation of user configuration file directory.
Add missing 66-inresovle and 66-instate* documentation.

In 0.3.0.1¶

Fix @optsdepends and @extdepends behavior: These two fields now respect correctly the start order of the service's dependencies.
66-inservice change: The field Optional dependencies and External dependencies displays now the name of the tree where the service is currently enabled after the : colon mark if any:
```
External dependencies : dbus-session@obarun:base gvfsd:desktop
Optional dependencies : picom@obarun:desktop
```

In 0.3.0.0¶

This is a Major release, you need to update your trees with 66-update tool.

Bugs fix
skalibs dependency bumped to 2.9.2.1
exeline dependency bumped to 2.6.0.1
s6 dependency bumped to 2.9.1.0
s6-rc dependency bumped to 0.5.1.2
oblibs dependency bumped to 0.0.6.0
Add -z to all tools to enable colorization:
- 66-in{service,tree} and 66-update tools change the -c option to -z option to be consistent between all tools.
66-start and 66-stop exit 0 instead of 110 if the service is not enabled.
Frontend service file change:
- The field @type accepts a new kind of service called module. A module can be considered as an instantiated service. It works the same way as a service frontend file but allows to configure a set of different kind of services before executing the enable process. Also, the set of the services can be configured with the conjunction of a script called configure which can be written on any language.
- module also comes with a new section named [regex] which contains the following field:
  - @configure
  - @directories
  - @files
  - @infiles
  Please see the frontend documentation for futher information.
- Allow commenting out of a service inside @contents, @depends, @optdepends, @extdepends field with the # character. The service name must begin with a # character without any space between the # and the name of the service, so it can be ignored.
```
@depends = ( foo #bar fooB )
```
- Add @version field: This field is currently not mandatory to allow time to adapt the existing service files on your system, but it will be mandatory in a future release.
- Comments in the [environment] section is now kept at parsing process and written to the final service configuration file. This is useful to explain the use of a variable without the need to look on the executable script.
66-in{service,tree} display now up,down or unintialized on status and graph dependencies field for oneshot, bundle, module services. This allows us to know if e.g an oneshot service is currently up or down.
66-enable now accepts the new option -m. This option reacts the same as the -c option for new key=value pair but overwrites the change of the admin on existing key=value pair. A key=value pair which doesn't exist on the frontend file remains untouched.
Oneshot now accepts the option log at the field @options. This allows a oneshot to have its own logger. As any other service the log file can be seen with the 66-inservice tool. The log destination can be controlled by the @destination field at the [logger] section. Default is set at compile time by the --with-system-log or --with-user-log flag.
New debug tools:
- 66-inresolve: This tool allows to read the contents of the resolve file.
- 66-instate: This tool allows to read the contents of the state file.
--disable-s6-log-notification was added on the configure script to disable logger's readiness notification. By default it use the file descriptor number 3.
--with-s6-log-timestamp flag was added on the configure script to set the default output date format for a logger at the compile time. See configure --help command.
--with-system-module, --with-sysadmin-module, --with-user-module flags was added on the configure script to set the default system, sysadmin and user module directory respectively.
--with-system-script, --with-user-script flags was added on the configure script to set the default system and user script directory respectively.

In 0.2.5.2¶

Fix build: Remove bytes.h oblibs header file

In 0.2.5.1¶

Bugs fix
66-tree -S options: if after_tree and tree have the same name, tree is considered as the very first tree to start.
66-unmountall: do not umount SS_LIVE
Add SIGPWR control file at creation of .s6-svscan directory

In 0.2.5.0¶

Bugs Fix: Bad memory allocation
66-tree: add -S option which allows to order the start tree process even after it creation.

In 0.2.4.1¶

Hot fix: fix html documentation

In 0.2.4.0¶

Bugs fix: Bad memory allocation.
Enable again RB_DISABLE_CAD but don't crash if its not available and warn user
- Add @optsdepends and @extdepends field at [main] section:
  - @optsdepends can be considere as: "enable one on this service or none"
  - @extdepends can be considere as: "enable the service if it is not already declared on a tree"
- 66-in{tree,service}:
  - in case of empty value the tools return None
  - add -n option: this avoids to display the name of the field.
  - add field at 66-intree:
    - start: displays the list of tree(s) started before
    - allowed: displays a list of allowed user to use the tree
    - symlinks: displays the target of tree's symlinks
  - add field at 66-inservice:
    - optsdepends: displays the service's optional dependencies
    - extdepends: displays the service's external dependencies
- 66-shutdown skeleton: be safier and check if options are past
- New tool:
  - 66-update: The 66-update program makes a complete transition of trees and the live directory using a old 66 format (the one being replaced) with the new 66 format.

In 0.2.3.2¶

Fix bad annoucement at 66-svctl
Warn in case of bad key at parenthesis parsing process

In 0.2.3.1¶

Bugs fix: bad allocation memory
Makefile: remove creation of empty directories. 66-tree will check all directories and create it if it missing.

In 0.2.3.0¶

Adapt to oblibs 0.0.4.0
Remove RB_DISABLE_CAD at 66-boot by default and allow to enable it with -c option. Starting on virtual system like lxc crash with this option. Well be safe on every case.
Use of new function log_? familly from oblibs:
- standardization of the exit code
- standardization of the verbosity output
- allow verbosity to 4 to display debug message
Doc typo fix

In 0.2.2.2¶

Adapt to skalibs 2.9.1.0
Remove insta_? deprecated function

In 0.2.2.1¶

Typo fix at html documentation
Fix 66-intree without options

In 0.2.2.0¶

Bugs fix: Always check the existing of the 66 heart directories
66-info is now deprecated and splitted into two differents API:
- 66-intree give informations about tree
- 66-inservice give informations about services

In 0.2.1.2¶

Hot fix: fix 66-info graph display.

In 0.2.1.1¶

Hot fix: fix the build of the service dependencies graph.

In 0.2.1.0¶

Bugs fix.
A synchronization is now made on reboot even with force option.
The field @name has no effects except for instantiated service but can be omitted. The name of the frontend file is took as name. In case of instantiated service, the field @name must contain the complete name of the frontend service file -- Refer to the frontend documentation page.
New skeleton file: rc.shutdown.final. This skeleton file will be run at the very end of the shutdown procedure, after all processes have been killed and all filesystems have been unmounted, just before the system is rebooted or the power turned off. This script normally remains empty.
Extra tools has removed and provided as an independent programs at ttps://framagit.org/Obarun/66-tools.git except for 66-echo and 66-unmountall program which are a dependent part of 66.
The account to run the s6-log program at the associate service logger can be set at compilation time by the --with-s6-log-user=USER options. The default is root. Obviously, the @runas field at the [logger] section overwritte it. This options set the account to run the uncaught-logs too. Also, this option can be overwritten with the '-l' option at 66-boot invocation and the '-L' option at 66-scandir invocation.

In 0.2.0.4

Bugs fix on memory allocation
Respect policies decision of user:
- Remove -m option from 66-boot on init file
Add -d feature to 66-scandir

In 0.2.0.3¶

66-parser: Fix write of configuration file, add -c|C features

In 0.2.0.2¶

Fix the written of user configuration file
Fix oneshot and bundle status check
Add -l features to 66-tree, rewrite tree_unsupervise function

In 0.2.0.0¶

New tools:
- 66-boot.
- 66-scanctl.
- 66-shutdown.
- 66-shutdownd.
- 66-hpr.
- 66-env.
- 66-parser.
- 66-which.
- 66-echo.
- 66-unmountall.
- execl-subuidgid.
New @key field and change on frontend file:
- [Logger] section accept a @depends field, @timestamp accept none as value, readiness notification is a default.
- @hiercopy in [main] allow to copy any file or directory coming from the directory of the service.
66-envfile is now deprecated, use execl-envfile in replacement.
66-enable: add -F, -c, -C features.
66-svctl is now asynchrone to launch services.
Man pages are available.
Environment variables limitation: maximum 100 files by directory, each file cannot contain more than 4096 bytes or 50 variables.
Syntax to unexport variable with execl-envfile change: the exclamation mark '!' need to be placed at begin of value instead of key.
Bugs fix

In 0.1.0.1¶

Bugs fix
Add 66-getenv,66-writenv and 66-gnwenv extra-tools

In 0.1.0.0¶

Logger for nested tree at scandir creation time is disabled.
Pass graph function to new ss_resolve_graph_? function.
Split resolve file and state flags.
Add ss_state_? function.
Really unsupervise rc service, add rc_manage, rc_unsupervise function.
Fix -U option for 66-tree.
Fix 66-info, add -c option.

In 0.0.2.2¶

Hot fix, do not stop an empty db.

In 0.0.2.1

Hot fix, do not append inner bundle with empty word.

In 0.0.2.0¶

Bugs fix.
Add 66-tree -U options to unsupervise a tree from a scandir.
Rewrite 66-info to provide color and more informations.
Empty database cannot be initialized.
Respect /etc/66/sysadmin/service even for user.

In 0.0.1.1¶

Bugs fix at rc_init function.

Changelog for 66¶

In 0.9.2.0¶

Overview¶

What you must do¶

New things to explore¶

Interface changes¶

Behaviour changes¶

What happens by itself¶

Bug fixes¶

For packagers¶

In 0.9.1.2¶

Overview¶

What you must do¶

Bug fixes¶

For packagers¶

In 0.9.1.1¶

Overview¶

What you must do¶

Bug fixes¶

In 0.9.1.0¶

Overview¶

What you must do¶

New things to explore¶

Bug fixes¶

In 0.9.0.0¶

Overview¶

What you must do¶

Your frontend files¶

Keys recentered on their section¶

Keys removed¶

Build is deprecated and ignored — the shebang decides¶

s6log becomes 66log¶

MaxDeath becomes a crash budget, and MaxDeathInterval is new¶

The logger default timestamp is honored again¶

execl-envfile rejects newline and NUL in a value¶

Your boot¶

How 66 boots now¶

TREE is still there and now optional¶

init.conf¶

The rc.* skeleton files are gone¶

Containers¶

The 66-shutdown program is gone¶

Other boot fixes¶

What happens by itself¶

New things to explore¶

The event system¶

Per-session variables now reach your services¶

New commands¶

Long options everywhere¶

66 status tells you more¶

Fields are selectable¶

Options that no longer exist¶

Under the hood¶

Bug fixes¶

For packagers¶

Notes¶

In 0.8.2.2¶

Overview¶

Behavior Changes¶

Bug Fixes¶

In 0.8.2.1¶

Overview¶

Bug Fixes¶

In 0.8.2.0¶

Overview¶

New Features¶

Bug Fixes¶

Enhancements¶

Migration Notes¶

Notes¶

In 0.8.1.1¶

Bug Fixes¶

In 0.8.1.0¶

Overview¶

New Features¶

Bug Fixes¶

Improvements¶

Notes¶

In 0.8.0.2¶

Bug Fixes¶

`Build` is deprecated and ignored — the shebang decides¶

`s6log` becomes `66log`¶

`MaxDeath` becomes a crash budget, and `MaxDeathInterval` is new¶

`execl-envfile` rejects newline and NUL in a value¶

`TREE` is still there and now optional¶

`init.conf`¶

The `rc.*` skeleton files are gone¶

The `66-shutdown` program is gone¶

`66 status` tells you more¶