Public Spec: Callbacks as named, exported functions #51
Conversation
Previously, callbacks were identified by indices in an exported table, which are hard to keep stable across upgrades. By identifying callbacks with names, and exposing them via named function exports, those developers who aim to have stable callbacks across upgrades have a reasonable chance. Since we are touching this anyways, it also changes the size of the `env` to `i64`; this does not affect the system-side implementation greatly, but allows canisters to do interesting tricks, such as never re-using an `env` value, or encoding both version information and a 32-bit counter in it. This PR is an iteration of #183, which did away with “names” for callbacks completely and only used the `env` value for dispatch, using a single entry point, but subsequent discussion declared that as undesirable.
✅ Deploy Preview for ic-interface-spec ready!
To edit notification comments on pull requests, go to your Netlify site settings. |
|
@nomeata authored this proposal in the internal archived public spec repo. The patch didn't apply cleanly; I hope my resolution of conflicts didn't break anything. |
8 bytes ought to be enough for everyone? At this point, I wonder if we should just allow blobs as environments. This allows canisters to put even more information into the callback without having to touch it’s own memory, which can for example be otherwise completely stateless. It’d have to be accounted for, of course, but that seems doable. |
Co-authored-by: John Plevyak <[email protected]>
|
To me, this looks like a breaking change to the canister API. Isn't it? |
|
Closing the PR for now because there is no active work and it is unclear that converting it to markdown is beneficial at this point. The PR can be converted and reopened at any point in time, such as when the engineering teams actually have capacity to work on the implementation. |
Previously, callbacks were identified by indices in an exported table,
which are hard to keep stable across upgrades. By identifying callbacks
with names, and exposing them via named function exports, those
developers who aim to have stable callbacks across upgrades have a
reasonable chance.
Since we are touching this anyways, it also changes the size of the
envtoi64; this does not affect the system-side implementationgreatly, but allows canisters to do interesting tricks, such as never
re-using an
envvalue, or encoding both version information and a32-bit counter in it.
This PR is an iteration of #183, which did away with “names” for
callbacks completely and only used the
envvalue for dispatch, using asingle entry point, but subsequent discussion declared that as
undesirable.