Contract info should specify the defining lib for ADTs #786
Conversation
There was a problem hiding this comment.
I think we will need different versions of pp_typ for scilla-checker and scilla-runner. The former should use library names as the qualifier for custom ADTs, whereas the latter should use addresses.
This feature keeps being hairy, but I do think we are on the home stretch wrt. design decisions. To slightly rephrase a Danish expression, "this goat will soon have been shaved". :-)
| @@ -20,7 +20,8 @@ | |||
| "end_location": { "file": "", "line": 0, "column": 0 } | |||
| }, | |||
| { | |||
| "error_message": "Type Test cannot be used as transition parameter", | |||
| "error_message": | |||
| "Type unserializable_param.scilla.Test cannot be used as transition parameter", | |||
There was a problem hiding this comment.
In error messages it is probably better to use the name of the library (in this case UnserializableADT) rather than the file name (and even if we use the file name, the ".scilla" part should be stripped). Also, the type Test is defined in the contract's local library, so we don't need the qualified name here.
More importantly, though, this raises an issue regarding ADTs defined locally (and the sample contract I wrote for you will probably highlight that): A locally defined ADT T will have the qualified name _this_address.T. When we perform a dynamic typecheck of that type, the database won't know about the contract at _this_address yet, so we have to deal with that address as a special case.
It gets worse, though, because what if I have the following contract:
library L
type MyTyp =
| Foo
contract (typed_self : ByStr20 with f : MyTyp end)
field f : MyTyp = Foo
If this contract is deployed with typed_self = _this_address, then the address type refers to the unqualified type MyTyp, which means it must be defined locally, and thus wouldn't be in scope in the remote contract. However, since typed_self refers not just to a copy of the same contract but to the same contract itself, we have to consider whether MyTyp is in fact in scope.
It's probably not hugely important, but it would be a pretty cool feature, and I think all we need to do is to treat _this_address as a special case when performing the dynamic typecheck. Self-application RULES! :-)
(Side note: For version 1 I really think we ought to think about providing an implicit immutable field typed_self for this purpose)
| @@ -15,14 +15,14 @@ | |||
| "events": [], | |||
| "ADTs": [ | |||
| { | |||
| "tname": "TestType1", | |||
| "tname": "adt_test.scilla.TestType1", | |||
There was a problem hiding this comment.
Should we use the unqualified name here? I think the "ADTs" group is used by savant-IDE, so we should provide something that is easy to use for IDEs. Also, since this is output from scilla-checker we can't assume we have an address to output at this point, since the contract won't have been deployed yet.
Come to think of it, though, maybe it is better to use the qualified name (using the library name as the qualifier), since it makes it easy to distinguish between user-defined and built-in ADTs.
I'm still not certain, though, so alternative opinions are welcome. :-)
@jjcnn Even during checking, the blockchain can provide us the actual address to which this contract will be deployed, so we can print that for ADTs defined inside the contract.
We don't have much control over printing addresses. In the case of running in a blockchain, the blockchain will ensure that the filename it uses for a library is the address of that library. Hence we'll print the addresses. In the case of an IDE, users typically name the filename based on the library, and we end up printing the filename (= library name). Getting the actual library name specified inside the |
Hmm... Let's try this way of doing it. I just suspect it will be an issue once users start using the feature. |
|
Should we merge this then? |
I think so. |
* Create new branch * Define Address type (#763) * Define Address type * Define legal Address types for messages, parameters and fields. * Syntax for address types (#767) * Added address type to syntax, and fixed ambiguous grammar * Add location info to Address field identifiers * Type inference rules for addresses (#778) * assert_type_assignable * Use assert_type_assignable in TypeUtil. * Type inference rules for address types * Fix type_assignable and tests of same, introduce legal_procedure_parameter_type, and fix parser bug * Fix parser bug and typechecker issue * Refactor type equivalence tests * Assignable tests * Duplicate field check for address types * Remove fixed TODO * Move duplicate field check to Recursion.ml * Remove blank line * Reintroduce is_ground_type * Cosmetic change to pp_typ in case the address contains no fields * Addresses from JSONs should be read as ByStr20s * Reorder arguments to for_all2_exn in builtin argument traversal * Do not use type_assignable when comparing types derived from literal values * Use type_equiv in Schnorr calculations, and rename json_to_lit to json_to_lit_exn * Remote state read read syntax (#783) * Remote state read syntax and typechecking * _balance field must be accessible even when not declared * Change <- to <-- for remote reads * Add address type to map key and value types * Removed incorrect comment * Simple example contract for remote state reads, plus minor parser fix (#785) * Field initialiser bugs fixed (#787) * Fixed two bugs re. initialisers for address fields. * Update src/base/TypeChecker.ml Co-Authored-By: Vaivaswatha N <vaivaswatha@users.noreply.github.com> Co-authored-by: Vaivaswatha N <vaivaswatha@users.noreply.github.com> * Contract info should specify the defining lib for ADTs (#786) * Provide an external_fetch function in StateService (#805) * Add IPC methods to fetch external contract state * Provide means to specify external states via input JSONs * Add a dummy fetch_ext_state_value to StateIPCTestServer * Provide external_fetch service in StateService.ml * Fix compare function for identifiers and names. (#926) * Fix outstanding merge issues (#925) * Fix outstanding merge issues * fmt * Use assert_type_assignable rather than assert_type_equiv * fmt * `_sender` and `_this_address` are addresses (#801) * _sender is an address, eq should compare addresses as ByStr20 * Added addresses as legal ADT constructor arguments * Addresses in event fields output as ByStr20 * fmt * Missing merge file * Remote state reads practicalities (#927) * Name parameters to type_assignable and assert_type_assignable * address_typ to construct address type nodes * fmt * Allow parenthesised address types (#937) * Fetch external value IPC query cannot provide expected type as that isn't exactly known. * Allow fetching type only for remote fields * fmt * Remote state reads bugfixes (#945) * Use declared type rather than actual type in field env * Allow _this_address to be declared in address types * Disallow spids other than _this_address in address types * Remote state reads testserver (#948) * JSON parsing of external states * Add testsuite support for external states * Added remote state reads test contract * Test contract * implement remote map read * Add comment clarifying computing accessed value type * Fix issue with address types being lost * fmt Co-authored-by: Vaivaswatha Nagaraj <vaivaswatha@zilliqa.com> * Merge * Push immutable parameters via state service, along with fields * Push contract parameters to state service during migration * Revert "Push contract parameters to state service during migration" This reverts commit 2ea064c. * Revert "Push immutable parameters via state service, along with fields" This reverts commit 2cf3978. * Remote state reads dynamic typecheck (#953) * Parameter typechecker for addresses * Ignore types in init.json (not backward compatible), remove faulty assignability check for field initialisers * Dynamic typecheck of contract parameters * Missing commit of TypeUtil.mli * Dynamic typecheck of contract parameters * Minor tidying up * Comments added * Separate prepare_message from handle_message to allow for dynamic typechecks * Dynamic typecheck of transition parameters, plus some (not all) tests * Negative tests for addresses and ADTs * Remote state reads misc bugfixes (#954) * Address types are legal map value types * Fix type_assignable * Added comment about contravariance in custom ADTs * Tests of _this_address in address types and remote reads * Tests that messages, events and excetions use ByStr20 and not Address types * Forgot to add the new tests to Testcontracts.ml * Added assignability tests * Decrease sender balance on acceptance (#955) * make fmt * Do not pretty-print Lists as JSON arrays. This confuses the (#956) state parser in the blockchain unit test which cannot know if its a map or not (yet). I ended up running `make fmt` which resulted in some other changes as well, but just whitespace changes. Co-authored-by: Jacob Johannsen <jajocnn@gmail.com> * Fix missing stuff in #956 (#957) * Fix missing stuff in #956 * Change _this_address of test contract * Update owner address and balance (to cover gas fee in C++ test) * Remote state reads nonce check (#959) * Added nonce check * Tests for nonce > 0 || balance > 0 * New syntax for address types and remote state reads. (#960) * Added nonce check * Tests for nonce > 0 || balance > 0 * Changed remote read operator from <-- to <-& * Change address type syntax * Remove _this_address production from address_type_field * Remove _this_address from Address AST nodes * make fmt * Change nonce_type to Uint64 (#961) * Scilla-server wrapper for the disambiguation tool (#964) * Disambiguator should output init JSON to file (#965) * Ignore _balance output (#966) * Remote state reads checker tests (#963) * Equality tests on address types * Non-storable types not allowed in address fields. Contract parameters checked for serializability. eq tests on addresses * Address list traversal tests * Namespaces not allowed when remote reading * Polymorphic address types * Test remote read from non-address * Additional negative remote read tests * Test of deep remote reads * Various deep remote reads tests * Fix problem map builtins on maps with addresses as key types * Only address argument types to put should be converted to ByStr20 * fmt * Merge fixes * fmt * Additional tests of assignability (#962) * Fix contract parameter validation check (#969) * Disable check. Another check needs to be enabled before merging * Allow maps as contract parameters * Validate contract parameters * fmt * Fixed elaboration of to_string, strrev and to_ascii for address types (#971) * Handle migration of contracts with missing state (#972) * Handle migration of contracts with missing state * fmt * Make contract init tests aware of scilla server tests Co-authored-by: Vaivaswatha N <vaivaswatha@users.noreply.github.com> Co-authored-by: Vaivaswatha Nagaraj <vaivaswatha@zilliqa.com>
No description provided.