Re: promised trace flags change - a bit late

On 05/02/2011 12:34, Martin J. Evans wrote:
> On 05/02/2011 12:29, Tim Bunce wrote:
>> On Fri, Feb 04, 2011 at 08:00:13PM +0000, Martin J. Evans wrote:
>>> Tim,
>>>
>>> At the LPW I think I promised to add connection and encoding trace
>>> flags to DBI which add to the already existing SQL flag. I never got
>>> around to it and Merijn reminded me today on #dbi. The change is
>>> trivial and once implemented I will replace DBD::ODBC trace flags
>>> for connection and encoding with the standard DBI ones. There only
>>> remains 2 small issues wrt to this which need confirming:
>>>
>>> 1. DBI currently says
>>>      return 0x00000100 if $name eq 'SQL';
>>>
>>>     and as there is no macro for 0x00000100 I have the following in
>>> DBD::ODBC:
>>>
>>> /* Can't find a constant in DBI for SQL tracing but it is 256 */
>>> #define SQL_TRACE_FLAG 0x100
>>>
>>> and then code like this:
>>>
>>>     if (DBIc_TRACE(imp_dbh, SQL_TRACE_FLAG, 0, 3)) {
>>>         TRACE1(imp_dbh, "    SQLExecDirect %s\n", 
>>> SvPV_nolen(statement));
>>>     }
>>>
>>> Should there be macros for DBI's trace flags
>> Yes. I'd suggest
>>
>>      #define DBIf_TRACE_SQL   0x00000100
>>
> ok
>>> and if so where do they go (in DBIXS.h?)
>> Just above DBIc_TRACE_MATCHES.
>>
> ok
>>> 2. I propose adding connection and encoding flags to DBI - so far
>>> they've kept pretty short:
>>>
>>>    ALL
>>>    SQL
>>>
>>> so I've replicated this with:
>>>
>>>      sub parse_trace_flag {
>>>      my ($h, $name) = @_;
>>>      #      0xddDDDDrL (driver, DBI, reserved, Level)
>>>      return 0x00000100 if $name eq 'SQL';
>>>      return 0x00000200 if $name eq 'CON';
>>>      return 0x00000400 if $name eq 'ENC';
>>>      return;
>>>      }
>>>
>>> and:
>>>
>>> Currently the DBI only defines four trace flags:
>>>
>>>    ALL - turn on all DBI and driver flags (not recommended)
>>>    SQL - trace SQL statements executed
>>>          (not yet implemented in DBI but implemented in some DBDs)
>>>    CON - trace connection process
>>>          (not yet implemented in DBI but implemented in some DBDs)
>>>    ENC - trace encoding (unicode translations etc)
>>>          (not yet implemented in DBI but implemented in some DBDs)
>>>
>>> is this ok?
>> I'm not a huge fan of short names in general, but in this case I think
>> it's reasonable.
>>
> neither am I but on the DBI_TRACE=x|y|z line it is nicer for them to 
> be short.
>>> and 2 supplementals:
>>>
>>> 3. I could look into implementing some/all those flags in DBI but it
>>> seems these areas are pretty much covered with trace levels right
>>> now
>> I think it's worth adding CON. It would enable connect and disconnect to
>> be traced *with no other output*.
>>
> ok
>>> and in any case, I cannot find prepare in DBI (for SQL tracing)
>>> anyway.
>> The trace flag checking would need to be applied in XS_DBI_dispatch
>> and controlled by the 'Internal Method Attributes' (dbi_ima_st).
>>
>> I'm thinking it would work along these lines:
>>
>>      When a method with the CON trace flag is called and the current
>>      trace setting on the handle has the CON trace flag set, then the 
>> trace
>>      level will be set to min(trace_level,2) for the duration of the 
>> call.
>>
>> Sound ok?
>>
>> (I've partly implemented that but not tested it yet.)
>>
> this is an area I am not that familiar with - bit out of my depth here 
> at the moment. Similar issue with handle attributes and method 
> attributes discussion - I understood what you said but am struggling 
> to implement it.
>>> 4. Lastly (and not wanting to get in Merijn's way who agreed to look
>>> at it) but at the same workshop you discussed dbd_verbose (or
>>> whatever some DBDs have added support for) with Merijn and we think
>>> you said there is space in the trace settings for DBD trace
>>> settings. Apparently some DBDs only want DBD tracing and not DBI
>>> tracing so trace level is no good as it always includes DBI tracing.
>>> I think you were talking about reserving some space for DBDs only -
>>> perhaps you could remind us - the idea being settings (or a setting)
>>> which outputs DBD tracing but not DBI tracing.
>> As you note above, the current trace settings are encoded into an int:
>>
>>        #      0xddDDDDrL (driver, DBI, reserved, Level)
>>
>> Where L is the DBI trace level (0-16) and the rest are bit flags.
>> The 4 reserved bits could be used as a driver trace level (0-16).
>>
>> Tim.
>>
> I'll leave that bit for Merijn as this seemed to be something he 
> wanted/needed.
>
> I'll check in the changes I put forward in the next hour.
>
> Martin

ok, the 3 new trace flags are in and I added macros (slight shame you 
cannot use them in DBI.pm too but not a great deal). I've also 
implemented the new DBI trace flags in DBD::ODBC and am testing now. 
Because DBD::ODBC already had ENC and CON trace flags I've simply ored 
then in DBD::ODBC with DBI's new ones so you can use either for now then 
some time after DBI is released I will remove the DBD::ODBC ones.

I've also changed all DBD::ODBC tracing to look at DBIf_TRACE_DBD so you 
should be able to get only DBD tracing - still to test.

Martin