further notice.
Looks like a bug in the lynx2.2.2 kernel scroggs the stack pointer in
these tests leading to massive failures. I've sent a nice small concise
example to Mark Rome @ lynx. Severe braindamage.
after the inferior gdb has started.
'cuz lynx may take a couple minutes to start the inferior gdb...
* gdb.base/a2-run.exp: Handle gdb echoing more than one cr-lf
after each command.
* gdb.base/funcargs.exp: Likewise.
* gdb.base/gdbvars.exp: Likewise.
* gdb.base/interrupt.exp: Likewise.
* gdb.base/list.exp: Likewise.
* gdb.base/watchpoint.exp: Likewise.
* gdb.c++/demangle.exp: Likewise.
* gdb.c++/inherit.exp: Likewise.
* gdb.chill/chexp.exp: Likewise.
* gdb.fortran/exprs.exp: Likewise.
Just in case we can't get a version of expect that sets the tty modes
right on lynx to avoid this problem.
targets except some specific ones that are known to not support signals.
Also clean up formatting.
* gdb.base/Makefile.in (EXECUTABLES): Remove sigall and signals.
(signals, sigall): Remove rules to build.
* gdb.base/{sigall.c, sigall.exp, signals.c, signals.exp}:
Moved to new gdb.signals directory.
* gdb.signals: New test directory.
* gdb.signals/{Makefile.in, configure.in, sigall.c, sigall.exp,
signals.c signals.exp}: New or moved files.
config/rom68k.exp: New tcl glue for the appropriate monitors.
These all just end up calling monitor.exp.
* config/monitor.exp (gdb_target_monitor): Handle `Connection
refused' by retrying. Cleanup some timeout issues.
* (gdb_load): Reduce timeout. Cleanup some timeout issues.
* lib/gdb.exp (gdb_test gdb_test_exact): Upvar timeout so that
callers don't have to set/restore global timeout variable.
* The following set of changes centralizes management of the global
timeout variable. This way, it can be set in one target dependent
place instead of dozens of places scattered throughout the test suite.
If you need to lengthen a timeout, then you should either set timeout
in one of the config/{target}.exp files, or multiply it by a factor.
Setting it to an absolute value is always going to lose for some
targets.
* gdb.base/a1-selftest.exp (test_with_self): Only use local timeout.
* gdb.base/callfuncs.exp (do_function_calls): Don't set timeout.
* gdb.base/funcargs.exp: Don't set timeout.
* gdb.base/list.exp (test_forward_search): Only use local timeout.
* gdb.base/printcmds.exp (test_print_string_constants
test_print_array_constants): Don't set timeout.
* gdb.base/ptype.exp: Don't set timeout.
* gdb.base/recurse.exp: Don't set timeout.
* gdb.base/return.exp: Don't set timeout.
* gdb.base/watchpoint.exp: Don't set timeout.
* gdb.c++/classes.exp (do_tests): Don't set timeout.
* gdb.c++/virtfunc.exp (test_virtual_calls): Don't set timeout.
* TODO: Remove item about printing variables in nodebug.exp. Add
item about enabling tests which are only run on some targets.
* gdb.base/watchpoint.exp: Various cleanups (make each test pass or
fail, make test names unique, use new gdb_test convention
regarding pattern, use gdb_test more, etc.).
number (we are already testing that the correct source line text
gets printed).
* gdb.base/break.exp: Make one test if $usestubs. I'm not sure
that is what is intended, but something needed to be done to get
sunos4 native working again.
* gdb.c++/misc.cc (main): Fix typo (#iffef -> #ifdef).
* findvar.c: Move default def of CANNOT_STORE_REGISTER closer to
the beginning of the code.
* (write_register_gen): New routine. Analogous to
read_register_gen.
* (write_register_bytes): Another rewrite! Make it smarter about
not updating regs with the same value.
* monitor.c (printf_monitor readchar): Use stderr instead of
stdout to output debug info. Also cleanup readchar a little.
* (expect): Make sure that excessive responses are null
terminated.
* (monitor_open): Check for magic number in monitor_ops struct.
Allow multiple commands as init strings. Also, clear all
breakpoints.
* (monitor_resume monitor_wait): Send a command to dump all the
regs for those targets which don't do so when waking up after a
continue command.
* (monitor_wait): Handle excessive response output better.
* (monitor_write_memory): Use block fill, word, and long word
commands (if they exist) to write memory more efficiently.
* General cleanups to use flag bits instead of individual flag
words in monitor_ops struct.
* (monitor_command): Return output from command.
* (monitor_load_srec): Allocate buffer only once. Use alloca.
Wait for load response string instead of using a timeout to start
sending S-records. Fix bug where value of srec_frame shrinks. If
hashmark is set, print `-' for retransmissions. General cleanups.
* (monitor_make_srec): Get rid of S-record default type kludge.
* monitor.h: Use seperate struct for memory and register
read/write commands. Memory commands can come in byte, word,
long, and longlong forms.
* (monitor_ops): Change lots of fields. Generalize some stuff.
Put all flags into flags word. Allow init to be a list of commands.
Add command for clearing all breakpoints, block fill, dumping all
registers.
* remote-est.c: Rewrite to use new monitor conventions.
* config/m68k/est.mt (TDEPFILES): Add monitor.o.
* config/m68k/tm-est.h: Set NUM_REGS to 18.
* testsuite/gdb.base/break.exp: Lots of cleanups. Use gdb_test
more thoroughly.
runto_main is a proc in gdb.exp that will do 'step' for target use
stubs.
* gdb.base/exprs.exp: ditto.
* gdb.base/interrupt.exp: ditto.
* gdb.base/opaque.exp: ditto.
* gdb.base/printcmds.exp: ditto.
* gdb.base/ptype.exp: ditto.
* gdb.base/scope.exp: ditto.
* gdb.base/setvar.exp: ditto.
* gdb.base/signals.exp: ditto.
* gdb.base/twice.exp: ditto.
* gdb.base/break.exp: no run and hit main for stubs. Change line
numbers for breakpoints and info breakpoint.
* gdb.base/break.c: Add #ifdef usestubs for set_debug_traps() and
breakpoint().
* gdb.base/callfuncs.c: ditto.
* gdb.base/exprs: ditto.
* gdb.base/interrupt.c: ditto.
* gdb.base/opaque0.c: ditto.
* gdb.base/printcmds.c: ditto.
* gdb.base/ptype.c: ditto.
* gdb.base/scope0.c: ditto.
* gdb.base/setvar.c: ditto.
* gdb.base/signals.c: ditto.
* gdb.base/twice.c: ditto.
somewhere between `read' and `main.c'. I'm pretty sure the
pattern ".*#\[0-9\].*" was slowing down pattern matching a lot,
and it isn't particularly useful.
* TODO: Remove "handle all print". Also remove item about
checking copyright date (I don't like the idea of a spurious FAIL
based on when we run the tests).
* gdb.base/a2-run.exp: Skip this file if noargs is set.
* gdb.base/break.c: Change things around so that this program
doesn't depend upon args. This is necessary to make remote
targets work (in general, they can't take args).
* gdb.base/break.exp: Don't try to send args to program. Don't
expect output. Also, replace lots of code with gdb_test.
only \r\n, not .*. The test can pass .* as the last thing in
$pattern if that is what it wants. In addition to providing this
flexibility, this change should speed up pattern matching in cases
where the pattern already ended with .* (there were a number of
them). This change also helps catch bad patterns--in the old
scheme the typo "char \*" instead of "char \\*" would pass. Now
it is caught.
* Many .exp files: Update callers.
"p '\'", which is the same as "p ''" once tcl gets done with
quoting, to "p '\\'", which I suspect is what is intended (one
backslash gets sent to GDB).
* gdb.base/printcmds.exp (test_artificial_arrays): New tests.
test, it causes stepping out of call_after_alloca_subr with
optimizing compilers.
(localvars_in_indirect_call): Consume GDB prompt if `finish'
fails.
* gdb.c++/misc.cc (class Contains_static_instance,
class Contains_nested_static_instance),
gdb.c++/classes.exp (test_static_members): Test printing of
a class that contains a static instance of the class.
remove them.
* gdb.base/nodebug.exp: Accept "function" in addition to "text
variable and "variable" in addition to "data variable".
* gdb.base/nodebug.exp: xfail datalocal and bsslocal tests for AIX.
restart GDB for h8300 after a continue.
(return): Add pattern to match a29k-udi behavior.
* gdb.base/nodebug.exp: Remove obsolete comment.
* gdb.base/regs.exp: New file, testing of register displays.
* gdb.base/return.exp: Fix typo in reference to $objdir.
* gdb.base/whatis.exp: Always start with a fresh GDB.
code following the catch silently ignored some errors; without the
catch they should be thrown like any other tcl error. Also, the
catch used "" instead of {} which meant all the callers who wanted
to include one of the characters "[]$ had to quote it an extra time.
* gdb.base/{callfuncs.exp,commands.exp,gdbvars.exp,printcmds.exp,
ptype.exp,signals.exp,watchpoint.exp}, gdb.c++/{cplusfuncs.exp,
demangle.exp}, gdb.chill/chexp.exp, gdb.fortran/exprs.exp:
Remove extra quoting.
gdb.c++/inherit.exp, gdb.c++/templates.exp, gdb.c++/virtfunc.exp,
gdb.fortran/exprs.exp, gdb.fortran/types.exp, gdb.chill/chexp.exp,
gdb.base/printcmds.exp: Remove passcount, failcount, etc., stuff;
it makes the tests harder to understand and confuses test-o-matic.
The preferred style is that each test provides a PASS or a FAIL,
and has a unique message (e.g. "continue #54" not just "continue")
which is the same for the pass and the fail.
* gdb.fortran/exprs.exp, gdb.fortran/types.exp,
gdb.chill/chexp.exp: Move test_print_accept and test_print_reject
to lib/gdb.exp.
* gdb.base/printcmds.exp: Use test_print_accept. Remove
prt_accept which was basically the same thing. Likewise for
test_print_reject and prt_reject.
* lib/gdb.exp (test_print_reject): Add some more error message
patterns to match from the former printcmds.exp (prt_reject).
* gdb.c++/classes.exp, gdb.base/scope.exp: Remove spurious xfails.
One defect of the passcount stuff is that some of it failed to
report XPASS where appropriate.
* gdb.c++/cplusfuncs.exp (print_addr_of): No longer accept extra
stuff before and after arg in expected pattern.
(test_paddr_operator_functions): Re-do test without print_addr_of;
this is the only test which seems to want extra stuff there.
compiler warning with dec alpha compiler.
* gdb.base/commands.exp (if_while_breakpoint_command_test): Remove
extra \n from gdb_test call (this explains why the prompt test
that I removed yesterday was there).
(user_defined_command_test): Remove extra \n and prompt test.
* gdb.base/sigall.c: If SIGABRT == SIGLOST, just act as if SIGLOST
doesn't exist. Likewise if SIGIO == SIGPOLL, pretend SIGPOLL
doesn't exist.
* gdb.base/sigall.exp: Remove special handling for SIGIO and SIGPOLL.
Also remove alpha workaround, which turns out to be because
SIGABRT == SIGLOST.
was a vestige of the manual TESTSTRATEGY, and the purpose of it
was to make sure that gdb can read a core file. But (a) the "read
the corefile" part didn't make it to a1-selftest.exp, we just
delete the corefile after creating it, and (b) we test reading
corefiles in corefile.exp anyway. Also, this test left around a
xgdb process on Solaris.
* gdb.base/corefile.exp: Also test backtrace.
when we'll be getting lots of output from gdb.
* gdb.base/a1-selftest.exp: Move code that was inside expect -re
"init_malloc" to after the expect statement. Run it if that
regexp is taken, but also for two others which happen if source
cannot be found.
* Makefile.in (site.exp): Don't insert host_os, etc.
(just-check): Just use RUNTEST with RUNTESTFLAGS only.
(distclean): Check existence of subdir.
* gdb.base/configure.in (hppa*-*-hpux*): Recognize.
* gdb.base/Makefile.in: General cleanup and simplification.
Use CC, CFLAGS, and LDFLAGS instead of ..._FOR_TARGET versions
to build executables. Use default rule to build .o files.
(EXPECT, RUNTEST, CC): Fix relative pathname.
(site.exp): Don't insert host_os, etc.
* gdb.base/signals.c (main): Don't call signal unless SIGALRM
and/or SIGUSR1 defined.
* gdb.base/watchpoint.c: Don't include stdio.h a second time.
* gdb.c++/Makefile.in: General cleanup and simplification.
Use CC, CFLAGS, and LDFLAGS instead of ..._FOR_TARGET versions
(CFLAGS): Remove.
(EXPECT, RUNTEST_FOR_TARGET, CXX): Fix relative pathname.
Jeff Law
Fred Fish
Stu Grossman
Jim Kingdon
Kung Hsu
Peter Schauer
David Edelsohn
Angela Marie Thomas
J.T. Conklin
Stan Shebs
Ian Lance Taylor