sdk
sdk copied to clipboard
dotnet
Use architecture specific DOTNET_ROOT variable name
Corrects parsing of runtime identifier added by https://github.com/dotnet/sdk/pull/19860 and reuses the code in dotnet watch to do the same.
Other than parsing the rid the behavior of https://github.com/dotnet/sdk/pull/19860 is preserved. However, it doesn't seem consistent with https://github.com/dotnet/sdk/issues/19743#issuecomment-901404323 in the case the architecture specified in rid is unknown. The implementation in this case sets DOTNET_ROOT variable, while it seems it shouldn't since an unknown architecture doesn't match the current architecture.
Implements https://github.com/dotnet/aspnetcore/issues/35793
![azure-pipelines[bot] avatar](https://avatars.githubusercontent.com/in/9426?v=4 "Published by a pub.dev verified publisher"){kind=small}
However, it doesn't seem consistent with https://github.com/dotnet/sdk/issues/19743#issuecomment-901404323 in the case the architecture specified in rid is unknown. The implementation in this case sets DOTNET_ROOT variable, while it seems it shouldn't since an unknown architecture doesn't match the current architecture.
That comment was mainly about the case where both architectures are known and the SDK architecture is different from the target architecture - in that case SDK should not set anything, since it doesn't know anything about the target's architecture situation on the machine - thus it should fallback to existing state on the machine to figure it out. The case where the target architecture is unknown typically means in the "dotnet run" case that it was not specified (I think that can happen), in which case the desire for "dotnet run" is to run the app with the SDK environment - the app should default to SDK's RID in such case basically (which should be the behavior of "dotnet build" as well). I don't know if this is something which actually happens, since I think the msbuild targets will eventually default the target RID to the SDK RID, and thus the command should see the defaulted RID, but maybe it doesn't happen always.
I don't know what is the behavior for "dotnet watch" in this case. The logic should be:
- If the target is to run on the same architecture as the SDK -> modify the environment to use the SDK's runtime location
- If the target is to run on a different architecture -> don't modify anything, as the SDK has no clue what that architecture situation is.
I don't feel strongly about the behavior when the target architecture is unknown (I think and hope that it's rare anyway), but since "dotnet run" already defaults to "assume SDK's architecture", then I guess "dotnet watch" should be consistent.
@MarcoRossignoli as an FYI, since he implemented (and maintains) a similar logic in "dotnet test".
@tmat the linked issue is from a year ago and listed for .net 8. Is this not too risky to take in rc2 at this time?
@marcpopMSFT I'm fine moving to .NET 8. Note though that this also corrects the parsing of RID in dotnet run originally implemented in https://github.com/dotnet/sdk/pull/19860.
