Re: PATCH: error reading variable: value has been optimized out

[Jan Kratochvil <jan dot kratochvil at redhat dot com>]

On Fri, 31 Aug 2012 12:32:32 +0200, Andrew Burgess wrote:
> Included a test case (below) that covers 4 different cases, all of which are
> currently broken in different ways. These are,

I would prefer to give arbitrary names suggesting what each test does, instead
of 1, 2, 3 and 4.  The same applies to the name of tests in resulting gdb.sum.

Also I had to create same map of the test for myself, it could be given there:

ABI registers order: rdi rsi rdx rcx r8 r9
==amd64_dummy_call_integer_regs

pre-set:   rdi rsi rdx rcx    rbx
undefined: rdi rsi     rcx
used:      rdi rsi rdx rcx r8 rbx

       operand0    operand1    operand2
test1: rcx         rbx         rsi
test2: rdx (+rcx)  rcx (+r8)   rsi (+rdx)
test3: rdx  +rcx   rcx  +rbx   rsi  +rdi
test4: rdx/4+rcx/4 rcx/4+rbx/4 rsi/4+rdi/4

This shows it tries to print also r8 but r8 is not initialized.

Also I would prefer to initialize each 32-bit part of each register by
a unique value, $0xdeadbeefdeadbeef everywhere may needlessly hide possible
ordering errors.  That means some 0xdeadbe00deadbe01, 0xdeadbe02deadbe03 or
anything like that.


> My original patch fixes (1) and (2),

In such case (3) and (4) should be KFAILed with PR number(s) filed and stated
there.


> though as Tom suggested we could do better in case (2) if we used computed
> values to describe partially optimized out values, but as gcc is no longer
> creating examples like this I think we can accept the more limited solution
> I originally suggessted, if that's ok.

Although formally there should be also filed PR and to KFAIL the case 2.

(Unfixed GDB would FAIL and not KFAIL.)



gdb_test "set print frame-arguments all" ""
->
gdb_test_no_output "set print frame-arguments all"

gdb_test "break breakpt" "Breakpoint.*at.*"
->
gdb_breakpoint "breakpt"


Thanks,
Jan