[WIP] sig data trim #12674
[WIP] sig data trim #12674
Conversation
|
@vzarytovskii I'm closing this for now, as it's not critical for us. However we may want to return to this |
|
Tracked in #12746 |
|
|
||
| type InterfacePublic = interface end | ||
|
|
||
| type private InterfacePrivate = interface end |
There was a problem hiding this comment.
@dsyme is my understanding correct here that this should be kept only when there is an InternalsVisibleTo present on the assembly level?
There was a problem hiding this comment.
I guess, it should be kept if a public type inherits it.
We should verify what C# emits into refassembly in this case.
There was a problem hiding this comment.
Internal interfaces appear to be present in C#:
https://sharplab.io/#v2:DYLgZgzgNAJiDUAfALgTwA4FMAE6BOAlgG4CGyOAkgHbl5gkDGmACoaedgLzYE2Z2McmKjCA
Even without any InternalsVisibleTo attributes.
|
|
||
| type InterfacePublic = interface end | ||
|
|
||
| type private InterfacePrivate = interface end |
There was a problem hiding this comment.
@dsyme is my understanding correct here that this should be kept only when there is an InternalsVisibleTo present on the assembly level?
|
|
||
| val canAccessFromSomewhere: Accessibility -> bool | ||
|
|
||
| val canAccessFromSomewhereOutside: 'T -> Accessibility -> bool |
There was a problem hiding this comment.
@dsyme what was the train of thought with this first generic parameter?
I'm currently only able to deduce that the top-level Entity has the attribute inside the .Attribs property.
3 participants
While working on #11521 with @vzarytovskii we've identified that the F# signature data metadata resource blob contains Too Much Information.
In particular it contains
This means the reference assembly feature will not be quite as effective as we would like , as reference assemblies will get updated even when private implementation details change.
However, this fix to reduce the information on the signature bloc is of a different nature to the rest of the work on reference assemblies and is in a sense completely independent from a logical/correctness point of view. It also has some subtle edge cases, and thus we think this fix is best made separately, and perhaps released under preview or else have a command line option to disable it.
The reference assembly feature is still highly useful without this fix as there are plenty of cases where it's still useful without this.