Problem with 6.3 and Cygwin

[tbutler at caf dot net]

I've built Insight for the arm-elf target, using the sources from the CVS
tree as of 7-Mar-2005. My host environment is WinXP with Cygwin, using the
latest packages as of 8-March-2005.

First of all, I noticed that the Target Settings menu item didn't work.
Based on a search of the archive, I confirmed that this was caused by
problems with the images/ and images2/ directories. Overwriting these with
an older copy corrects the problem (both 6.1 and 5.3 vintages seem to be
OK). As noted in other posts, it appears that some of the .gif files are
simply corrupt; viewing them with Windows Explorer reveals several that are
undisplayable.

Having overcome that hurdle, I found that Insight crashes when I do just
about anything (eg. open a symbol file, connect to target). Taking the
advice of an earlier archive message, I opened the console to launch a
debug window ("Tk ManagedWin::open DebugWin"), but this crashed upon
entering the command. I tried a simple "help" in the console with the same
results. I don't think this is a core GDB problem, because if I run Insight
with --interpreter=console, I can perform these actions successfully.

Debugging the simple "help" scenario reveals the following trace:

(gdb) run
Starting program: /gnutools/bin/arm-elf-insight.exe

Program received signal SIGSEGV, Segmentation fault.
do_my_cleanups (pmy_chain=0x6865b8, old_chain=0x1061c6e0)
    at /Insight_2005-07-Mar/src/gdb/utils.c:351
351           *pmy_chain = ptr->next;   /* Do this first incase recursion
*/
(gdb) bt
#0  do_my_cleanups (pmy_chain=0x6865b8, old_chain=0x1061c6e0)
    at /Insight_2005-07-Mar/src/gdb/utils.c:351
#1  0x00409329 in do_cleanups (old_chain=0x1061c6e0)
    at /Insight_2005-07-Mar/src/gdb/utils.c:317
#2  0x0040b3c8 in vfprintf_filtered (stream=0x1024af60,
    format=0x60a290 "List of classes of %scommands:\n\n",
    args=0x22dac8 "ÔÚ\"") at /Insight_2005-07-Mar/src/gdb/utils.c:2159
#3  0x0040b4af in fprintf_filtered (stream=0x1024af60,
    format=0x60a290 "List of classes of %scommands:\n\n")
    at /Insight_2005-07-Mar/src/gdb/utils.c:2191
#4  0x0042798b in help_list (list=0x10226278, cmdtype=0x609fd0 "",
    class=all_classes, stream=0x1024af60)
    at /Insight_2005-07-Mar/src/gdb/cli/cli-decode.c:837
#5  0x00427919 in help_cmd (command=0x0, stream=0x1024af60)
    at /Insight_2005-07-Mar/src/gdb/cli/cli-decode.c:744
[ snip ]

The 'pmy_chain' argument which do_my_cleanups() receives points to the
value zero, which ultimately leads to a null pointer dereference. It is
actually the address of the global variable 'cleanup_chain', which I found
gets overwritten with zero during the execution of vfprintf_filtered() (I'm
assuming that it shouldn't be 0). I followed the flow a bit, and found that
control arrives at gdbtk_fputs(), and the actual overwriting of the occurs
deep in the resulting flow of Tcl/Tk processing.

Any suggestions as to what might be wrong, or where to investigate further,
would be greatly appreciated.

Regards,
Tim