eacl: Improvements and test coverage #605

cthulhu-rider · 2024-07-18T18:28:32Z

No description provided.

codecov · 2024-07-23T08:08:21Z

Codecov Report

Attention: Patch coverage is 97.80822% with 8 lines in your changes missing coverage. Please review.

Project coverage is 54.43%. Comparing base (c30d691) to head (402b83c).
Report is 1 commits behind head on master.

Files	Patch %	Lines
eacl/record.go	91.66%	2 Missing and 2 partials ⚠️
eacl/table.go	88.46%	2 Missing and 1 partial ⚠️
eacl/target.go	97.87%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #605      +/-   ##
==========================================
+ Coverage   53.69%   54.43%   +0.74%     
==========================================
  Files         164      164              
  Lines       19113    19172      +59     
==========================================
+ Hits        10262    10436     +174     
+ Misses       8447     8338     -109     
+ Partials      404      398       -6

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

eacl/enums.go

carpawell · 2024-07-23T11:46:18Z

eacl/enums.go

-	// OperationRangeHash is an object payload range hashing Operation.
-	OperationRangeHash
+	OperationUnknown   Operation = iota // Deprecated: use zero instead.
+	OperationGet                        // ObjectService.Get RPC


what is ObjectService?

https://github.com/nspcc-dev/neofs-api/blob/v2.16.0/proto-docs/object.md#neo.fs.v2.object.ObjectService

yeah, i understand, just not sure if go SDK should fix such "proto package links" in comments. however, i think everybody can understand what is meant here

so - keep it as is?

eacl/enums.go

carpawell · 2024-07-23T12:00:04Z

eacl/table.go

-
-	return true
-}
+// Deprecated: compare [Table.Marshal] instead.


it was added in #180 and i really think this comparison is faster than marshaling + bytes comparison. waiter does not use it just because it happened so i think. otherwise, #180 (approved and closed) was a mistake

waiter.ContainerSetEACLWaiter
compares binary representations which is correct

i do not agree, why? binary representation depends on many things

In particular, the eACL does not change when targets are shuffled

this is one more pro to manual comparison, not just "are bytes equal?"

waiter does not use it just because it happened so i think

instead, i think it behaves as expected. Since specific method to exact table presence in the network, it compares its state in the Container contract

binary representation depends on many things

false, it is stable

this is one more pro to manual comparison, not just "are bytes equal?"

i mean its really hard to declare comparison op for such a complex type which is not comparable overall

i do not agree, why?

instances may equal binary, structurally, essentially, etc. when code compares bytes explicitly, it emphasizes the criterion of equality. Again, i think waiter is doin well

i dont mind to keep it, but i wouldnt recommend using it in the app code, only for test purposes mb

binary representation depends on many things

false, it is stable

array order? table version? both look like the same rule set to me, although may have a little bit different byte representation

i do not agree, why?

instances may equal binary, structurally, essentially, etc. when code compares bytes explicitly, it emphasizes the criterion of equality. Again, i think waiter is doin well

can understand you. but to me, acl table equality is a table that leads to the same access for every possible request sender

ofc array order will change proto binary, so equality is very context-dependent. Binary equality is what, for example, waiter expects. Otherwise, nobody should compare eACLs outside testing purposes imo

acl table equality is a table that leads to the same access for every possible request sender

ur talkin about behavioral identity, and the current implementation violates it since a false negative return is possible: array of independent records may be shuffled w/o access changes. In this case, EqualTables returns false. Such tables are essentially the same, but bin diff

eacl/filter.go

eacl/enums.go

roman-khimov · 2024-07-23T16:26:27Z

eacl/enums.go

-}
+//
+// Deprecated: do not use it! The protocol does not define a consistent string
+// format. Declare your own string mapping.


Why can't SDK provide this? Why should everyone implement some mappings?

these are not string types in their essence, so it is redundant to introduce a consistent string format imo. Can be implemented in a subpackage if this is such a popular need. But i wouldnt leave it as a core interface

SDK is far from being just a "core interface". It has to provide some sugar as well, this is one of the cases like that. Yeah, we've got String(), but then missing a DecodeString() it can make SDK less user-friendly.

i prefer to leave the ability of changing String() return values. To me, they must be printable and human-readable, nothin more. And i dont mind to provide string mapping, i just dont think it should be a type interface

implemented as functions around enum types. Will be a typical example if the app decides to implement custom mapping

eacl/filter.go

eacl/target.go

eacl/filter.go

eacl/record.go

eacl/filter.go

carpawell

But I would wait for @roman-khimov's threads.

roman-khimov · 2024-08-07T16:04:43Z

Conflicts.

To get closer to the protocol. `Unknown` constants are deprecated. Signed-off-by: Leonard Lyubich <leonard@morphbits.io>

Also drop mention of to-be-deprecated elements. Signed-off-by: Leonard Lyubich <leonard@morphbits.io>

Enum types provide `EncodeToString`/`DecodeString` interface representing stable stringers. Since the string representation is not declared by the NeoFS protocol, applications are allowed to use this functionality only for internal conversions (e.g. parsing command line input). In addition, enumeration types are simple integers. Thus, implementing this functionality as type methods is quite redundant. At the same time, exporting consistent in-box mapping can simplify development of some apps. `XToString` / `XFromString` function are provided now. Being global functions they emphasize that if the app is satisfied with the lib strings - it can take and use them, otherwise reimplement them by analogy. Methods are deprecated for the sake of new functions. Signed-off-by: Leonard Lyubich <leonard@morphbits.io>

Continues 71381ff. Signed-off-by: Leonard Lyubich <leonard@morphbits.io>

The function is unused. For example, `waiter.ContainerSetEACLWaiter` compares binary representations which is correct. Overall, NeoFS protocol does not declare comparison operation b/w two eACLs. In particular, the eACL does not change when targets are shuffled. Thus, to avoid introducing a synthetic property, the func will be removed soon. Signed-off-by: Leonard Lyubich <leonard@morphbits.io>

All other types are components and are not transported autonomously outside the eACL. Thus, for the sake of minimal support, these functions will be removed soon. Signed-off-by: Leonard Lyubich <leonard@morphbits.io>

Refs #483. Signed-off-by: Leonard Lyubich <leonard@morphbits.io>