Can't find the bug

[Maurizio Cimadamore]

Hi Bernd,
the problem in your code lies here:

```
((MemorySegment)strcat.invokeExact(buf, other))
                                        .getUtf8String(0);  // 
<----------------------
```

That is, you are dereferencing the result of a "strcat" call. Now, while 
you "know" that the resulting pointer will be a zero-terminated string 
of some known size, the Java runtime has no way to know that. Attached 
to the pointer there could be one char, two char, or infinite - the 
linker has no way to know.

For this reason, all pointers returned by the linker are modelled as 
"zero-length memory segments" (or ZLM in short). This is described here:

https://download.java.net/java/early_access/jdk20/docs/api/java.base/java/lang/foreign/MemorySegment.html#wrapping-addresses

To "resize" a ZLM you can create a _new_ segment, using the 
MemorySegment::ofAddress factory, which takes a raw address, a size and 
a scope.

Or, more simply, you can use an *unbounded* address layout when 
constructing the downcall method handle, as this:

```
strcat = linker.downcallHandle(linker.defaultLookup().find("strcat").get(),
FunctionDescriptor.of(ADDRESS.asUnbounded(), ADDRESS, ADDRESS));
```

This will tell the linker to resize the returned segment to the maximal 
size (Long.MAX_VALUE), which will make your dereference work.

Which path you take is up to you - the former gives you more control - 
and more safety; the latter might be useful if you just want "things to 
work". But both involve a "leap of faith" safety-wise, as the Java 
runtime is trusting you to know what's coming out of that pointer. For 
this reason, both paths will require a call to a so called "restricted" 
method, which will produce a runtime warning (which can be useful when 
diagnosing the source of unsafety in case of e.g. a JVM crash).

Hope this helps.

Maurizio

On 04/01/2023 07:52, Bernd Müller wrote:
> Dear Gavin,
>
> thank you very much for extracting the code.
> Eventually, it would be much easier to do a "mvn test". To do so
> please clone:
> https://urldefense.com/v3/__https://github.com/BerndMuller/ffm2__;!!ACWV5N9M2RV99hQ!PmaiPHl2BbBSjKjJr5SwpasFBcCCILMjVmcHQjs4n10ixKwpJyHK3aucRIBDoLKroPXW46Wtfvqm_mHjWecm3dfSiaicIK_yzGM$ 
>
> Error still:
> StdLibTest.foo:49->strcat:60 » IndexOutOfBounds Out of bound access on 
> segment MemorySegment{ array: Optional.empty address:139631472807584 
> limit: 0 }; new offset = 0; new length = 1
>
> looks for me like strcat changes the array size
>
> Kind Regards,
>
> Bernd
>
> Am 04.01.23 um 02:12 schrieb Gavin Ray:
>> Maurizio, the code I was able to get out of the email is below:
>>
>> public class StdLibTest {
>>
>> private static MethodHandle strlen; // C's strlen() function
>> private static MethodHandle strcat; // C's strcat() function
>>
>>      @BeforeAll
>> public static void locateFunctions() {
>> Linker linker = Linker.nativeLinker();
>> strlen = 
>> linker.downcallHandle(linker.defaultLookup().find("strlen").get(),
>> FunctionDescriptor.of(JAVA_LONG, ADDRESS));
>> strcat = 
>> linker.downcallHandle(linker.defaultLookup().find("strcat").get(),
>> FunctionDescriptor.of(ADDRESS, ADDRESS, ADDRESS));
>>      }
>>
>>
>>      @Test
>> public void strlen() throws Throwable {
>> try (Arena arena = Arena.openConfined()) {
>> MemorySegment segment = arena.allocateUtf8String("Hello");
>> int length = (int) (long) strlen.invoke(segment);
>> Assertions.assertSame(5, length);
>>          }
>>      }
>>
>>      @Test
>> public void foo() throws Throwable {
>> String result = strcat("hello", "world");
>>      }
>>      /**
>>       * This is a one to one copy of strcat test taken from 
>>  StdLibTest.java test from OpenJDK
>>       */
>> String strcat(String s1, String s2) throws Throwable {
>> try (var arena = Arena.openConfined()) {
>> MemorySegment buf = arena.allocate(s1.length() + s2.length() + 1);
>> buf.setUtf8String(0, s1);
>> MemorySegment other = arena.allocateUtf8String(s2);
>> return ((MemorySegment)strcat.invokeExact(buf, other)).getUtf8String(0);
>>          }
>>      }
>> }
>>
>> On Tue, Jan 3, 2023 at 12:55 PM Maurizio Cimadamore 
>> <maurizio.cimadamore at oracle.com 
>> <mailto:maurizio.cimadamore at oracle.com>> wrote:
>>
>>     Hi Bernd,
>>     this mailing list truncates attachments. If you want us to look 
>> at some
>>     code you will have to paste it somewhere :-)
>>
>>     Cheers
>>     Maurizio
>>
>>     On 03/01/2023 17:14, Bernd Müller wrote:
>>      > Dear all,
>>      >
>>      > I try to train FFM. I even master quick sort but became 
>> desperate on
>>      > strcon.
>>      >
>>      > Attached you will find a Maven project with the strcon test 
>> copied one
>>      > to one from
>>      > OpenJDK 20 sources.
>>      >
>>      > "mvn test" results in
>>      >
>>      > [ERROR] Errors:
>>      > [ERROR]   StdLibTest.foo:49->strcat:60 » IndexOutOfBounds Out 
>> of bound
>>      > access on segment MemorySegment{ array: Optional.empty
>>      > address:139636573277136 limit: 0 }; new offset = 0; new length 
>> = 1
>>      >
>>      > I am running OpenJDK 64-Bit Server VM (build 20-ea+29-2280, mixed
>>      > mode, sharing) on Fedora 36.
>>      >
>>      > Please, can someone help ?
>>      >
>>      > Kind Regards,
>>      >
>>      > Bernd
>>      >
>>
>
>