[libraries-pgo] BBF_HAS_IDX_LEN is not set ... but is required because of the following tree

[jakobbotsch]

Failure in System.Globalization.Extensions.Tests on multiple Linux configurations.

Work item: https://dev.azure.com/dnceng-public/public/_build/results?buildId=21469&view=ms.vss-test-web.build-test-results-tab&runId=432716&paneView=dotnet-dnceng.dnceng-anon-build-release-tasks.helix-anon-test-information-tab&resultId=192990
Console log: https://helixre107v0xdeko0k025g8.blob.core.windows.net/dotnet-runtime-refs-heads-main-7b6005bf99b2486999/System.Globalization.Extensions.Tests/1/console.8d45794a.log?helixlogtype=result

BBF_HAS_IDX_LEN is not set on BB10 but is required because of the following tree 
N002 (  3,  3) [000126] ---X-------                         *  ARR_LENGTH int   
N001 (  1,  1) [000125] -----------                         \--*  LCL_VAR   ref    V11 tmp8         u:2

Assert failure(PID 21939 [0x000055b3], Thread: 21945 [0x55b9]): Assertion failed 'false' in 'Xunit.Assert:RecordException(System.Func`1[System.Object]):System.Exception' during 'Early Value Propagation' (IL size 48; hash 0xd54f2d8c; Tier1)

    File: /__w/1/s/src/coreclr/jit/earlyprop.cpp Line: 66
    Image: /datadisks/disk1/work/A44E0904/p/dotnet

[ghost]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Failure in System.Globalization.Extensions.Tests on multiple Linux configurations.

Work item: https://dev.azure.com/dnceng-public/public/_build/results?buildId=21469&view=ms.vss-test-web.build-test-results-tab&runId=432716&paneView=dotnet-dnceng.dnceng-anon-build-release-tasks.helix-anon-test-information-tab&resultId=192990
Console log: https://helixre107v0xdeko0k025g8.blob.core.windows.net/dotnet-runtime-refs-heads-main-7b6005bf99b2486999/System.Globalization.Extensions.Tests/1/console.8d45794a.log?helixlogtype=result

BBF_HAS_IDX_LEN is not set on BB10 but is required because of the following tree 
N002 (  3,  3) [000126] ---X-------                         *  ARR_LENGTH int   
N001 (  1,  1) [000125] -----------                         \--*  LCL_VAR   ref    V11 tmp8         u:2

Assert failure(PID 21939 [0x000055b3], Thread: 21945 [0x55b9]): Assertion failed 'false' in 'Xunit.Assert:RecordException(System.Func`1[System.Object]):System.Exception' during 'Early Value Propagation' (IL size 48; hash 0xd54f2d8c; Tier1)

    File: /__w/1/s/src/coreclr/jit/earlyprop.cpp Line: 66
    Image: /datadisks/disk1/work/A44E0904/p/dotnet

Author:	jakobbotsch
Assignees:	-
Labels:	area-CodeGen-coreclr, untriaged
Milestone:	-

[jakobbotsch]

cc @dotnet/jit-contrib

[EgorBo]

@jakobbotsch didn't you want to remove the check in #71707 ? 🙂

[jakobbotsch]

Yes. But I have run into some regressions in #71707 that I need to think about how to fix before that can be merged.

[jakobbotsch]

But I will assign to myself, it should give me some motivation to work on fixing that. 😉

[jakobbotsch]

#71707 will probably take a while, so we should look at this separately given that it is failing often.

From a preliminary look this is some case of nested inlining not correctly propagating BB flags back.

[jakobbotsch]

At one point here inlining ends up replacing the inline candidate call node with the return expression in-place:

runtime/src/coreclr/jit/fginline.cpp

Line 1420 in 10fc8ae

iciCall->ReplaceWith(pInlineInfo->retExpr, this);

The return expression here is itself a GT_RET_EXPR node. This means we end up with multiple copies of that particular node, and the flags are not propagated through the right one.

Not yet sure how to fix it, this is all very convoluted.