Build failure, HEAD, java/lang/Class.o

Build failure, HEAD, java/lang/Class.o

[Michael Chastain]

I haven't seen this build failure reported yet.

The new tree-ssa-loop-ivopts.c code is language-independent, but it uses
language-specific type nodes such as unsigned_type_node and
long_unsigned_type_node, which are not available in some languages such
as Java.

My system:

  native i686-pc-linux gnu, red hat 8, 128 megabytes of memory
  building gcc HEAD with gcc 3.3.4.

The build log says:

  /berman/fsf/_today_/berman/build/target/native/gcc/gcc-HEAD-as-2.15-ld-2.15/gcc/gcj -B/berman/fsf/_today_/berman/build/target/native/gcc/gcc-HEAD-as-2.15-ld-2.15/gcc/ -B/berman/fsf/_today_/berman/install/target/native/gcc/gcc-HEAD-as-2.15-ld-2.15/i686-pc-linux-gnu/bin/ -B/berman/fsf/_today_/berman/install/target/native/gcc/gcc-HEAD-as-2.15-ld-2.15/i686-pc-linux-gnu/lib/ -isystem /berman/fsf/_today_/berman/install/target/native/gcc/gcc-HEAD-as-2.15-ld-2.15/i686-pc-linux-gnu/include -isystem /berman/fsf/_today_/berman/install/target/native/gcc/gcc-HEAD-as-2.15-ld-2.15/i686-pc-linux-gnu/sys-include -ffloat-store -fno-omit-frame-pointer -fclasspath= -fbootclasspath=/berman/fsf/_today_/berman/build/target/native/gcc/gcc-HEAD-as-2.15-ld-2.15/i686-pc-linux-gnu/libjava --encoding=UTF-8 -Wno-deprecated -g -O2 -c /berman/fsf/_today_/source/gcc/HEAD/gcc/libjava/java/lang/Class.java -o java/lang/Class.o
  /berman/fsf/_today_/source/gcc/HEAD/gcc/libjava/java/lang/Class.java: In class `java.lang.Class':
  /berman/fsf/_today_/source/gcc/HEAD/gcc/libjava/java/lang/Class.java: In method `java.lang.Class.internalGetFields()':
  /berman/fsf/_today_/source/gcc/HEAD/gcc/libjava/java/lang/Class.java:416: internal compiler error: Segmentation fault
  Please submit a full bug report,
  with preprocessed source if appropriate.
  See <URL:http://gcc.gnu.org/bugs.html> for instructions.
  make[2]: *** [java/lang/Class.lo] Error 1
  make[2]: Leaving directory `/berman/fsf/_today_/berman/build/target/native/gcc/gcc-HEAD-as-2.15-ld-2.15/i686-pc-linux-gnu/libjava'
  make[1]: *** [all-recursive] Error 1
  make[1]: Leaving directory `/berman/fsf/_today_/berman/build/target/native/gcc/gcc-HEAD-as-2.15-ld-2.15/i686-pc-linux-gnu/libjava'
  make: *** [all-target-libjava] Error 2

This worked on:

  :pserver:anoncvs:@gcc.gnu.org:/cvs/gcc
  2004-08-31 16:33:12 UTC

And it fails with:

  :pserver:anoncvs:@gcc.gnu.org:/cvs/gcc
  2004-09-05 22:32:35 UTC

Here is a stack trace:

  #0  0x08237e17 in fold_convert (type=0x0, arg=0x401e46d8)
      at /berman/fsf/_today_/source/gcc/HEAD/gcc/gcc/fold-const.c:1847
  #1  0x08176831 in add_standard_iv_candidates (data=0xbffff330)
      at /berman/fsf/_today_/source/gcc/HEAD/gcc/gcc/tree-ssa-loop-ivopts.c:1563
  #2  0x081773b4 in find_iv_candidates (data=0xbffff330)
      at /berman/fsf/_today_/source/gcc/HEAD/gcc/gcc/tree-ssa-loop-ivopts.c:1757
  #3  0x0817df43 in tree_ssa_iv_optimize_loop (data=0xbffff330, loop=0x8667200)
      at /berman/fsf/_today_/source/gcc/HEAD/gcc/gcc/tree-ssa-loop-ivopts.c:4336
  #4  0x0817e0cc in tree_ssa_iv_optimize (loops=0x8658350)
      at /berman/fsf/_today_/source/gcc/HEAD/gcc/gcc/tree-ssa-loop-ivopts.c:4395
  #5  0x08163d5b in tree_ssa_loop_ivopts ()
      at /berman/fsf/_today_/source/gcc/HEAD/gcc/gcc/tree-ssa-loop.c:267
  #6  0x080fd1f2 in execute_one_pass (pass=0x855bd40)
      at /berman/fsf/_today_/source/gcc/HEAD/gcc/gcc/tree-optimize.c:458

The problem is in add_standard_iv_candidates:

  static void
  add_standard_iv_candidates (struct ivopts_data *data)
  {
    /* Add 0 + 1 * iteration candidate.  */
    add_candidate (data,
		   fold_convert (unsigned_type_node, integer_zero_node),
		   fold_convert (unsigned_type_node, integer_one_node),
		   true, NULL);

    /* The same for a long type.  */
    add_candidate (data,
		   fold_convert (long_unsigned_type_node, integer_zero_node),
		   fold_convert (long_unsigned_type_node, integer_one_node),
		   true, NULL);
  }

unsigned_type_node equals zero when this function is called.
integer_types is initialized in language-specific code.  For java, this
code is java_init_decl_processing in java/decl.c.  unsigned_type_node is
not used by Java.  That causes the seg fault in fold_convert.

  /* tree.h */
  #define unsigned_type_node		integer_types[itk_unsigned_int]

  (gdb) print integer_types
  $21 = {0x401e7e80, 0x0, 0x0, 0x0, 0x0, 0x401e73a0, 0x0, 0x0, 0x0, 0x0, 0x0}

  (gdb) print global_trees
  $22 = {0x401e24c0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0,
  0x401e46c0, 0x401e46d8, 0x401e4720, 0x401e4810, 0x401e40d8, 0x401e4150,
  0x401e45a0, 0x401e4768, 0x401e4138, 0x0, 0x0, 0x0, 0x401e4888, 0x401e48a0,
  0x0, 0x0, 0x0, 0x0, 0x401f3414, 0x401f34fc, 0x0, 0x0, 0x0, 0x0, 0x0,
  0x401e7c3c, 0x401e7d24, 0x0, 0x401e7910, 0x0, 0x0, 0x0, 0x401e7f68, 0x0,
  0x401f59d8, 0x0}

I don't quite know the right way to fix this.  I can file all this info
with bugzilla if you want.

Michael Chastain
:q

Re: Build failure, HEAD, java/lang/Class.o

[Andreas Jaeger]

[-- Attachment #1: Type: text/plain, Size: 263 bytes --]


Jan Hubicka and Steven Bosscher are looking into this AFAIK,

Andreas
-- 
 Andreas Jaeger, aj@suse.de, http://www.suse.de/~aj
  SUSE Linux AG, Maxfeldstr. 5, 90409 Nürnberg, Germany
   GPG fingerprint = 93A3 365E CE47 B889 DF7F  FED1 389A 563C C272 A126

[-- Attachment #2: Type: application/pgp-signature, Size: 188 bytes --]

Re: Build failure, HEAD, java/lang/Class.o

[Steven Bosscher]

On Sep 06, 2004 05:57 PM, Michael Chastain <mec.gnu@mindspring.com>
wrote:

> I haven't seen this build failure reported yet.
> 
> The new tree-ssa-loop-ivopts.c code is language-independent, but it
> uses
> language-specific type nodes such as unsigned_type_node and
> long_unsigned_type_node, which are not available in some languages
> such
> as Java.

Yup, it appears that we need some other magic there,
we're doing the following right now:

Index: tree-ssa-loop-ivopts.c 
 =================================================================== 
 RCS file: /cvs/gcc/gcc/gcc/tree-ssa-loop-ivopts.c,v 
 retrieving revision 2.2 
 diff -c -3 -p -r2.2 tree-ssa-loop-ivopts.c 
 *** tree-ssa-loop-ivopts.c 5 Sep 2004 15:24:15 -0000 2.2 
 --- tree-ssa-loop-ivopts.c 6 Sep 2004 15:25:59 -0000 
 *************** add_standard_iv_candidates (struct ivopt 
 *** 1561,1575 **** 
 { 
 /* Add 0 + 1 * iteration candidate. */ 
 add_candidate (data, 
 ! fold_convert (unsigned_type_node, integer_zero_node), 
 ! fold_convert (unsigned_type_node, integer_one_node), 
 true, NULL); 
 
 ! /* The same for a long type. */ 
 ! add_candidate (data, 
 ! fold_convert (long_unsigned_type_node, integer_zero_node), 
 ! fold_convert (long_unsigned_type_node, integer_one_node), 
 ! true, NULL); 
 } 
 
 
 --- 1561,1576 ---- 
 { 
 /* Add 0 + 1 * iteration candidate. */ 
 add_candidate (data, 
 ! build_int_cst (unsigned_intSI_type_node, 0), 
 ! build_int_cst (unsigned_intSI_type_node, 1), 
 true, NULL); 
 
 ! /* The same for a long type if it is still fast enought. */ 
 ! if (BITS_PER_WORD > 32) 
 ! add_candidate (data, 
 ! build_int_cst (unsigned_intDI_type_node, 0), 
 ! build_int_cst (unsigned_intDI_type_node, 1), 
 ! true, NULL); 
 } 
 
 
Only, that doesn't fix all the problems we've been running
into.  It is really odd, on i686 I had it bootstrapping,
and apparently so had a few other people, but on just about
every other target it's totally broken.

We're working overhours here to get the bootstrap going
again ;-)

Gr.
Steven