SUGGESTED FIX
make/java/fdlibm>sccs sccsdiff -r1.21 -r1.22 Makefile
------- Makefile -------
57,58c57,58
< # Override optimization primarily for Linux-i386-egcs-1.1.1 to get
< # correctness on Intel.
---
> # Disable optimization to get correctly reproducible
> # floating-point results.
61d60
< ifeq ($(ARCH),i586)
64d62
< endif
###@###.### 2003-10-31
|
EVALUATION
Will investigate; suspect change in gcc version or flags on Linux itanium is the cause.
###@###.### 2003-06-18
Similar failures reported in 4910006; suspect common cause.
###@###.### 2003-08-26
See analysis in 4910006.
###@###.### 2003-08-28
As discussed in detail in the comments for 4910006 this does *not* appear to be a C compiler flag issue. When I linked the generated *.o files with a C program, they returned the proper results.
###@###.### 2003-09-05
This may be related to when we shifted OS to build on from 7.2 to Redhat AS 2.1.. Not sure of the date/build integration that we switched to the new OS to build on..
###@###.### 2003-09-08
I've tested this on Redhat 7.2 and it reproduces the same crash.. I've tested
B06 and all is well. B07 was a dead JVM build. However JDK did get updated.
I've taken a b06 JVM and plugged into a B07 JDK and test fails. Thus looking like libraries issue change and not JVM..
###@###.### 2003-09-09
I tested a B08 JVM and plugged into a B06 JDK and the test passed. This is not a VM issue..
###@###.### 2003-09-09
Since the generated object files produce the correct result when used in a C program, the problem is neither with the object files nor the underlying source code. This implicatese the jni layer.
###@###.### 2003-10-24
As it turns out, it was just a compile flag problem after all.
###@###.### 2003-11-03
|
Java