[PATCH 2/2] Use enums for human-readable exception information.

[Simon Marchi]

On 2020-01-17 10:31 a.m., Hannes Domani via gdb-patches wrote:
> Changes to $_siginfo type to this:
> 
> (gdb) pt $_siginfo
> type = struct EXCEPTION_RECORD {
>     enum ExceptionCode ExceptionCode;
>     unsigned int ExceptionFlags;
>     struct EXCEPTION_RECORD *ExceptionRecord;
>     void *ExceptionAddress;
>     unsigned int NumberParameters;
>     union {
>         void *ExceptionInformation[15];
>         struct {...} AccessViolationInformation;
>     };
> }
> (gdb) pt $_siginfo.ExceptionCode
> type = enum ExceptionCode {FATAL_APP_EXIT = 1073741845,
>     DBG_CONTROL_C = 1073807365, DBG_CONTROL_BREAK = 1073807368,
>     DATATYPE_MISALIGNMENT = 2147483650, BREAKPOINT, SINGLE_STEP,
>     ACCESS_VIOLATION = 3221225477, IN_PAGE_ERROR,
>     ILLEGAL_INSTRUCTION = 3221225501, NONCONTINUABLE_EXCEPTION = 3221225509,
>     INVALID_DISPOSITION, ARRAY_BOUNDS_EXCEEDED = 3221225612,
>     FLOAT_DENORMAL_OPERAND, FLOAT_DIVIDE_BY_ZERO, FLOAT_INEXACT_RESULT,
>     FLOAT_INVALID_OPERATION, FLOAT_OVERFLOW, FLOAT_STACK_CHECK,
>     FLOAT_UNDERFLOW, INTEGER_DIVIDE_BY_ZERO, INTEGER_OVERFLOW,
>     PRIV_INSTRUCTION, STACK_OVERFLOW = 3221225725, FAST_FAIL = 3221226505}
> (gdb) pt $_siginfo.AccessViolationInformation
> type = struct {
>     enum ViolationType Type;
>     void *Address;
> }
> (gdb) pt $_siginfo.AccessViolationInformation.Type
> type = enum ViolationType {READ_ACCESS_VIOLATION, WRITE_ACCESS_VIOLATION,
>     DATA_EXECUTION_PREVENTION_VIOLATION = 8}
> 
> Which makes it easier to understand the reason of the exception:
> 
> (gdb) p $_siginfo
> $1 = {
>   ExceptionCode = ACCESS_VIOLATION,
>   ExceptionFlags = 0,
>   ExceptionRecord = 0x0,
>   ExceptionAddress = 0x401632 <main+18>,
>   NumberParameters = 2,
>   {
>     ExceptionInformation = {0x1, 0x123, 0x0 <repeats 13 times>},
>     AccessViolationInformation = {
>       Type = WRITE_ACCESS_VIOLATION,
>       Address = 0x123
>     }
>   }
> }

Well, this kind of goes against the last comment I left on patch 1 (to use the
actual types found in the real structure), but I think that what you are
suggesting here is quite convenient, so I wouldn't be against doing something
like that.

Please add a bit of comments to your code to explain what you are doing.  Since
you are inventing types that don't exist in the documentation, it's not obvious
to know what this is doing.

Simon