Java compiler lossage

[Tim Van Holder]

GCC Version: 2.95.2
System:      i386-pc-msdosdjgpp

Problem: Serious lossage in the Java compiler (and libgcj).

Issues:
* No proper DOS support. DOS has a case-insensitive filesystem, so for
    DOS platforms, stricmp/strnicmp should be used instead of strcmp/strncmp
    to check paths.

* gcj should not always pass '-g1' if no -g option is given; if gcc is set 
up
    to emit SDB (coff) debugging by default, the compiler will generate
    debugging info GAS (including recent snapshots) can't grok.

   Example: gnu/gcj/convert/BytesToUnicode.java
   Command line: gcj -gcoff1 -c BytesToUnicode.java
   Output and corresponding lines:
----
   1188: Error: Rest of line ignored. First ignored character is `['.
->	.def	_byte[];	.scl	10;	.type	010;	.size	12;	.endef
   1191: Error: Rest of line ignored. First ignored character is `['.
->	.def	.eos;	.val	12;	.scl	102;	.tag	_byte[];	.size	12;	.endef
   1194: Error: Rest of line ignored. First ignored character is `['.
->	.def	_inbuffer;	.val	8;	.scl	8;	.tag	_byte[];	.size	12;	.type	030;	.endef
----

   Solution: default to -gstabs+1 if stabs is supported, or fix GAS so it
              accepts square brackets.

* ResourceBundle.java (part of libgcj) wouldn't compile if -O is given; with 
-O1
    it generated an internal compiler error, and with any other level it
    froze (which on my system often indicates a stack overflow).
   Now I have recompiled the GCC suite using a recent snapshot of binutils
    (with support for weak symbols), using -Os will cause jc1 to freeze and
    using any numbered -O option will always generate a SIGSEGV:
----
ResourceBundle.java: In class `java.util.ResourceBundle':
ResourceBundle.java: In method 
`trySomeGetBundle(java.lang.String,java.lang.String)':
ResourceBundle.java:142: Exiting due to signal SIGSEGV
----

* BigInteger.java (part of libgcj) can't generate a class file:
----
BigInteger.java: In class `java.math.BigInteger':
BigInteger.java: In method `isNegative()':
BigInteger.java:283: internal error - SP mismatch
----

   However if this:
----
  private final boolean isNegative()
  {
    return (words == null ? ival : words[ival - 1]) < 0;
  }
----
   is replaced by this less efficient, but equivalent code:
----
  private final boolean isNegative()
  {
    if(words == null) return (ival < 0);
    else return (words[ival - 1] < 0);
  }
----
   it works just fine.
   <This bug has also been submitted to the libjava bug list>

* If libgcj.a is built with optimization, gij and jv-convert won't link:
   Affected files:
    java/lang/Class.java
     -> if -O3 or higher is used, hackTrampoline and checkMemberAccess are 
no
        longer present in the object file.
     Probable cause: -O3 turns on -finline-functions, and the Java compiler
      incorrectly inlines them
    java/util/GregorianCalendar.java
    java/util/TimeZone.java
    java/util/zip/ZipEntry.java
     -> These miss LJv0, LJv1, ... entries if optimizing; -O0 is required.
   <This bug has also been submitted to the libjava bug list>


________________________________________________________________________
Get Your Private, Free E-mail from MSN Hotmail at http://www.hotmail.com