The end goal is to get rid of `db/transact!` and send outliner ops to
the db worker.
Currently, some property ops are async, set-block-property! will also
need to be async because when setting a non-ref value (e.g. a number
str "2"), we need to query whether a block with the value exists, this
unfortunately, will be an async query, so we're risking turning more
functions to async in the future which makes it hard to reason about
the implementation.
Replace [:block/schema :properties] with ref type property/schema.properties.
This commit also introduces :logseq.class/base so that all the other
classes will be its children.
A :number closed value saved as a string is problematic for a couple
reasons:
* Not storing a number by the user's intended value means a loss of any
numeric functionality the property provides. Basic things like sorting
by the numeric property don't work. This has caused us problems in the
past and we invented hacky workarounds like logseq.query/nlp-date to
compensate for this
* Creates inconsistency in how numbers are saved as non-closed values
are not saved as string. This affects not only the validation layer
but also creates bugs when querying for these property values as the
querying interface is inconsistent
closes LOG-3042. Moved :default-queries to default-config as this allows
for any future fix to automatically update for a user. There's also a
breaking change with this config key from :block/journal? so moving this
to default-config makes the migration easier e.g. just tell users to
remove the config key
Some changes:
1. use :block/closed-value-property instead to record which property
this value entity belongs to.
2. use :block/content instead of [:block/schema :value].
3. :page and :date have been removed from closed-value-property-types
because there could be a page/date that's a closed value for multiple
properties, we have no idea what's the `:block/order` for a specific
property.
Finally no need to worry about parent-left conflicts and broken chain.
With :block/order, we only need to re-compute new orders for siblings
with same order (it can happens if there're bugs in our code, or
updates from rtc), but it doesn't break UI.
Another huge potential benefit after discussing with Zhiyuan is:
Ee might be able to simplify both RTC and undo/redo, currently, we
need to handle each new op for both of them, with recently
refactorings like properties being db attributes, :block/order
is a string instead of a ref, we can handle most property value
conflicts using last-write-wins, and others (e.g. :block/parent,
property with :default type) specifically.
I haven't fixed the issues of using :block/left in RTC and undo/redo,
because we might change both soon.
This allows imported and eventually user classes to have db idents.
Updated schema example graph which now imports classes as
:schema.class/X. Also fix schema properties which were removed from
the debug file and remove unused property uuids
