Document contract execution semantics #898
Conversation
ethanfrey
commented
Apr 22, 2021
•
ethanfrey
commented
- SDK/Tendermint Environment
- Basic Execution
- Handling Messages
- Handling Submessages
- Query Semantics
SEMANTICS.md
Outdated
| area in Ethereum. | ||
|
|
||
| The `reply` call may return `Err` itself, in which case it is treated like the | ||
| caller errored, and aborting the transaction. (Unless, of course, the original |
There was a problem hiding this comment.
(Unless, of course, the original...
The sentence in brackets make it hard for me to follow the flow. Let's get the reply flow described first and explain this at the end of the paragraph?
Maybe with an example:
Please keep in mind that submessage execution and reply can happen within the context of another submessage. For example contract-A --submessage--> contract-B --submessage->contract-C.
Then contract-B can revert the state for contract-C and itself by returning an error in the submessage reply handling. contract-A can revert state for itself, contract-B and contract-C with it's submessage reply error result..
SEMANTICS.md
Outdated
| just as if the original `execute` and returned `data: Some(b"better idea")`. If | ||
| `reply` returns `data: None`, it will not modify any previously set data state. | ||
| If there are multiple submessages all setting this, only the last one is used | ||
| (they all overwrite any previous `data` value). |
There was a problem hiding this comment.
Some details on multiple submessages would be good:
A contract may return multiple submessages with different ReplyOn settings. For example
a) submessage with reply on success
b) submessage with reply on failure
If we return a,b then we have the following conditions:
| processing a) | processing b) | reply called | may overwrite result from reply |
|---|---|---|---|
| ok | ok | a) | a) |
| err | err | b) | b |
| err | ok | none | none |
| ok | err | a)b) | b) |
Any error returned when handling the replies in the contract, would abort the process and revert the state as if the original execute failed
There was a problem hiding this comment.
Almost. The first submessage to non handle an error, or return an error in reply will be treated like an error via message, thus aborting the transaction.
| processing a) | processing b) | reply called | may overwrite result from reply | note |
|---|---|---|---|---|
| ok | ok | a) | a) | returns success |
| err | err | none | none | returns error (abort parent transaction) |
| err | ok | none | none | returns error (abort parent transaction) |
| ok | err | a)b) | a)b) | if both a) and b) overwrite, only b) will be used |
SEMANTICS.md
Outdated
| just as if the original `execute` and returned `data: Some(b"better idea")`. If | ||
| `reply` returns `data: None`, it will not modify any previously set data state. | ||
| If there are multiple submessages all setting this, only the last one is used | ||
| (they all overwrite any previous `data` value). |
There was a problem hiding this comment.
Almost. The first submessage to non handle an error, or return an error in reply will be treated like an error via message, thus aborting the transaction.
| processing a) | processing b) | reply called | may overwrite result from reply | note |
|---|---|---|---|---|
| ok | ok | a) | a) | returns success |
| err | err | none | none | returns error (abort parent transaction) |
| err | ok | none | none | returns error (abort parent transaction) |
| ok | err | a)b) | a)b) | if both a) and b) overwrite, only b) will be used |
SEMANTICS.md
Outdated
| In the Cosmos SDK, a transaction returns a number of events to the user, along | ||
| with an optional data "result". This result is hashed into the next block hash | ||
| to be provable and can return some essential state (although in general client | ||
| apps rely on Events more). This result is more commonly used to pass results | ||
| between contracts or modules in the sdk. |
There was a problem hiding this comment.
Are events hashed into the app state as well?
There was a problem hiding this comment.
No they are not. There was a long discussion about that a few months ago.
Just Data and Code (0 = success) are hashed. They are stored in ResultHash (not AppHash, which is only the state the app writes, not other info that goes to tendermint). I can explain this a bit more.
There was a problem hiding this comment.
Alright, thanks. I don't think we need to go into details here but worth making that fact explicit.
It's all wasmd and @alpe is on that. We needed to define what was expected behavior first. I will address the comments and prepare to merge |
ethanfrey
force-pushed
the
document-execution-semantics
branch
from
d4059d6
to
59b6e8c
Compare
|
Okay, this is complete as far as I am concerned. In integrated your feedback and added a small section about queries. Please review and suggest any last changes to approve it. I will rebase the IBC doc on this and refer to these new techniques. |
|
Nice document. Thanks for writing it. |
