summaryrefslogtreecommitdiff
path: root/src/Propellor/Types
AgeCommit message (Collapse)Author
2019-11-13introduce propsInfoSean Whitton
Code from Joey by private e-mail.
2019-11-11Changed the ChrootBootstrapper type class's buildchroot method to take a ↵Joey Hess
Info parameter, instead of Maybe System. The System can be extracted from the Info; this also allows the chroot's Info to be introspected for eg, the apt mirror. (API change)
2019-07-16update from Stable "stretch" to Stable "buster" in some examplesJoey Hess
2019-07-02use ConstraintKindsJoey Hess
This is just a bit prettier code than manually needing to use constraint ~ True
2019-07-02reword to be more like ghc wordingJoey Hess
2019-07-02better explain Property <unknown>Joey Hess
I had thought it could happen when passing the wrong type to a Property constructor, but happily that is not the case. For example & osDebian Unstable False Still produces a good error: • Couldn't match expected type ‘Architecture’ with actual type ‘Bool’ So Property <unknown> only happens when too many or too few parameters are passed, eg: & osDebian Unstable X86_64 False & osDebian Unstable Before commit 14f6ae30809d8bbdb10b91cc59757e865a365df8, the former of those resulted in: • Couldn't match expected type ‘[Char] -> p0’ with actual type ‘Property (HasInfo + Debian)’ • The function ‘osDebian’ is applied to three arguments, but its type ‘DebianSuite -> Architecture -> Property (HasInfo + Debian)’ has only two Unfortunately, it's now just Property <unknown>. This is the only reversion I've identified from commit 14f6ae30809d8bbdb10b91cc59757e865a365df8. My ghc bug https://gitlab.haskell.org/ghc/ghc/issues/16894 isn't quite explaining this problem yet, and I need to find a minimal case of it. As to the latter with too few parameters, it used to result in a massive ugly useless error. Now it is also Property <unknown>, which is an improvement.
2019-07-02use DelayErrorJoey Hess
Syrak looked at this branch and said: Cool! I'd suggest that if it's working, that's an accident! You probably want IfStuck e (DelayError err1) (DelayErrorFcf err2) rather than IfStuck e (TypeError err1) (TypeError err2)
2019-07-02Revert "Revert "custom type error messages""Joey Hess
This reverts commit 665ea0d3d9e1b0e90278fd659dee0ef8642030da.
2019-07-02Revert "custom type error messages"Joey Hess
This reverts commits 14f6ae30809d8bbdb10b91cc59757e865a365df8 de21ef26861db458b0dfb0212cf501f9f8ed459b e20662e6a8881db55394a6366be17ca4e509bc2a Until this bug is resolved, these custom error types hide more basic errors. https://gitlab.haskell.org/ghc/ghc/issues/16894
2019-07-01optionally use type-errors to detect stucknessJoey Hess
Use the type-errors library to detect when the type checker gets stuck unable to reduce type-level operations on MetaTypes, and avoid displaying massive error messages in such a case. But, since type-errors is a new library not available in eg Debian yet, added a WithTypeErrors build flag. When the library is not available, cabal will automatically disable that build flag, and it will build without the type-errors library. This is most often used when combining properties of different types. If the MetaTypes don't have an OS in common, the error message used to be "Property " followed by pages of MetaTypes operations. Now it looks like this: • Cannot combine Properties: Property <unknown> Property HasInfo + Debian + Buntish + ArchLinux + FreeBSD (Property <unknown> is often due to a partially applied Property constructor, or due to passing the wrong type to a Property constructor.) Also it's used in ensureProperty to detect a case where the outer MetaTypes need to be inferred in order to check if the inner MetaTypes match, but the type checker is unable to infer it: • ensureProperty outer Property type is not able to be inferred here. Consider adding a type annotation. • When checking the inferred type writeConfig :: forall (outer :: [Propellor.Types.MetaTypes.MetaType]) t. And it's used in tightenTargets to detect when ghc is unable to infer the desired type of Property: • Unable to infer desired Property type in this use of tightenTargets. Consider adding a type annotation. • When checking the inferred type mk :: forall (tightened :: [Propellor.Types.MetaTypes.MetaType]).
2019-07-01custom type error messagesJoey Hess
* Avoid displaying an excessive amount of type error messages when many properties have been combined in a props list. * Added custom type error messages when Properties don't combine due to conflicting metatypes. * Added custom type error messages for ensureProperty and tightenTargets. * ensureProperty: The constraints have been simplified to EnsurePropertyAllowed. (API change) * ensureProperty: The contraints have been simplified to TightenTargetsAllowed. (API change) * CheckCombinable generates a Bool. (API change) This commit was sponsored by Jake Vosloo on Patreon.
2019-04-08change from inChroot to granular container capabilities infoJoey Hess
* Removed inChroot, instead use hasContainerCapability FilesystemContained. (API change) * Hostname: Properties that used to not do anything in a systemd or docker container will now change the container's hostname, since it's namespaced. More container capabilities can easily be added later, to fine grain control what properties will run in different kinds of containers. This changed CmdLine's Read instance, which should be ok, because propellor inside the container is always updated at the same time as propellor outside, so when it chains into the chroot, it will know to expect the capability list. Docker was not setting InChroot before, but now sets both container capabilities, so chroot setting will also work in it. Note that, things that used to check inChroot before would not work in docker, but things that check FilesystemContained now will. It may be that some of those properties don't really work properly in docker. And, Grub.installed used to run grub-mkconfig in a docker container before, I doubt that made sense (was it even safe?); it doesn't do it now. This commit was sponsored by Trenton Cronholm on Patreon.
2018-12-30stop importing deprecated Network.BSDJoey Hess
Luckily only the HostName type is needed, which Network.Socket still exports
2018-10-24no longer need UndecidableInstances hereJoey Hess
2018-10-13name bikesheddingJoey Hess
2018-10-08Add a DirectBoot bootloader type.David Bremner
This doesn't actually do anything to the chroot in question, just marks it as needing to be boot directly, e.g. by some external kernel and initrd.
2018-05-18modernized and simplified the MetaTypes implementationJoey Hess
now that compatability with ghc 7 is no longer needed. Data.Type.Bool contains effectively the same stuff that was implemented here, so removed my code. Tried to use Data.Type.Equality instead of my EqT, but it seems to be some other type of (type level) equality, and didn't compile. Instead went with the simpler EqT implementation that newer ghc versions allow. The rest of the changes are simply better syntax for defining type families. And upon using that syntax, ghc noticed that `type family a + b` does not have kind "ab" like I wrote before, but is kind *. Tested on debian stable with ghc 8.0.1. This commit was sponsored by John Pellman on Patreon.
2018-04-30fix broken SemigroupMonoid transition <<loop>>Joey Hess
Turns out that with ghc 8.2.2, the instructions given on the page don't work. And the cppless variant that I had compiles, but into effectively mappend = mappend so it loops. The only way I can see to make it work without cpp is to use mappend = (Sem.<>) which is ugly and a land mine waiting to explode if someone changes it to a nicer mappend = (<>) with a newer version of ghc which will compile it and work ok, while breaking it with 8.2.2. Sigh. I posted to haskell-cafe about this.
2018-04-23more ghc 8.4 build fixesJoey Hess
Tested build with ghc 8.4 now.
2018-04-23update note about singletons libraryJoey Hess
ghc version is ok to use it now. However, seems debian has not packaged it, so it won't be in stable for years, so I can't even think about using it for propellor. Proably ghc will get built-in singletons support before I can drop the current singletons code.
2018-04-23semigroup monoid change fallout; drop ghc 7 supportJoey Hess
Fix build with ghc 8.4, which broke due to the Semigroup Monoid change. See https://prime.haskell.org/wiki/Libraries/Proposals/SemigroupMonoid Dropped support for building propellor with ghc 7 (as in debian oldstable), to avoid needing to depend on the semigroups transitional package, but also because it's just too old to be worth supporting. If we indeed drop ghc 7 support entirely, some code to support "jessie" can be removed; concurrent-output can be de-embedded, and the Singletons code can be simplified. This commit was sponsored by Jack Hill on Patreon.
2017-12-21Grub.boots, Grub.bootsMounted: Pass --target to grub-install.Joey Hess
This is to support eg, coreboot. The GrubTarget passed to Grub.installed is introspected to determine --target. If multiple grubs are installed, it currently doesn't pass any --target. Might make more sense to run grub-install repeatedly, but I don't know if that case is sane at all. The Xen -> "x86_64-xen" mapping is kind of arbitrarily chosen since there's a i386-xen available too. I don't know when that case would be used in any case though; chainPVGrub uses installed Xen, but it does not run grub-install. If this does become a problem, would probably need to split it into Xen64 and Xen32. Renamed BIOS to GrubTarget in passing to match grub's terminology; BIOS was kind of a joke term for this in propellor. This commit was sponsored by Francois Marier on Patreon.
2017-11-19partition table in InfoJoey Hess
Diskimage.imageBuiltFor: New property to build a disk image for a Host, using partition table information configured via the new properties hasPartitionTableType, hasPartition and adjustPartition. This lets Machine properties include eg /boot partitions that are known to be needed by the bootloader, and the user can adjust those partitions and add others. This commit was sponsored by Brock Spratlen on Patreon.
2017-11-18reorganizedJoey Hess
2017-11-16Uboot: New module.Joey Hess
Installing u-boot to the boot sector is not needed by some boards (my CubieTruck boots without it), but may be by others. Tricky part was making u-boot be written to a disk image when building one. This commit was sponsored by Jake Vosloo on Patreon.
2017-11-16flash-kernel supportJoey Hess
Can be used to create disk images for arm boards using flash-kernel. This commit was sponsored by Ewen McNeill.
2017-10-25Make addInfo accumulate Info in order properties appear, not reverse orderJoey Hess
This fixes a bug involving reverting Systemd.resolvConfed or Systemd.linkJournal. addInfo was prepending to the list for efficiency. But, that was in conflict with mappend of two Info, which appended the second to the first. In the case where Systemd.resolvConfed was added reverted, to override the one added by default, that led to a list of info that had first the reversion and then the default, so the default won. Which was wrong. So, make addInfo accumulate in the same order mappend combines things, even though it's a little less efficient. The efficiency probably does not matter; there is not typically a whole lot of info. There's some risk this change has unexpected consequences, if something relied on the old addInfo order without using fromInfo to access to info. But if so, that something would have been broken before when two Info properties were combined. With this change, it would just be broken the other way around. This commit was sponsored by Jochen Bartl on Patreon.
2017-10-04avoid propagating non-alias DNS info from container to hostJoey Hess
* When the ipv4 and ipv6 properties are used with a container, avoid propagating the address out to the host. * DnsInfo has been replaced with DnsInfoPropagated and DnsInfoUnpropagated. (API change) * Code that used fromDnsInfo . fromInfo changes to use getDnsInfo. * addDNS takes an additional Bool parameter to control whether the DNS info should propagate out of containers. (API change) This commit was sponsored by Trenton Cronholm on Patreon.
2017-10-04improve commentJoey Hess
2017-07-29Added PROPELLOR_TRACE environment variableJoey Hess
which can be set to 1 to make propellor output serialized Propellor.Message.Trace values, for consumption by another program. This commit was sponsored by Ewen McNeill.
2017-07-27Propellor.Types.Info: add missing import for 'cast'Sergei Trofimovich
The change fixes the following build failure on ghc-8.2.1: [ 6 of 160] Compiling Propellor.Types.Info ( src/Propellor/Types/Info.hs, dist/dist-sandbox-2ceaa3c0/build/Propellor/Types/Info.o ) src/Propellor/Types/Info.hs:38:34: error: • Variable not in scope: cast :: v1 -> Maybe v • Perhaps you meant one of these: data constructor ‘Last’ (imported from Data.Monoid), ‘last’ (imported from Prelude) | 38 | extractInfoEntry (InfoEntry v) = cast v | ^^^^ Signed-off-by: Sergei Trofimovich <slyfox@gentoo.org>
2017-07-26add DeriveDataTypeableJoey Hess
2017-07-21calcPartTable using PartSpec DSLJoey Hess
* Generalized the PartSpec DSL, so it can be used for both disk image partitioning, and disk device partitioning, with different partition sizing methods as appropriate for the different uses. (minor API change) * Propellor.Property.Parted: Added calcPartTable function which uses PartSpec DiskPart, and a useDiskSpace combinator. This commit was sponsored by Thomas Hochstein on Patreon.
2017-07-05DiskImage: Removed grubBootedJoey Hess
Properties that used to need it as a parameter now look at Info about the bootloader that is installed in the chroot that the disk image is created from. (API change) This is a simplication, and avoids the user needing to repeat themselves in the propellor config, thus avoiding mistakes. When no boot loader is installed, or multiple different ones are, disk image creation will fail, which seems reasonable. This commit was sponsored by Jake Vosloo on Patreon.
2017-06-17use stretch in debian stable examplesJoey Hess
2017-05-15Removed dependency on MissingH, instead depends on split and hashable.Joey Hess
MissingH is a heavy dependency, which pulls in parsec and a bunch of stuff. So eliminating it makes propellor easier to install and less likely to fail to build. changesFileContent now uses hashable's hash. This may not be stable across upgrades, I'm not sure -- but it's surely ok here, as the hash is not stored. socketFile also uses hash. I *think* this is ok, even if it's not stable. If it's not stable, an upgrade might make propellor hash a hostname to a different number, but with 9 digets of number in use, the chances of a collision are small. In any case, I've opned a bug report asking for the stability to be documented, and I think it's intended to be stable, only the documentation is bad. NB: I have not checked that the arch linux and freebsd packages for the new deps, that Propellor.Bootstrap lists, are the right names or even exist. Since propellor depends on hashable, it could be changed to use unordered-containers, rather than containers, which would be faster and perhaps less deps too. This commit was sponsored by Alexander Thompson on Patreon.
2017-03-15Added Monoid instances for Property and RevertableProperty.Joey Hess
* Added Monoid instances for Property and RevertableProperty. * Removed applyToList. Instead, use mconcat. (API change) Eg, if you had: applyToList accountFor [User "joey", User "root"] use instead: mconcat (map accountFor [User "joey", User "root"]) mappend x y is basically the same as x `before` y. In particular, if x fails to be ensured, it won't ensure y. This seems to make sense, since applyToList had that behavior, and so does the Monoid for Propellor Result. The alternative would be to try to ensure both and combine the results. However, I don't see any reason to do it that way. It would be nice if the description of both properties were displayed when ensuring the combination. But then, it would need to display eg: ensuring x..ok ensuring y..failed ensuring x and ensuring y..failed Without a way to get rid of that redundant last line, I don't want to do that. Note that the haddocks for the Monoid instances need a really wide screen to display! This is IMHO an infelicity in haddock, and I can't do anything about it really. This commit was sponsored by Fernando Jimenez on Patreon.
2017-03-15Property types changed to use a Maybe (Propellor Result). (API change)Joey Hess
* Property types changed to use a Maybe (Propellor Result). (API change) * When Nothing needs to be done to ensure a property, propellor will avoid displaying its description at all. The doNothing property is an example of such a property. This is mostly in preparation for Monoid instances for Property types, but is's also nice that anything that uses doNothing will avoid printing out any message at all. At least, I think it probably is. It might potentially be confusing for something that sometimes takes an action and sometimes resolves to doNothing and in either case has a description set to not always show the description. If this did turn out to be confusing, the change to doNothing could be reverted. This commit was sponsored by Boyd Stephen Smith Jr. on Patreon.
2017-03-11don't propagate DNS info from DiskImage chrootsJoey Hess
* DiskImage building properties used to propagate DNS info out from the chroot used to build the disk image to the Host. That is no longer done, since that chroot only exists as a side effect of the disk image creation and servers will not be running in it. * The IsInfo types class's propagateInfo function changed to use a PropagateInfo data type. (API change) This is particularly important when using hostChroot, since the host could well have DNS settings then. This commit was sponsored by Ole-Morten Duesund on Patreon.
2017-02-26noteJoey Hess
2017-02-26convert fromIPAddr to valJoey Hess
2017-02-26use val instead of showJoey Hess
2017-02-26ConfigurableValue instances for User, GroupJoey Hess
2017-02-26fix tabs for indentationJoey Hess
2017-02-26use ConfigurableValue where applicableJoey Hess
* Removed fromPort (use val instead). (API change) * Removed several Show instances that were only used for generating configuration, replacing with ConfigurableValue instances. (API change) It's somewhat annoying that IsInfo requires a Show instance. That's needed to be able to display Info in ghci, but some non-derived Show instances had to be kept to support that.
2017-02-26Added ConfigurableValue type classJoey Hess
* Added ConfigurableValue type class, for values that can be used in a config file, or to otherwise configure a program. * The val function converts such values to String. This was motivated by the bug caused by type Port = Int changing to newtype Port = Port Int deriving Show After that change, some things that used show port to generate config files were broken. By using the ConfigurableValue type class instead, such breakage can be prevented.
2017-02-04improve layoutJoey Hess
2017-02-03add types for Arch LinuxZihao Wang
Signed-off-by: Zihao Wang <dev@wzhd.org>
2017-01-18add Ord instance for PortFélix Sipma
2016-12-26Added --build option, which makes propellor simply build itself.Joey Hess