|
I guess this was used to circumvent the restriction that a switch statement can't be used on unmanaged strings.
|
|
|
|
|
A very smart way of circumventing the restrictions of C--!
|
|
|
|
|
I did similar in C about twenty years ago, but I could only fit two characters into a character literal.
If you'll pardon my rusty C:
char* s ;
// read a line from a file into s
short* i = s ;
switch ( *i ) ...
This enabled me to easily parse some reports. I wish I still had the code. 
|
|
|
|
|
PIEBALDconsult wrote: I did similar in C about twenty years ago, but I could only fit two characters into a character literal.
If you'll pardon my rusty C:
char* s ;
// read a line from a file into s
short* i = s ;
switch ( *i ) ...
On most hardware platforms, this will cause some alignment indigestion for every second pair of chars. Assigning a short off of an even byte boundary will usually cause a lot of work for the processor, perhaps involving exceptions and retries.
I like the thinking though...
--
Harvey
|
|
|
|
|
I only did it with the first two characters per line.
|
|
|
|
|
For a string, that could still be at an odd address.
--
Harvey
|
|
|
|
|
It's ugly, yes.
But it also has the potential to be fast. By loading the three characters and one space into a machine word, it is a single machine instruction to compare each value, and if the compiler spots that - and it damn well should - then it can use a machine-word look up table and index into a machine-word jump table. And you can't get much faster than that!
If this is in time-critical code, then it could be justified. Provided it was well commented. If it's in user input handling, then the author needs to meet Mr Firing Squad for being too "clever". 
The universe is composed of electrons, neutrons, protons and......morons. (ThePhantomUpvoter)
|
|
|
|
|
It's in a thread where it's listening for something from a TCPIP connection (sort of a command channel). It's low-volume, and as such, I don't think the "cleverness" is warranted. And no, it's not commented at all.
It's not that I don't understand what he was doing. I just think he was being too clever for his own good. And in the (unlikely?) event that the code is ever ported to a big-endian system, it would be completely broken.
WE ARE DYSLEXIC OF BORG. Refutance is systile. Your a$$ will be laminated.
There are 10 kinds of people in the world: People who know binary and people who don't.
|
|
|
|
|
Perhaps not a meeting with Mr Firing Squad than, but a meeting with Mr Big Stick seems warranted.
The universe is composed of electrons, neutrons, protons and......morons. (ThePhantomUpvoter)
|
|
|
|
|
I don't think it would, as it's doing the + and shift manually. If it was just switch(recv[0]) then it would be endian vulnerable.
In the absence of language or framework features that allow you to switch on a string, I have no problem with this.
|
|
|
|
|
You've missed the point ... all multibyte char literals are endian-sensitive.
To make it endian-insensitive, it should be
#define VB(a,b,c) (((((a)<<8)+(b))<<8)+(c))
verb = VB(upper[curr->recv[0]], upper[curr->recv[1]], upper[curr->recv[2]]);
switch (verb)
{
case VB('C','M','D'):
...
|
|
|
|
|
Oh really? So in that case doing switch(*recv) should be safe.
|
|
|
|
|
I have no idea which of the things I wrote are the subject of your "oh really", but all of them are correct. As for the rest, it's a non sequitur. Both *recv and multi-char literals are endian-sensitive, so you appear to have committed a fallacy of affirmation of the consequent ... a rather basic failure of logic. But if you want to go around switching on 4-char literals thinking that it's portable, be my guest ... just don't do it in any code that might affect my life.
|
|
|
|
|
You can use big words all you like, but either 'CMD ' is endian sensitive, i.e. it is the byte stream 'C', 'M', 'D', ' ' on one system and ' ', 'D', 'M', 'C' on another, or it isn't, i.e. it's always 'C', 'M', 'D', ' ' and therefore maps to a different integer.
In the first case, *recv won't be correct because the byte stream you're checking for is always the same, but the code in the original example would be, because it manually makes the integer in the big-endian manner, and that will be the value that the constant has if it switches the byte order. And if not, *recv will be correct, even if the integer interpretation of that value will be different.
There is one good argument you could have deployed, which is that the standard doesn't actually define whether a multi character constant refers to byte order or integer value. But if it has a consistent meaning in real world compilers, that doesn't really matter.
|
|
|
|
|
"You can use big words all you like"
Try "git". That's a little word.
|
|
|
|
|
You're not proposing he should use VB are you?
|
|
|
|
|
jibalt wrote: You've missed the point ... all multibyte char literals are endian-sensitive.
To make it endian-insensitive, it should be
That however ignores the point that the code was written to support a specific protocol over TCP.
If the character set of the protocol changed then that would be just one thing that would likely break.
|
|
|
|
|
No, I did not ignore your irrelevant point.
|
|
|
|
|
jibalt wrote:
No, I did not ignore your irrelevant point.
Your comments are only relevent if the character set changes - and thus the protocol changed.
Thus your comment, in terms of the original question, is irrelevant.
And I seriously doubt it would ever be relevant because it presumes that the protocol is compiler dependent. Even if the protocol did specially use a multi-byte character set which varied by OS then then the correct thing for the protocol to do is send a precursor to determine the ordering. Your code presumes and ordering and ignores the possibility that the other end is running on a different OS.
|
|
|
|
|
Was that meant to be English?
|
|
|
|
|
A wee bit of commmenting would be nice, but other than that I cannot see a problem here.
As for porting to big-endian system: First of all, that will never happen. Secondly, there will be more places to fix if it did happen. Thirdly, the code is platform-restricted anyway (to Windows), so there is not a problem...
|
|
|
|
|
Why is that code platform-restricted to Windows?
I've seen DWORDs on many platforms, so there's really nothing here to suggest that.
|
|
|
|
|
Ah, but it was Windows that started it. Nyah, nyah... 
As for the two other points, they are still valid...
|
|
|
|
|
Tom Delany wrote: event that the code is ever ported to a big-endian system, it would be completely broken.
No it wouldn't.
The code POSTED would work regardless of that.
As mentioned it is a TCP stream so it sequential. The switch statement is using ascii. The only way endianess would matter would be if the stream started to use a different character set. And if it did that it would fail if
1. The character system was not using a 8 bit lower representation of that matched ascii (UTF8 does where UTF16 does not.)
2. AND if the protocol changed.
And certainly if item 2 is true then all sorts of problems could result. Such as the rest of the code, following the switch, failing as well.
|
|
|
|
|
That's how you handle tags in ICC profiles, for example. It's a perfectly reasonable thing to do.
Tom Delany wrote: WE ARE DYSLEXIC OF BORG. Refutance is systile. Your a$$ will be laminated.
Still one of my favorite sigs on CP.
Software Zen: delete this;
|
|
|
|
General 
News 
Suggestion 
Question 
Bug 
Answer 
Joke 
Praise 
Rant 
Admin 
Submit an article or tip
