RE: Question: [PATCH] Change calculation of frame_id by amd64 epilogue unwinder

[Carl Love <cel@us.ibm.com>]

On Wed, 2022-11-02 at 08:36 -0700, Carl Love wrote:
> Bruno:
> > The idea of the patch looks fine, I just have a few style nits
> > inlined.
> > 
> > > ---
> > >   .../gdb.base/unwind-on-each-insn.exp          | 25
> > > +++++++++++++++++--
> > >   1 file changed, 23 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/gdb/testsuite/gdb.base/unwind-on-each-insn.exp
> > > b/gdb/testsuite/gdb.base/unwind-on-each-insn.exp
> > > index 3b48805cff8..c908a4b838e 100644
> > > --- a/gdb/testsuite/gdb.base/unwind-on-each-insn.exp
> > > +++ b/gdb/testsuite/gdb.base/unwind-on-each-insn.exp
> > > @@ -41,7 +41,6 @@ if ![runto_main] then {
> > >   proc get_sp_and_fba { testname } {
> > >       with_test_prefix "get \$sp and frame base $testname" {
> > >   	set sp [get_hexadecimal_valueof "\$sp" "*UNKNOWN*"]
> > > -
> > >   	set fba ""
> > >   	gdb_test_multiple "info frame" "" {
> > >   	    -re -wrap ".*Stack level ${::decimal}, frame at
> > > ($::hex):.*" {
> > > @@ -75,6 +74,23 @@ gdb_continue_to_breakpoint "enter foo"
> > >   
> > >   # Figure out the range of addresses covered by this function.
> > >   set last_addr_in_foo ""
> > > +
> > > +# The disassembly of foo on PowerPC looks like:
> > > +#     Dump of assembler code for function foo:
> > > +#  => 0x00000000100006dc <+0>:     std     r31,-8(r1)
> > > +#     0x00000000100006e0 <+4>:     stdu    r1,-48(r1)
> > > +#     0x00000000100006e4 <+8>:     mr      r31,r1
> > > +#     0x00000000100006e8 <+12>:    nop
> > > +#     0x00000000100006ec <+16>:    addi    r1,r31,48
> > > +#     0x00000000100006f0 <+20>:    ld      r31,-8(r1)
> > > +#     0x00000000100006f4 <+24>:    blr
> > > +#     0x00000000100006f8 <+28>:    .long 0x0
> > > +#     0x00000000100006fc <+32>:    .long 0x0
> > > +#     0x0000000010000700 <+36>:    .long 0x1000180
> > > +#     End of assembler dump.
> > > +#
> > > +# The last instruction in function foo is blr.  Ignore the .long
> > > +# entries following the blr instruction.
> > >   gdb_test_multiple "disassemble foo" "" {
> > >       -re "^disassemble foo\r\n" {
> > >   	exp_continue
> > > @@ -84,7 +100,11 @@ gdb_test_multiple "disassemble foo" "" {
> > >   	exp_continue
> > >       }
> > >   
> > > -    -re "^...($hex) \[^\r\n\]+\r\n" {
> > > +    -re "^...($hex) \[<>+0-
> > > 9:\s\t\]*\.long\[\s\t\]*\[^\r\n\]*\r\n" 
> > > {
> > 
> > I wonder if this pattern is unnecessarily strict. Correct me if
> > I'm 
> > wrong, but I think that no architecture has an instruction that
> > starts 
> > with a .  so this pattern could probably be simplified to
> > 
> > ^...($hex) \[<>+0-9:\s\t\]*\.[^\r\n\]*\r\n
> > And no other architectures would have a similar problem in the
> > future.
> > 
> > I'm not very knowledgeable on assembly so you may know better in
> > this
> > area.
> 
> I do not know of any architectures that have a dot at the beginning
> of
> an instruction name.  So yes the pattern could be changed as
> suggested.
> That will require changing the above comment as well to make it clear
> we are skipping any "instruction" that starts with a dot, such as
> .long.  

Well, I guess I should have tried the change before I sent the email! 
Using just a dot in the pattern doesn't work as the pattern matches on
the period in the line "End of assembler dump." resulting in a failure
to find the statement for the end of the dump.  Argh!  I will leave the
pattern as it was to check for ".long".  

                    Carl