Hi Folks,
When upgrading our build container from 1.11.10 to 1.12.5, some of our
tests failed in a surprising way. I've boiled down the issue to this short
example code fragment.
package main
// #include <string.h>
// struct S { unsigned char data[18]; };
import "C"
import "unsafe"
func main() {
var array [5]C.struct_S
for i := range array {
C.memset(unsafe.Pointer(&array[i].data[0]), 0xff, C.sizeof_struct_S)
}
for i := range array {
for j := range array[i].data {
if array[i].data[j] != C.uchar(0xff) {
panic("oops")
}
}
}
}
Under 1.11 this runs to completion; under 1.12 it panics.
Debugging and instrumenting shows that the problem is the first argument to
memset is not what one would expect when run under 1.12. (The address of a
heap temporary copy rather than the original perhaps?)
Strangely, either parenthesising the argument expression to the address
operand ("&(array[i].data[0])") or removing the final index expression
("&array[i].data") "fixes" the problem.
I've spelunked through the AST (identical under both); the liveness and
escape analysis (different); the parse tree (different); and the go asm
(different). Frankly, my gc-fu is very weak and I don't really know what
I'm looking for.
I assume that the root cause is that we were violating some sort of
liveness or escape policy, but I can't figure out what it would be (or why
minor syntactic changes "fix" things). I'd be quite curious to know.
The workaround in the real code is simple, so this isn't a huge issue, just
a a bit of a mystery.
Regards,
Neil
--
You received this message because you are subscribed to the Google Groups
"golang-nuts" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to golang-nuts+unsubscr...@googlegroups.com.
To view this discussion on the web visit
https://groups.google.com/d/msgid/golang-nuts/5528b76b-c182-4cab-b3a8-ed5216042259%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
