One of my pending work items for post-realm merge.
The lowest-level import task is no longer asynchronous, as we don't want
it to span multiple threads to allow easier interaction with realm.
Removing the `Task` spec simplifies a heap of usages.
Individual usages should decide whether they want to run the import
asynchronously, by either using an alternative override or spooling up a
thread themselves.
Without doing this, the JSON content is not written to the file. A user
assumption is that as soon as a skin shows up in the skin list as
exportable, it should export correctly, so it makes sense that it should
be in a sane state even if the user has not made any changes in the skin
editor yet.
Going forward, we might move more of the json serialisation logic out,
and run for consistency as part of the import process. This seems like
the simplest way to guarantee things for now, though.
As this has regressed twice now, let's play it safe and bail rather than
stack overflowing. Note that as all the underlying issues that could
trigger this have been fixed, no additional tests have been added. To
test, comment out `SkinManager.cs` line 228-229 to cause a failure. The
new logic will kick in and show a log output message, but all tests will
still (correctly) pass.
Apart from being cleaner (and allowing removal at the point of export,
if that is ever requires), this also gives us a better guarantee that
the metadata is going to take effect.
By writing the `[General]` section ourselves, we avoid any issues that
may be present from a user constructing the file (incorrectly)
themselves, such as multiple `Title:` entries or a malformed `[General]`
marker.
The main goal is to ensure that the name is updated to what we expect it
to be.
