|
Or a recursive CTE, but you can increase that I think.
|
|
|
|
|
I never make a large JOIN , I split it up:
A JOIN B JOIN C JOIN D
becomes:
( ( A JOIN B ) JOIN C ) JOIN D
|
|
|
|
|
Oh that's interesting! I'll have to try that!
|
|
|
|
|
For inner joins the optimizer will disregard any order of the joins, or any parentheses for that matter.
If you want to force a specific order there's a hint for that: OPTION (FORCE ORDER)
For outer joins this is not the case, they will happen in the relative order specified.
|
|
|
|
|
Question for you?
Can you verify this 1) uses less memory and/or 2) improves performance time?
Thanks! Craig
|
|
|
|
|
It has significantly reduced execution time in many cases.
Your mileage will vary.
|
|
|
|
|
This message is only shown to the SELECT few...
|
|
|
|
|
That's bang out of ORDER 
Life should not be a journey to the grave with the intention of arriving safely in a pretty and well-preserved body, but rather to skid in broadside in a cloud of smoke, thoroughly used up, totally worn out, and loudly proclaiming “Wow! What a Ride!" - Hunter S Thompson - RIP
|
|
|
|
|
What version of SQL was that?
What was the count of unions that caused this?
Thank you.
|
|
|
|
|
Which means "Stop doing this to us in here!" 
Just for more gig-gles, sent it over and over and see is the "rare event" message changes.
|
|
|
|
|
struct Record
{
unsigned short Length;
unsigned long ObjectId;
LevelType Level;
unsigned long SendBlock;
struct
{
unsigned long SendBlock : 31;
unsigned long SendBlockMarker : 1;
};
__time32_t SendTime;
struct
{
Type SendType : 16;
unsigned SendVersion : 16;
};
unsigned long SendNumber;
union
{
struct
{
char Message[1];
} Version0;
struct
{
__int64 EventTime;
char Message[1];
} Version1;
};
};
:face-palm:
Software Zen: delete this;
|
|
|
|
|
What's wrong isn't obvious to me. 
It looks like SendVersion tells you which union to look at.
If sent interprocessor, it will pack differently on 32-bit and 64-bit CPUs, ignoring endianism problems. The fact that there's a version number suggests that this could be the case.
|
|
|
|
|
There's a #pragma pack around the declaration.
I'll give you a couple hints: it's before SendVersion. This compiles successfully under VS2008 (don't ask), and I've not tried compiling it under VS2019. It should trigger a compile error.
Software Zen: delete this;
|
|
|
|
|
The two SendBlock fields? Maybe VS2008 accepted it because the second one was inside an unnamed struct.
|
|
|
|
|
BINGO!
Software Zen: delete this;
|
|
|
|
|
In short: everything!
Where should I start? The unsigned long of 1 bit? The SendBlock doubly defined? The version union with members in wrong order? The complete lack of comments?
I hope they pay you well to put up with this pile of manure 
Mircea
|
|
|
|
|
The union with members in the wrong order is likely correct. In Version1, EventTime was added to timestamp messages. It's followed by the Message contents, which are at the end of the message and claim to be of type char[1], but which will actually be indexed from 0 to Length - 1.
Writing types for message byte buckets is fun, especially when you add protocol version control!
|
|
|
|
|
Greg Utas wrote: type char[1], but which will actually be indexed from 0 to Length - 1
My instincts are telling me that it's a placeholder for variable-length data.
With MSVC you cannot use a zero-length array unless the following conditions are true:
1.) It must be the last field in the union or struct.
2.) Option /Za must be enabled.
3.) Option /permissive- must be unset/disabled
Or you can just do char[1] 
|
|
|
|
|
And then you receive a zero-length message (header only), and someone does a memset using the size of this type, trampling over the char[1] that isn't there. 
|
|
|
|
|
Heh,
I'm not advocating the technique, I am just pointing out what I see in the struct.
|
|
|
|
|
Randor wrote: My instincts are telling me that it's a placeholder for variable-length data.
Exactly right. The struct is actually header for the information content which follows.
Software Zen: delete this;
|
|
|
|
|
Mircea Neacsu wrote: The unsigned long of 1 bit?
An unsigned value 1 bit in width is slightly more useful than a signed value 1 bit in width, if you think about it.Mircea Neacsu wrote: The SendBlock doubly defined?
That's the error I was referring to. IMO the compiler should issue an error, since the identifier SendBlock is ambiguous when referencing fields in the struct. This code is compiled using VS2008 which does not issue a diagnostic.Mircea Neacsu wrote: The version union with members in wrong order?
They're in the correct order. The Message value must be at the end of the structure, since it is a placeholder for the beginning of the variable-length data that's included in the message.Mircea Neacsu wrote: The complete lack of comments?
This is a code fragment, extracted from a header file that provides more context.Mircea Neacsu wrote: hope they pay you well to put up with this pile of manure
They do pay me well, in the coin that really matters: appreciation. I've worked for the same company for over 30 years, and have survived numerous "workforce reduction" actions during the last period of financial instability.
As to its fecal quality, I didn't create this originally but I have maintained it over the last 20 years. This 'weird & wonderful' came up during some refactoring I'm doing while adding a new feature.
Software Zen: delete this;
|
|
|
|
|
Gary R. Wheeler wrote: An unsigned value 1 bit in width is slightly more useful than a signed value 1 bit in width, if you think about it.
Yes, but long... 
Gary R. Wheeler wrote: They do pay me well, in the coin that really matters: appreciation. I've worked for the same company for over 30 years,
I understand. I've worked for the same company for 22 years and enjoyed every minute of it... well, at least most minutes 
Mircea
|
|
|
|
|
Mircea Neacsu wrote: Yes, but long
In this environment, 32 bits. The original value was an unsigned long.
Software Zen: delete this;
|
|
|
|
|
Gary R. Wheeler wrote: struct
{
Type SendType : 16;
unsigned SendVersion : 16;
};
This appears to be a bit field shared between two different data types. I am assuming Type is an enumeration type which is a signed integer. The other 16 bits are unsigned.
|
|
|
|
General 
News 
Suggestion 
Question 
Bug 
Answer 
Joke 
Praise 
Rant 
Admin 
Submit an article or tip
