More API updates for the Allegra and Mary eras #2111
Conversation
Will be needed for constructing txs in later ledger eras.
Move them into the TxMetadata module and generalise toShelleyMetadataHash to work over all ledger eras. Will be needed for creating txs in later ledger eras.
Will be needed for making txs for later ledger eras.
Not yet generalised over ledger eras.
We would like the ProtocolParametersUpdate type to be able to be converted both to and from the equivalent underlying ledger type. We would like this for a couple reasons: 1. so that we can use this same type later for programs that analyse the blockchain and want to see the content of update proposals; 2. so we can switch UpdateProposal to use the representation of the surface API types, while still using the on-chain format for the binary serialisation. This requires converting in both directions. We want to do 2. now because we want a single era-independent representation for update proposals, and to convert into the era-dependent ledger types on-demand when we make a transaction.
So we now have conversions in both directions.
Switch the UpdateProposal to use the representation of the surface API types. We stick to the existing on-chain binary serialisation format. We want to do this now to add support for more ledger eras. We want a single era-independent representation for update proposals, and to convert into the era-dependent ledger types on-demand when we make a transaction for a specific era.
The protocol params should always have been exported since it is a API surface type. The UpdateProposal repreentation can now be public since it now uses the API surface types too.
Use Text instead of URI. We will want to allow conversion in both directions which means we cannot be more restrictive than the ledger's validity rules.
We'll use this next to change the representation of certificates
We will switch the representation shortly and so these conversion functions will become non-trivial.
This will make it easier to convert from a common era-independent type into ledger era-dependent types. We still use the same on-chain binary format.
Needed both directly and to be able to generalise other such conversion functions.
|
For some functions the Allegra and Mary cases are still TODOs but the types are now more general. This should allow CLI code to type check at the new eras. We'll fill in the remaining TODO cases. |
| toCBOR = toCBOR . toShelleyUpdate | ||
| toCBOR = toCBOR . toShelleyUpdate @StandardShelley | ||
| -- We have to pick a monomorphic era type for the serialisation. We use the | ||
| -- Shelley era. This makes no difference since era type is phantom. |
There was a problem hiding this comment.
Can we define a type synonym to make this clear. Something like:
type ShelleyOnwardsCrypto = StandardShelley or something with a better name 😁
Replace per-era instances with a single era-paramaterised instance. This will be simpler as we add more eras.
Replace per-era instances with a single era-paramaterised instance. This will be simpler as we add more eras.
Generalise over the IsShelleyBasedEra class. The Allegra and Mary cases remain to do.
Switch from the Shelley.TxBody concrete type, to using the Ledger.TxBody type family. The type family has different representations for different eras. For this step the specific era remains the same, so there are no other changes. The next step will be to cover multiple eras, at which point we will take advantage of the different tx body representations for each era.
The 'ShelleyBasedEra' GADT tells us what era we are in. The 'ShelleyLedgerEra' type family maps that to the era type from the ledger lib. The 'Ledger.TxBody' type family maps that to a specific tx body type, which is different for each Shelley-based era. Due to this use of GADTs We now need custom Eq and Show instances. Do the minimal changes elsewhere, inserting error cases to fill in next.
This is partial in that we generalise the type, but the Allegra and Mary cases are error TODOs for now. This is still helpful to allow downstream code to type check at other eras.
dcoutts
force-pushed
the
dcoutts/api-allegra-mary
branch
from
e20486f
to
134403f
Compare
All the Shelley-based eras are supported with one overloaded implementation used at different types.
Still TODOs for validity intervals and minting.
We sort-of already had implementations of these in the impl of to/fromShelleyAddress functions. Pull them out to top level functions and put them next to the definition of the ScriptHash type. We'll reuse these new functions in the next patch.
To/from the equivalent ledger library type. The ledger representation is a pair of the lovelace value and a nested map of the non-ada assets.
Now that we have toMaryValue
Order the ada case first.
| -- The era parameter in these types is a phantom type so it is safe to cast. | ||
| -- We choose to cast because we need to use an era-independent address | ||
| -- representation, but have to produce an era-dependent format used by the | ||
| -- Shelley ledger lib. |
There was a problem hiding this comment.
Nice 👍.
I'm slightly concerned that someone reading through the types will be confused, but doc like this should help.
| stakePoolMetadataURL :: URI.URI, | ||
| stakePoolMetadataURL :: Text, |
There was a problem hiding this comment.
Isn't it safer to use a URI to ensure that we've got a valid URI?
There was a problem hiding this comment.
Yes, but the on-chain format does not. And given that we want to use this type to read the chain as well as write it, then we cannot impose stronger constraints in the types.
We can do optional helpful validation of course.
|
|
||
| makeByronKeyWitness :: NetworkId | ||
| -> TxBody ByronEra | ||
| -> SigningKey ByronKey | ||
| -> Witness ByronEra | ||
| makeByronKeyWitness _ (ShelleyTxBody era _ _) = case era of {} |
There was a problem hiding this comment.
This is my first time encountering EmptyCase, but is there a benefit here as opposed to using error?
There was a problem hiding this comment.
Yes! The benefit is that the compiler can tell you that it is correct!
With error you're promising the compiler that this can never happen. With empty case the compiler check for us that it can never happen. We have pattern match warnings turned on (and as errors in CI) and so the compiler can check that we're right that there are no cases here that we have missed. There are simply no possible cases at all.
| {-# LANGUAGE PatternSynonyms #-} | ||
| {-# LANGUAGE NamedFieldPuns #-} |
There was a problem hiding this comment.
| {-# LANGUAGE PatternSynonyms #-} | |
| {-# LANGUAGE NamedFieldPuns #-} | |
| {-# LANGUAGE NamedFieldPuns #-} | |
| {-# LANGUAGE PatternSynonyms #-} |
| {-# LANGUAGE FlexibleInstances #-} | ||
| {-# LANGUAGE FlexibleContexts #-} |
There was a problem hiding this comment.
| {-# LANGUAGE FlexibleInstances #-} | |
| {-# LANGUAGE FlexibleContexts #-} | |
| {-# LANGUAGE FlexibleContexts #-} | |
| {-# LANGUAGE FlexibleInstances #-} |
| && case era of | ||
| ShelleyBasedEraShelley -> txbodyA == txbodyB | ||
| ShelleyBasedEraAllegra -> txbodyA == txbodyB | ||
| ShelleyBasedEraMary -> txbodyA == txbodyB |
There was a problem hiding this comment.
Can this be simplified?
| && case era of | |
| ShelleyBasedEraShelley -> txbodyA == txbodyB | |
| ShelleyBasedEraAllegra -> txbodyA == txbodyB | |
| ShelleyBasedEraMary -> txbodyA == txbodyB | |
| && txbodyA == txbodyB |
or is there a reason for doing it this way?
There was a problem hiding this comment.
No. Can you see why?
| ByronEra -> "TxUnsignedByron" | ||
| ShelleyEra -> "TxUnsignedShelley" | ||
| AllegraEra -> "TxBodyAllegra" | ||
| MaryEra -> "TxBodyMary" |
There was a problem hiding this comment.
Do we want to maintain consistent naming here?
There was a problem hiding this comment.
I was thinking that we change it to be consistent with the type names (ie TxBody), but keep existing ones for compatibility. I'm happy to be talked out of it.
| (Seq.fromList (map toShelleyCertificate txCertificates)) | ||
| (toShelleyWithdrawal txWithdrawals) | ||
| (toShelleyLovelace fee) | ||
| (error "TODO: support validity interval") |
There was a problem hiding this comment.
| (error "TODO: support validity interval") | |
| (error "TODO: makeShelleyTransaction support validity interval") |
| => Mary.Value ledgerera -> Value | ||
| fromMaryValue (Mary.Value lovelace other) = | ||
| Value $ | ||
| --TODO: write QC tests to show it's ok to use Map.fromAscList here |
There was a problem hiding this comment.
And this would be okay because we're deriving an Ord instance for AssetId in which the earliest constructor declared is AdaAssetId, right?
So an appropriate test would basically just verify that AdaAssetId < AssetId x y.
There was a problem hiding this comment.
Both order of ada vs other assets, and order amongst non-ada assets. This should be the case because converting to/from the ledger lib PolicyId and AssetName should not affect the order.
So an appropriate test would basically just verify that AdaAssetId < AssetId x y
No, it'd be the whole order of converting the value to a list, including all assets (including cases of assets with a common policy id).
Move toByronLovelace to the Value module, and split the TxBody module section on TxIn and TxOut so they're in separate sections. We're about to add a bunch more sections for types used in the body.
Move the TxOutValue, MintValue and related functions from the Value module to the TxBody module. We're going to end up with a whole bunch of these era-dependent field types for the tx body and it makes most sense to keep them all together in one module.
dcoutts
force-pushed
the
dcoutts/api-allegra-mary
branch
2 times, most recently
from
d0e1015
to
dde3664
Compare
|
bors merge |
|
Build succeeded: |
2121: Completing API support for Allegra and Mary eras r=dcoutts a=dcoutts Following on from #2111 with what is hopefully the last remaining API changes needed to support all the Shelley-based eras. All the representations and functions should now work for all appropriate eras. 2126: Updated help text for query ledger/protocol-state r=intricate a=kevinhammond Clarified that these commands are not intended for general use, and pointed to the underlying data structures for more information about content Co-authored-by: Duncan Coutts <duncan@well-typed.com> Co-authored-by: Luke Nadur <19835357+intricate@users.noreply.github.com> Co-authored-by: Kevin Hammond <12563287+kevinhammond@users.noreply.github.com>
Generalise most of the API tx construction functions over all the Shelley-based eras.