Another --disable-checking bootstrap failure

[Steven Bosscher]

On Sep 14, 2004 12:19 PM, Richard Kenner <kenner@vlsi1.ultra.nyu.edu>
wrote:

>     Is anyone seeing this on a --disable-checking bootstrap?
> 
>     (gdb) bt
>     #0  0x000000000090bba3 in lookup_page_table_entry (p=0xd3a170)
>         at bld/gcc/src.x86-64/gcc/ggc-page.c:581
>     #1  0x0000000000909b3c in ggc_set_mark (p=0xd3a170)
>         at bld/gcc/src.x86-64/gcc/ggc-page.c:1254
> #2 0x00000000006d7751 in gt_ggc_mx_basic_block_def (x_p=0x2a962032a0)
>         at gtype-desc.c:437
> #3 0x0000000000428e77 in gt_ggc_mx_lang_tree_node (x_p=0x2a96207c80)
>         at gt-c-decl.h:228
>     #4  0x00000000006d71fa in gt_ggc_mx_dataflow_d (x_p=0x2a9622da00)
>         at gtype-desc.c:288
>     #5  0x00000000006d7512 in gt_ggc_mx_tree_ann_d (x_p=0x2a96202690)
>         at gtype-desc.c:365
> 
> This looks similar to the SIGSEGV I get compiling the Ada rts module
> g-awk.adb with checking *enabled*.

Most likely another instance of PR17437.

What appears to be the problem is that somehow a dataflow object
for immediate uses is being collected while there is still a pointer
to it from somehwere.  Unfortunately, I can not reproduce the but
with an always-collect (--enable-checking=gcac) compiler, which is
really strange if this really is a dangling reference to a GC object.

Gr.
Steven