Today I released a new version of warmy, the
warmy-0.6.0 release. That release kicks in with a
few additions, among:
That bug caused long-lasting reloads on stream-copied resources go into weird behavior.
In order to get the bug, I must give a bit of context.
Imagine you have a large resource, like a 4K texture or a big mesh. Whenever your favourite
software writes it to disk, it’s very likely it’ll stream chunks by chunks of bytes. For
instance, it might choose to copy the resource 2 MB by 2 MB on disk. On each copy, your
file system will generate
WRITE events, that warmy will intercept. Before
default behavior was to reload the resource on the first
WRITE event, which was already
wrong, because only a very small part of the resource would have changed – I actually witnessed
weird behaviors with textures in a demo of mine I’m working on, not seeing the new textures in my
But there’s worse. There’s a parameter you can set of your
That parameter gives warmy a hint about how much time must have passed since the last update in
order to effectively call the
Load::reload function. However, this is a bit twisted, because if
the resource takes more time to reload than
update_await_time_ms, it’ll get repeatedly loaded –
for as many as
WRITE events were generated. This is a bit sick, yeah.
The fix was pretty simple: change the semantics of that
0.5.2, it has
the default value of 1s, meaning that a resource wouldn’t reload if it was reloaded less than a
second ago. The new semantics works on the future. Whenever a
WRITE event is intercepted, warmy
will call the
Load::reload function only if no
WRITE event is intercepted in the next
update_await_time_ms. It’s a bit like the implementation of a click and a double click: you must
wait a bit after you got a
MouseRelease event in order to interpret is as a
MouseRelease could arrive soon (generating a
DoubleClick if it’s soon enough).
You’ll also notice that
update_await_time_msname sticks better to the new semantics!
0.5.2, the default value for
update_await_time_ms was 1s. If we kept that value, it would
result in a pretty bad overall latency. The value was lowered to 50ms instead.
You can still tweak that value if it doesn’t suit your needs.
More information can be found in the changelog.
I’ve been very happy with what warmy has brought to me so far. Other people also gave it a try
and for now seem to enjoy it. I’ve gathered a few ideas for the future, based on IRL talks over
a beer several beers, and GitHub issues / pull requests:
Feel free to test it, and as always, keep the vibes!