interp: ptrtoint integer size does not equal pointer size

I think this error was supposed to be used internally for rollback, but it seems that it replicates on AVR:

package main

import "fmt"

func printOne() {
        fmt.Println("one")
}

func main() {
        f := printOne
        f()
}

This example is pulled from #349 which I was checking on.

[niaow@finch tinygo]$ build/tinygo run -target arduino /tmp/bleh.go
error: interp: ptrtoint integer size does not equal pointer size

Is this the issue that was mentioned in the patch for the conservative collector here: f870ed5 ?

No, it is not. Although idk if the patch might impact it.

Is this patch then still something I should create a PR for?

diff --git a/src/runtime/gc_conservative.go b/src/runtime/gc_conservative.go
index cabd41427..faa4091a5 100644
--- a/src/runtime/gc_conservative.go
+++ b/src/runtime/gc_conservative.go
@@ -410,7 +410,7 @@ func markRoots(start, end uintptr) {
 		}
 	}
 
-	for addr := start; addr < end; addr += unsafe.Alignof(addr) {
+	for addr := start; addr+unsafe.Sizeof(unsafe.Pointer(nil)) <= end; addr += unsafe.Alignof(addr) {
 		root := *(*uintptr)(unsafe.Pointer(addr))
 		markRoot(addr, root)
 	}

I do not know. I will need to look into that again. That should not be related to this issue.

So the answer there is that yes that code still needs to be fixed but also my fix had a bug in it because markRoots already had another load-bearing bug.

For future reference, that patch has been turned into #2396

This issue is also happening when trying to build https://github.com/google/open-location-code/, and I'm pretty sure that's a recent failure.

This issue is most likely unrelated to any runtime GC that might happen. The interp package intercepts calls to runtime.alloc and replaces them with creating a new global.

I think this error was supposed to be used internally for rollback

No, what it actually shows is that there is a bug somewhere in the interp package. It is usually caused when something stores a pointer in a non-pointer field (such as int, uintptr, [4]byte, etc). It might be fixable by passing a correct layout parameter to runtime.alloc: there are many places where this isn't yet done (for example, in llvmutil.EmitPointerPack).

While working on #2458, I found that this can be triggered by testdata/json.go, testdata/stdlib.go, and testdata/testing.go on AVR.

Not sure if this will help, but I had a similar issue until I removed all fmt and log references from my code

(This is no longer crashing open-location-code/go either wasi or native.)

os: add stubs for Readlink and File.Seek for baremetal targets, with …

#2565 includes a regression test for this, but doesn't run it yet from Makefile, as it is still failing.

os: add stubs for Readlink and File.Seek for baremetal targets, with …

I reduced one of the above testcases. To reproduce:

$ tinygo build -target arduino -o main  main.go

where main.go is

package main
func main() {
	println(Foo())
}
func Foo() int { return global.Foo() }
var global = &Bar{NewThing()}
type Bar struct {
	src Thing
}
func (r *Bar) Foo() int { return r.src.Foo() }
type Thing interface {
	Foo() int
}
func NewThing() Thing {
	return &subThing{}
}
type subThing struct {}
func (s *subThing) Foo() int {
	return 7
}

@deadprogram

@deadprogram @dkegel-fastly In the spirit of prioritizing a fix for #2732, does it make sense to go ahead and enable the regression test Dan mentions above and in #2565? I'd suggest the build should fail until the fmt regression is fixed -- not supporting the arduino is not only blocking adoption but may be masking some other issues, like the apparent println() regression I also mention over there.

I've marked the two bugs with the 0.23 milestone, that might suffice.

Thanks Dan -- that should help. After I wrote that I realized that all I'm suggesting is TDD, but decided not to press it.

Noticing that this bug is causing some uncertainty in other repos as well -- tinygo-org/drivers#328 for instance.

Yeah, it's serious, and now that it has a minimal-ish reproducer, the compiler guys have something to chew on...
thanks for making a ruckus.

os: add stubs for Readlink and File.Seek for baremetal targets, with …

I think I heard this is AVR-specific, and the AVR backend is so unstable we had to disable tests.

Navigation Menu

Repository navigation

interp: ptrtoint integer size does not equal pointer size #2389

1 remaining item

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

interp: ptrtoint integer size does not equal pointer size #2389

Description

Activity

dgryski commented on Dec 15, 2021

niaow commented on Dec 15, 2021

dgryski commented on Dec 15, 2021

niaow commented on Dec 15, 2021

niaow commented on Dec 15, 2021

niaow commented on Dec 16, 2021

dgryski commented on Dec 16, 2021

aykevl commented on Dec 17, 2021

niaow commented on Jan 1, 2022

markoxley commented on Jan 10, 2022

dgryski commented on Jan 10, 2022

dkegel-fastly commented on Jan 20, 2022

1 remaining item

dkegel-fastly commented on Mar 22, 2022

stevegt commented on Mar 22, 2022

dkegel-fastly commented on Mar 22, 2022

stevegt commented on Mar 22, 2022

dkegel-fastly commented on Mar 22, 2022

dkegel-fastly commented on Apr 7, 2022

Emreu commented on Jul 11, 2022

deadprogram commented on Oct 10, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions