diff options
Diffstat (limited to 'doc/todo')
12 files changed, 209 insertions, 20 deletions
diff --git a/doc/todo/editor_for_privdata__63__.mdwn b/doc/todo/editor_for_privdata__63__.mdwn new file mode 100644 index 00000000..8b91338c --- /dev/null +++ b/doc/todo/editor_for_privdata__63__.mdwn @@ -0,0 +1,4 @@ +Would adding a way to call $EDITOR to edit privdata be possible? +It would make sense for editing data like logcheck files. + +> [[done]] diff --git a/doc/todo/editor_for_privdata__63__/comment_2_4fcbdf36f32ca7cf82593a8992167aff._comment b/doc/todo/editor_for_privdata__63__/comment_2_4fcbdf36f32ca7cf82593a8992167aff._comment new file mode 100644 index 00000000..bbe93fe3 --- /dev/null +++ b/doc/todo/editor_for_privdata__63__/comment_2_4fcbdf36f32ca7cf82593a8992167aff._comment @@ -0,0 +1,9 @@ +[[!comment format=mdwn + username="http://joeyh.name/" + subject="comment 2" + date="2014-11-11T21:16:09Z" + content=""" +Already exists in `propellor --edit` + +Documentation patches accepted! :) +"""]] diff --git a/doc/todo/etckeeper.mdwn b/doc/todo/etckeeper.mdwn new file mode 100644 index 00000000..7dc80cef --- /dev/null +++ b/doc/todo/etckeeper.mdwn @@ -0,0 +1 @@ +It would be cool to have an etckeeper module :-). diff --git a/doc/todo/etckeeper/comment_1_8766da27c69bbae357d497e0e557fad2._comment b/doc/todo/etckeeper/comment_1_8766da27c69bbae357d497e0e557fad2._comment new file mode 100644 index 00000000..f080f70e --- /dev/null +++ b/doc/todo/etckeeper/comment_1_8766da27c69bbae357d497e0e557fad2._comment @@ -0,0 +1,9 @@ +[[!comment format=mdwn + username="joey" + subject="""comment 1""" + date="2014-11-06T15:46:56Z" + content=""" +All I've needed for this is `& Apt.installed ["etckeeper"]` + +Patches welcome, I suppose. +"""]] diff --git a/doc/todo/fail_if_modification_not_commited_when_using_--spin.mdwn b/doc/todo/fail_if_modification_not_commited_when_using_--spin.mdwn new file mode 100644 index 00000000..046f4a6f --- /dev/null +++ b/doc/todo/fail_if_modification_not_commited_when_using_--spin.mdwn @@ -0,0 +1,3 @@ +Sometimes I forget to commit a modification, and running "propellor --spin" automatically commits this stuff. It would be better if "propellor --spin" failed (or, even better, warned the user) that there are uncommited changes, and "propellor --spin" would just always add an empty commit. + +> --merge added; [[done]] --[[Joey]] diff --git a/doc/todo/fail_if_modification_not_commited_when_using_--spin/comment_1_7267d62ccc8db44bccb935836536e8a1._comment b/doc/todo/fail_if_modification_not_commited_when_using_--spin/comment_1_7267d62ccc8db44bccb935836536e8a1._comment new file mode 100644 index 00000000..19b2fab6 --- /dev/null +++ b/doc/todo/fail_if_modification_not_commited_when_using_--spin/comment_1_7267d62ccc8db44bccb935836536e8a1._comment @@ -0,0 +1,30 @@ +[[!comment format=mdwn + username="joey" + subject="""comment 1""" + date="2014-11-23T18:41:40Z" + content=""" +Letting --spin commit is part of my workflow. It's great when you're just +changing config.hs to quickly blast out the changes. + +Granted, it is not so nice when doing Property development, as changes get +fragmented across the spins used to test them. I'd be happy to find some +way to improve that. Perhaps a way could be found to get this structure of +git commits: + + manual commit------------------------->manual commit--merge + \--spin--spin--spin--spin--spin------------/ + +Where the second manual commit has an identical tree committed as does the +spin just underneath it, and so the following merge doesn't change any files, +just grafts the two branches back together. + +I guess that could be handled by haing a checkpoint command, that squashes +all the previous spins since the last checkpoint together into one commit, +lets the user edit the commit message of that, and the juggles the branches +into place and creates the merge commit -- which then becomes the new last +checkpoint. + +I'll take patches for such a thing, or more simply a way to configure --spin's +auto-committing behavior. However, I don't want to change the default +behavior to not commit. +"""]] diff --git a/doc/todo/fail_if_modification_not_commited_when_using_--spin/comment_2_e4d170a14d689bef5d9174b251a4fe6f._comment b/doc/todo/fail_if_modification_not_commited_when_using_--spin/comment_2_e4d170a14d689bef5d9174b251a4fe6f._comment new file mode 100644 index 00000000..3e8e5f62 --- /dev/null +++ b/doc/todo/fail_if_modification_not_commited_when_using_--spin/comment_2_e4d170a14d689bef5d9174b251a4fe6f._comment @@ -0,0 +1,7 @@ +[[!comment format=mdwn + username="gueux" + subject="comment 2" + date="2014-11-23T20:23:24Z" + content=""" +Your solution seems a lot better :-). +"""]] diff --git a/doc/todo/fail_if_modification_not_commited_when_using_--spin/comment_3_c69eaa9c6ae5b07b5c2dd2591de965a3._comment b/doc/todo/fail_if_modification_not_commited_when_using_--spin/comment_3_c69eaa9c6ae5b07b5c2dd2591de965a3._comment new file mode 100644 index 00000000..8ad6ab49 --- /dev/null +++ b/doc/todo/fail_if_modification_not_commited_when_using_--spin/comment_3_c69eaa9c6ae5b07b5c2dd2591de965a3._comment @@ -0,0 +1,19 @@ +[[!comment format=mdwn + username="joey" + subject="""comment 3""" + date="2014-11-23T21:12:19Z" + content=""" +Here's a almost-script to do it, which worked when it did it by hand: + +<pre> +get old-head (git show-ref HEAD -s) +get curr-branch (refs/heads/master eg) +find old-commit (look back through git log for the first commit that was not "propellor spin") +git reset old-commit +git commit -a # user gets to edit commit message for all the spins and any staged changes here +git merge -S -s ours old-head +get current-commit (result of merge) +git update-ref curr-branch current-commit +git checkout curr-branch +</pre> +"""]] diff --git a/doc/todo/git_push_over_propellor_ssh_channel.mdwn b/doc/todo/git_push_over_propellor_ssh_channel.mdwn new file mode 100644 index 00000000..c6d42fcf --- /dev/null +++ b/doc/todo/git_push_over_propellor_ssh_channel.mdwn @@ -0,0 +1,13 @@ +Propellor currently needs a central git server. And it has a special-cased +protocol during bootstrap that transfers the git repo over to a new host, +using the ssh connection that will be used to run propellor. + +This could be improved by making a git push be done whenever +`propellor spin $host` runs. The remote propellor runs `git receive-pack`; +the local one runs `git send-pack`. + +Then there would be no need for a central git repo. Although still very +useful if you have multiple propellor driven hosts and you want to just git +commit and let cron sort them out. + +> [[done]]! --[[Joey]] diff --git a/doc/todo/info_propigation_out_of_nested_properties.mdwn b/doc/todo/info_propigation_out_of_nested_properties.mdwn index 1a586be6..b08e326f 100644 --- a/doc/todo/info_propigation_out_of_nested_properties.mdwn +++ b/doc/todo/info_propigation_out_of_nested_properties.mdwn @@ -1,7 +1,7 @@ Currently, Info about a Host's Properties is manually gathered and -propigated. propertyList combines the Info of the Properties in the list. +propigated. propertyList combines the Info of the Properties in the list; Docker.docked extracts relevant Info from the Properties of the container -(but not al of it, intentionally!). +(but not al of it, intentionally!); etc. This works, but it's error-prone. Consider this example: @@ -9,26 +9,91 @@ This works, but it's error-prone. Consider this example: (Just (System (Debian Unstable) _)) -> ensureProperty foo _ -> ensureProperty bar -Here, the Info of `foo` is not propigated out. Nor is `bar`'s Info. -Of course, only one of them will be run, and only its info should be propigated -out.. +Here, the Info of `foo` is not propigated out. Nor is `bar`'s Info. Of +course, only one of them will be run, and only its info should be +propigated out.. -This commonly afflicts eg, privData. For example, `User.hasPassword'` -has this problem, and this prevents --list-fields from listing privdata -that's not set from that property. +---- One approach might be to make the Propellor monad be able to be run in two -modes. In one mode, it actually perform IO, etc. In the other mode, all -liftIO is a no-op, but all Info encountered is accumulated using a Reader -monad. This might need two separate monad definitions. +modes. In run mode, it actually performs IO, etc. In introspection mode, all +liftIO is a no-op, but all Info encountered is accumulated using a Reader. +This might need two separate monad definitions. -That is surely doable, but the withOS example above shows a problem with it -- -the OS is itself part of a Host's info, so won't be known until all its -properties have been examined for info! +That is surely doable, but consider this example: -Perhaps that can be finessed. We don't really need to propigate out OS info. -Just DNS and PrivDataField Info. So info could be collected in 2 passes, -first as it's done now by static propertyInfo values. Then take that -and use it as the Info when running the Properties in the Reader monad. -Combine what the Reader accumulates with the static info to get the full -info. + property "demo" = do + needfoo <- liftIO checkFoo + if needfoo + then ensureProperty foo + else ensureProperty bar + +In introspection mode, the liftIO is a no-op, but needs to return a Bool. +That seems unlikely (how to pick which?), but even if some defaulting is +used, only one of foo or bar's info will be seen. + +---- + +Another approach could be something like this: + + withInfoFrom foo $ \callfoo -> + withInfoFrom bar $ \callbar -> + property "demo" = do + needfoo <- liftIO checkFoo + if needfoo + then callfoo + else callbar + +Here withInfoFrom is able to add foo and bar's Info to the info of the +property that (may) call them. + +This approach is not fully type safe; it would be possible to call +withInfoFrom in a way that didn't let it propigate the info. + +Also it has the problem that it doesn't support this: + + property "demo" = do + needfoo <- liftIO checkFoo + if needfoo + then do + foop <- liftIO getFooParam + ensureProperty (foo foop) + else ensureProperty bar + +---- + +Another approach would be to add a new SimpleProperty, which is a property +that has no Info. Only allow calling ensureProperty on this new type. + +(Or, remove propertyInfo from Property, and add a new InfoProperty that +has the info.) + +But, propertyList can only contain one type at a time, +not a mixed list of Property and SimpleProperty. + +Could a GADT be used instead? + + {-# LANGUAGE GADTs #-} + {-# LANGUAGE EmptyDataDecls #-} + + data HasInfo + data NoInfo + + data Property = IProperty (GProperty HasInfo) | SProperty (GProperty NoInfo) + + data GProperty i where + GIProperty :: Desc -> Propellor Result -> Info -> GProperty HasInfo + GSProperty :: Desc -> Propellor Result -> GProperty NoInfo + + ensureProperty :: GProperty NoInfo -> Propellor Result + ensureProperty (GSProperty d r) = r + +That works. I made a `gadtwip` git branch that elaborated on that, +to the point that Property.File compiles, but is otherwise +unfinished. Most definitions of `Property` need to be changed to +`GProperty NoInfo`, so that ensureProperty can call them. It's a big, +intrusive change, and it may complicate propellor too much. + +(I may need to make instances of Prop for `GProperty NoInfo` and `GProperty +HasInfo`, if that's possible, and make more Property combinators work on +Prop.) diff --git a/doc/todo/port_info_for_properties_for_firewall.mdwn b/doc/todo/port_info_for_properties_for_firewall.mdwn new file mode 100644 index 00000000..efaaba05 --- /dev/null +++ b/doc/todo/port_info_for_properties_for_firewall.mdwn @@ -0,0 +1,24 @@ +The firewall module could be improved if properties that set up a service +on a port included info (see Propellor.Info and Propellor.Types.Info) +about the port(s) used. + +While currently the ports have to be explicitly listed: + + & Apache.installed + & Firewall.installed + & Firewall.addRule (Rule INPUT ACCEPT (Proto TCP :- Port 80)) + & Firewall.addRule (Rule INPUT ACCEPT (Proto TCP :- Port 443)) + +Instead the ports would be derived from the installed services. + + & Apache.installed + & Firewall.installed + +There could also be some combinators to adjust the exposed +ports of a property. + + & localOnly Apache.installed + & exposedPorts [443,80] (Apt.serviceInstalledRunning "apache2") + +Such port enformation is also going to be needed as a basis of +[[type_level_port_conflict_detection]]. --[[Joey]] diff --git a/doc/todo/type_level_port_conflict_detection.mdwn b/doc/todo/type_level_port_conflict_detection.mdwn new file mode 100644 index 00000000..5aec5775 --- /dev/null +++ b/doc/todo/type_level_port_conflict_detection.mdwn @@ -0,0 +1,5 @@ +See <http://stackoverflow.com/questions/26027765/using-types-to-prevent-conflicting-port-numbers-in-a-list> --[[Joey]] + +Needs ghc newer than 7.6.3. It may be possible to port Data.Type.Equality +and Data.Type.Bool to older versions; I got them to compile but they didn't +work right. --[[Joey]] |