JIT: Stop treating GT_LCLHEAP as inherently exceptional and remove dead StackOverflowException flag by Copilot · Pull Request #125362 · dotnet/runtime

Copilot · 2026-03-10T04:01:58Z

Description

GT_LCLHEAP was previously marked as inherently exception-producing via GTF_EXCEPT in the importer, and OperExceptions also reported StackOverflowException for it. Based on review feedback, this change removes that special exception treatment instead of trying to delete GT_LCLHEAP nodes in liveness.

As a follow-up cleanup, the now-unused ExceptionSetFlags::StackOverflowException enum value is also removed from the JIT exception flag set.

Changes

importer.cpp — Stopped marking imported GT_LCLHEAP nodes with GTF_EXCEPT. The node still gets GTF_DONT_CSE.
gentree.cpp — Removed the GT_LCLHEAP case from GenTree::OperExceptions, so it no longer reports StackOverflowException as an inherent exception side effect.
liveness.cpp — Reverted the earlier liveness-specific dead-code-elimination handling for GT_LCLHEAP.
morph.cpp — Updated the call-argument reordering logic to check for GT_LCLHEAP explicitly in the places where its stack-depth behavior still matters, rather than relying on GTF_EXCEPT.
utils.h — Removed the dead ExceptionSetFlags::StackOverflowException enum member after the GT_LCLHEAP exception handling change made it unused.
Behavioral intent — GT_LCLHEAP is no longer treated as inherently exceptional solely because it may grow the stack. Any exception-related behavior now comes from propagated operand flags rather than special casing on the node itself, and the unused stack overflow exception flag has been deleted from the JIT exception set.

Testing

./build.sh clr.jit -rc release
./build.sh libs -rc release -lc release
src/tests/build.sh -Test JIT/Directed/localloc/localloc3_cs_r.csproj x64 Release -priority1

Focused validation succeeded for the affected JIT and localloc test build path.

🔒 GitHub Advanced Security automatically protects Copilot coding agent pull requests. You can protect all pull requests by enabling Advanced Security for your repositories. Learn more about Advanced Security.

If the GT_LCLHEAP node's value is unused (no consumer in LIR), remove the node and mark its size arg as unused so it is still evaluated for side effects. This works even when the length is not a constant. Co-authored-by: EgorBo <523221+EgorBo@users.noreply.github.com>

…oval - Change LowerLclHeap from void to GenTree* return type to correctly communicate the next node for the lowering loop - Save node->gtNext before removing the node when value is unused, since Remove() sets gtNext to nullptr - Also convert all internal early returns to return node->gtNext - Update the GT_LCLHEAP case in LowerNode to use 'return LowerLclHeap(node)' Co-authored-by: EgorBo <523221+EgorBo@users.noreply.github.com>

jakobbotsch · 2026-03-10T06:05:09Z

Doesn't liveness remove these?

EgorBo · 2026-03-10T10:04:21Z

Doesn't liveness remove these?

given it produces diffs - it doesn't?

@copilot please fix liveness to delete unused LCLHEAP nodes instead.

Co-authored-by: EgorBo <523221+EgorBo@users.noreply.github.com>

Copilot · 2026-03-10T10:32:51Z

Doesn't liveness remove these?

Yes. I moved the removal out of lowering and into liveness in 6456523, so unused GT_LCLHEAP nodes are now deleted during DCE and their size operand is marked unused there.

Doesn't liveness remove these?

given it produces diffs - it doesn't?
...

Updated in 6456523. LowerLclHeap is back to its original behavior, and liveness now removes unused GT_LCLHEAP nodes via DCE while marking the size operand unused so any side effects are preserved.

jakobbotsch · 2026-03-10T10:59:49Z

I think the real problem is that we mark GT_LCLHEAP with GTF_EXCEPT. GTF_EXCEPT nodes cannot be removed by liveness.

It feels inconsistent to be deleting them anyway. If we think we can do that, then we should probably stop marking them with GTF_EXCEPT.

jakobbotsch · 2026-03-10T11:01:46Z

But it is not clear that is valid -- stackalloc int[-1] is probably expected to throw something? So perhaps the best we can do is transform to a non-negative check on the length.

Edit: well actually, localloc IL instruction takes an unsigned integer. So removing entirely is probably ok.

dotnet-policy-service · 2026-03-10T12:14:33Z

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

EgorBo · 2026-03-10T13:33:15Z

@copilot please revert the changes, just remove GTF_EXCEPT from LCLHEAP in importer (unless its argument propagates it). Remove it from OperExceptions. Check everywhere, we can assume StackOverflow is not a side-effect we care about.

Co-authored-by: EgorBo <523221+EgorBo@users.noreply.github.com>

Copilot · 2026-03-10T13:54:18Z

I think the real problem is that we mark GT_LCLHEAP with GTF_EXCEPT. GTF_EXCEPT nodes cannot be removed by liveness.

It feels inconsistent to be deleting them anyway. If we think we can do th...

Agreed. I changed the implementation in c59b003 to stop treating GT_LCLHEAP as inherently exceptional instead of deleting it specially in liveness. The liveness change is reverted, importer no longer sets GTF_EXCEPT, and OperExceptions no longer reports StackOverflowException for GT_LCLHEAP.

But it is not clear that is valid -- stackalloc int[-1] is probably expected to throw something? So perhaps the best we can do is transform to a non-negative check on the length.

Edit: well actu...

Went with that simpler direction in c59b003. Since localloc takes an unsigned size and we’re no longer modeling stack overflow here as an exception side effect, GT_LCLHEAP now just stops carrying the inherent GTF_EXCEPT/OperExceptions behavior.

@copilot please revert the changes, just remove GTF_EXCEPT from LCLHEAP in importer (unless its argument propagates it). Remove it from OperExceptions. Check everywhere, we can assume StackOve...

Done in c59b003. I reverted the liveness-specific removal, removed the importer-side GTF_EXCEPT from GT_LCLHEAP, dropped its OperExceptions entry, and updated the morph-side reordering logic to check for GT_LCLHEAP explicitly where it still matters.

EgorBo · 2026-03-10T18:13:42Z

PTAL @jakobbotsch @dotnet/jit-contrib Diffs

I think it also fixes an existing CQ github issue I had in my bookmarks, but I can't find it.

Copilot

Pull request overview

Updates CoreCLR JIT handling of GT_LCLHEAP (localloc) so it’s no longer treated as inherently exception-producing, and adjusts argument reordering logic to account for GT_LCLHEAP explicitly rather than via GTF_EXCEPT.

Changes:

Stop setting GTF_EXCEPT on imported GT_LCLHEAP nodes (while keeping GTF_DONT_CSE).
Remove GT_LCLHEAP from GenTree::OperExceptions so it no longer reports StackOverflowException as an inherent exception.
Update call-argument reordering constraints to explicitly detect GT_LCLHEAP in argument trees.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

File	Description
src/coreclr/jit/importer.cpp	Removes `GTF_EXCEPT` from imported `GT_LCLHEAP` while preserving `GTF_DONT_CSE`.
src/coreclr/jit/gentree.cpp	Removes inherent `StackOverflowException` reporting for `GT_LCLHEAP` in `OperExceptions`.
src/coreclr/jit/morph.cpp	Makes argument reordering constraints explicitly detect `GT_LCLHEAP` rather than relying on `GTF_EXCEPT`.
src/coreclr/jit/lower.cpp	Fixes parameter name in `LowerLclHeap` comment to match the function signature.

src/coreclr/jit/morph.cpp

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

src/coreclr/jit/gentree.cpp

Co-authored-by: EgorBo <523221+EgorBo@users.noreply.github.com>

…ad StackOverflowException flag (#125362) ## Description `GT_LCLHEAP` was previously marked as inherently exception-producing via `GTF_EXCEPT` in the importer, and `OperExceptions` also reported `StackOverflowException` for it. Based on review feedback, this change removes that special exception treatment instead of trying to delete `GT_LCLHEAP` nodes in liveness. As a follow-up cleanup, the now-unused `ExceptionSetFlags::StackOverflowException` enum value is also removed from the JIT exception flag set. ### Changes - **`importer.cpp`** — Stopped marking imported `GT_LCLHEAP` nodes with `GTF_EXCEPT`. The node still gets `GTF_DONT_CSE`. - **`gentree.cpp`** — Removed the `GT_LCLHEAP` case from `GenTree::OperExceptions`, so it no longer reports `StackOverflowException` as an inherent exception side effect. - **`liveness.cpp`** — Reverted the earlier liveness-specific dead-code-elimination handling for `GT_LCLHEAP`. - **`morph.cpp`** — Updated the call-argument reordering logic to check for `GT_LCLHEAP` explicitly in the places where its stack-depth behavior still matters, rather than relying on `GTF_EXCEPT`. - **`utils.h`** — Removed the dead `ExceptionSetFlags::StackOverflowException` enum member after the `GT_LCLHEAP` exception handling change made it unused. - **Behavioral intent** — `GT_LCLHEAP` is no longer treated as inherently exceptional solely because it may grow the stack. Any exception-related behavior now comes from propagated operand flags rather than special casing on the node itself, and the unused stack overflow exception flag has been deleted from the JIT exception set. ## Testing - `./build.sh clr.jit -rc release` - `./build.sh libs -rc release -lc release` - `src/tests/build.sh -Test JIT/Directed/localloc/localloc3_cs_r.csproj x64 Release -priority1` Focused validation succeeded for the affected JIT and localloc test build path.  --- 🔒 GitHub Advanced Security automatically protects Copilot coding agent pull requests. You can protect all pull requests by enabling Advanced Security for your repositories. [Learn more about Advanced Security.](https://gh.io/cca-advanced-security) --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Initial plan

6087630

Copilot AI assigned Copilot and EgorBo Mar 10, 2026

Copilot started work on behalf of EgorBo March 10, 2026 04:02 View session

github-actions bot added the area-System.Security label Mar 10, 2026

Copilot AI and others added 2 commits March 10, 2026 04:49

Copilot AI changed the title ~~[WIP] Remove unused GT_LCLHEAP node and set arg~~ JIT: Remove unused GT_LCLHEAP node in LowerLclHeap Mar 10, 2026

Copilot finished work on behalf of EgorBo March 10, 2026 04:55

Copilot started work on behalf of EgorBo March 10, 2026 10:04 View session

JIT: remove unused GT_LCLHEAP in liveness

6456523

Co-authored-by: EgorBo <523221+EgorBo@users.noreply.github.com>

Copilot AI changed the title ~~JIT: Remove unused GT_LCLHEAP node in LowerLclHeap~~ JIT: Remove unused GT_LCLHEAP node in liveness Mar 10, 2026

Copilot finished work on behalf of EgorBo March 10, 2026 10:33

vcsjones added area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI and removed area-System.Security labels Mar 10, 2026

Copilot started work on behalf of EgorBo March 10, 2026 13:33 View session

JIT: stop treating GT_LCLHEAP as exceptional

c59b003

Co-authored-by: EgorBo <523221+EgorBo@users.noreply.github.com>

Copilot AI changed the title ~~JIT: Remove unused GT_LCLHEAP node in liveness~~ JIT: Stop treating GT_LCLHEAP as inherently exceptional Mar 10, 2026

Copilot finished work on behalf of EgorBo March 10, 2026 13:55

build-analysis bot mentioned this pull request Mar 10, 2026

slow macOS - "##[error]The job running on agent Azure Pipelines 9 ran longer than the maximum time of 60 minutes." dotnet/dnceng#1883

Open

3 tasks

This was referenced Mar 10, 2026

The Operation will be canceled. The next steps may not contain expected logs. dotnet/dnceng#3008

Open

[android-arm64] The Operation will be canceled. The next steps may not contain expected logs. dotnet/dnceng#6408

Open

EgorBo marked this pull request as ready for review March 10, 2026 18:13

Copilot AI review requested due to automatic review settings March 10, 2026 18:13

EgorBo requested a review from jakobbotsch March 10, 2026 18:13

Copilot started reviewing on behalf of EgorBo March 10, 2026 18:15 View session

Copilot AI reviewed Mar 10, 2026

View reviewed changes

src/coreclr/jit/morph.cpp Outdated Show resolved Hide resolved

EgorBo and others added 2 commits March 10, 2026 19:29

Update src/coreclr/jit/morph.cpp

df61c51

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Update morph.cpp

118d144

build-analysis bot mentioned this pull request Mar 10, 2026

Checkout failure: "Git fetch failed with exit code 128" dotnet/arcade#9009

Open

2 tasks

Merge branch 'main' into copilot/remove-gt-lclheap-node-unused

2dd9c42

jakobbotsch reviewed Mar 10, 2026

View reviewed changes

src/coreclr/jit/gentree.cpp Show resolved Hide resolved

Copilot started work on behalf of EgorBo March 10, 2026 19:24 View session

JIT: delete ExceptionSetFlags::StackOverflowException

e5c4402

Co-authored-by: EgorBo <523221+EgorBo@users.noreply.github.com>

Copilot AI changed the title ~~JIT: Stop treating GT_LCLHEAP as inherently exceptional~~ JIT: Stop treating GT_LCLHEAP as inherently exceptional and remove dead StackOverflowException flag Mar 10, 2026

Copilot finished work on behalf of EgorBo March 10, 2026 19:41

jakobbotsch approved these changes Mar 10, 2026

View reviewed changes

EgorBo merged commit 85e141e into main Mar 10, 2026
133 checks passed

EgorBo deleted the copilot/remove-gt-lclheap-node-unused branch March 10, 2026 23:58

dotnet-maestro bot mentioned this pull request Mar 11, 2026

[main] Source code updates from dotnet/runtime dotnet/dotnet#5345

Merged

github-actions bot locked and limited conversation to collaborators Apr 10, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

JIT: Stop treating GT_LCLHEAP as inherently exceptional and remove dead StackOverflowException flag#125362

JIT: Stop treating GT_LCLHEAP as inherently exceptional and remove dead StackOverflowException flag#125362
EgorBo merged 9 commits intomainfrom
copilot/remove-gt-lclheap-node-unused

Copilot AI commented Mar 10, 2026 •

edited

Loading

Uh oh!

jakobbotsch commented Mar 10, 2026

Uh oh!

EgorBo commented Mar 10, 2026

Uh oh!

Copilot AI commented Mar 10, 2026 •

edited

Loading

Uh oh!

jakobbotsch commented Mar 10, 2026

Uh oh!

jakobbotsch commented Mar 10, 2026 •

edited

Loading

Uh oh!

dotnet-policy-service bot commented Mar 10, 2026

Uh oh!

EgorBo commented Mar 10, 2026

Uh oh!

Copilot AI commented Mar 10, 2026 •

edited

Loading

Uh oh!

EgorBo commented Mar 10, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Conversation

Copilot AI commented Mar 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Changes

Testing

Uh oh!

jakobbotsch commented Mar 10, 2026

Uh oh!

EgorBo commented Mar 10, 2026

Uh oh!

Copilot AI commented Mar 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jakobbotsch commented Mar 10, 2026

Uh oh!

jakobbotsch commented Mar 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dotnet-policy-service bot commented Mar 10, 2026

Uh oh!

EgorBo commented Mar 10, 2026

Uh oh!

Copilot AI commented Mar 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

EgorBo commented Mar 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Copilot AI commented Mar 10, 2026 •

edited

Loading

Copilot AI commented Mar 10, 2026 •

edited

Loading

jakobbotsch commented Mar 10, 2026 •

edited

Loading

Copilot AI commented Mar 10, 2026 •

edited

Loading

EgorBo commented Mar 10, 2026 •

edited

Loading