Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Panics can be caught from TCO loops #9705

Merged
merged 8 commits into from
May 1, 2024
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Original file line number Diff line number Diff line change
Expand Up @@ -38,6 +38,7 @@ public abstract class CatchPanicNode extends Node {
new CallArgumentInfo[] {new CallArgumentInfo()},
InvokeCallableNode.DefaultsExecutionMode.EXECUTE,
InvokeCallableNode.ArgumentsExecutionMode.PRE_EXECUTED);
// Note [Tail call]
this.invokeCallableNode.setTailStatus(BaseNode.TailStatus.TAIL_DIRECT);
}

Expand All @@ -59,7 +60,8 @@ Object doExecute(
@Cached BranchProfile otherExceptionBranchProfile,
@CachedLibrary(limit = "3") InteropLibrary interop) {
try {
return thunkExecutorNode.executeThunk(frame, action, state, BaseNode.TailStatus.TAIL_DIRECT);
// Note [Tail call]
return thunkExecutorNode.executeThunk(frame, action, state, BaseNode.TailStatus.NOT_TAIL);
Copy link
Member

@JaroslavTulach JaroslavTulach May 1, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Panic.catch cannot be in tail call position. I believe it has to be NOT_TAIL. All the tests in Error_Spec are fixed. Let's see what other collateral damage this change may have created. Btw. I have a feeling @hubertp was fighting with some similar issue a year back or so...

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What a good observation! At the end, it seems that we will not have to dig in the IR phases. I have just added some comments and made sure that even the handler callback is not in a tail call position, just to be sure - in commit 8539997.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

On contrary to the action invocation, handler can probably be in tail position. There is nothing in the CatchPanicNode that needs to run after handler...

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Good note. Add a comment about that in d86c7b5 . handler is invoked with TAIL_DIRECT.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Btw. I have a feeling @hubertp was fighting with some similar issue a year back or so...

It's been a while so I can't remember but maybe you are referring to #7116

} catch (PanicException e) {
panicBranchProfile.enter();
Object payload = e.getPayload();
Expand Down Expand Up @@ -93,4 +95,14 @@ private Object executeCallbackOrRethrow(
}
}
}

/* Note [Tail call]
* ~~~~~~~~~~~~~~~~~~~~~~
* The `NOT_TAIL` in `isTail` parameter is important here. Panic.catch cannot be in
* the tail call position. If it is, the panic may not be caught when a tail-call-optimized
* thunk is evaluated (as that will effectively discard this method call from the stack).
*
* This is important for the `action` parameter of `Panic.catch`. The `handler` callback can
* be executed as tail, as there is nothing in this class that needs to run after the `handler`.
*/
}
43 changes: 43 additions & 0 deletions test/Base_Tests/src/Semantic/Error_Spec.enso
Original file line number Diff line number Diff line change
Expand Up @@ -44,6 +44,14 @@ foreign js throw_js_str = """
foreign js throw_js_arr = """
throw [1,2,3];

fn_with_tco x =
if x == 1 then Panic.throw "fn_with_tco" else
@Tail_Call fn_with_tco x+1

global_wrapper_fn =
fn_with_tco 0


add_specs suite_builder =
suite_builder.group "No Method Errors" group_builder->
group_builder.specify "should be recoverable" <|
Expand Down Expand Up @@ -361,6 +369,41 @@ add_specs suite_builder =
c2 . should_equal "finalizer"
v2.to_vector . should_equal [1, 2]

group_builder.specify "should be caught in a loop" <|
fn_without_tco x =
if x == 1 then Panic.throw "fn_without_tco" else
fn_without_tco x+1

p = Panic.catch Any handler=(.payload) <| fn_without_tco 0
p . should_equal "fn_without_tco"

group_builder.specify "should be caught in a TCO loop" <|
fn_with_tco x =
if x == 1 then Panic.throw "fn_with_tco" else
@Tail_Call fn_with_tco x+1

p = Panic.catch Any handler=(.payload) <| fn_with_tco 0
p . should_equal "fn_with_tco"

# The same test as the one before, but it does not use the `<|` function.
group_builder.specify "should be caught in a TCO loop with a wrapper fn" <|
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

As suggested by @JaroslavTulach, I have added this test. And as expected, this test succeeds, but the former ("should be caught in a TCO loop") fails. It seems that your suspicion, @JaroslavTulach, was right - this is probably caused by the fact that the <| function is inlineable and therefore there is no frame for it.

Copy link
Member Author

@Akirathan Akirathan Apr 29, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@JaroslavTulach EDIT: The test added in cce3c03 suggests otherwise. In that commit, I have introduced a test that doesn't use the <| method, but uses a method defined in a top scope, and it fails as well.

fn_with_tco x =
if x == 1 then Panic.throw "fn_with_tco" else
@Tail_Call fn_with_tco x+1

wrapper_fn =
fn_with_tco 0

p = Panic.catch Any handler=(.payload) wrapper_fn
p . should_equal "fn_with_tco"

# The same test as the above one, but calls a global wrapper_fn that is declared
# in top scope.
group_builder.specify "should be caught in a TCO loop with a global wrapper fn" <|
p = Panic.catch Any handler=(.payload) global_wrapper_fn
p . should_equal "fn_with_tco"


suite_builder.group "Type Errors" group_builder->
my_func x y =
x + y
Expand Down
Loading