U++ forum: Welcome to the forum

Search on this site

Search in forums

Home » U++ Library support » U++ Core » PROPOSAL: small / usefull Stream iface extension

Show: Today's Messages :: Show Polls :: Message Navigator
E-mail to friend

PROPOSAL: small / usefull Stream iface extension [message #27952]

Wed, 11 August 2010 23:39

kohait00
Messages: 939
Registered: July 2009
Location: Germany

Experienced Contributor

hi all,

when implementing own, nontrivial protocols using stream, one sometimes (like i am now) in trouble, needing to access the underlying buffer directly, without always triggering a complete copy.. and also beeing able to skip read/write things.
thus here some very small but usefull functions. it would be really cool to have them there. otherwise please comment why not and how could i circumvent this need, without copying over and over.. (imagine having to put packets there that have some data, over which to calculate crc..etc).

Stream.h:85

	byte *    Base()                 { return buffer; }
	byte *    Head()                 { return ptr; }
	void      SkipRead(dword size = 1) { ptr += min((dword)(uintptr_t)(rdlim - ptr), size); }
	void      SkipWrite(dword size = 1){ ptr += min((dword)(uintptr_t)(wrlim - ptr), size); }

Head is the most important for me..
SkipRead makes sense as well (ignoring stuff, without need of dummy-copying just to advance ptr)
SkipWrite is kind of just for symetrical completeness, but might be usefull somewhere..

[Updated on: Wed, 11 August 2010 23:47]

Report message to a moderator

Re: PROPOSAL: small / usefull Stream iface extension [message #27953 is a reply to message #27952]

Thu, 12 August 2010 00:14

kohait00
Messages: 939
Registered: July 2009
Location: Germany

Experienced Contributor

and some 2 more..


	void      PutReserve(byte *& p, dword size = 1) { if(ptr + size <= wrlim) { p = ptr; ptr += size; } else p = NULL; }
	dword     GetReserve(byte *& p, dword size = 1) { if(ptr + size <= rdlim) { p = ptr; ptr += size; return size; } 
														else { dword s = dword((uintptr_t)(rdlim - ptr)); return ((s>0)?(p=ptr,s):(0)); } }

could maybe be named PutShadow, GetShadow, or something, to indicate that the buffer is refered only, no memcpy..
this is usefull to modify the data underneeth directly.

Report message to a moderator

Re: PROPOSAL: small / usefull Stream iface extension [message #27961 is a reply to message #27953]

Thu, 12 August 2010 08:36

kohait00
Messages: 939
Registered: July 2009
Location: Germany

Experienced Contributor

the use case of this is the following:

when coding or especially decoding some nontrivial protocols from a Stream (beeing serialized by another device actually, coming in serially) one needs to 'interpret' the data being made available in the stream, instead of consuming it. only if the paket is 'valid' in total, it may be consumed. if not, one may drop one byte and start 'interprete' again. if not all data is available yet one may postpone processing..without copying over and over again..

all this is not possible without accessing the data directly. otherwise a rebuffering would be nesseccary, which is not very well.

trying to avoid copying / reallocating stuff..(will probably run on an embedded system, whithout MPU / MMU (blackfin), which doesnt like a lot of new / delete)

Report message to a moderator

Re: PROPOSAL: small / usefull Stream iface extension [message #28002 is a reply to message #27961]

Fri, 13 August 2010 09:57

mirek
Messages: 14105
Registered: November 2005

Ultimate Member

Sorry, but this is poorly defined...

IMO, you should trust more to Seek. I believe that if you reimplement your code with Seek and regular stream interface, I believe it will be as effective as using proposed methods. Well, maybe just almost as effective..

Well, MAYBE we could add something like:

const byte *Peek(int size)

which would return NULL if size is not available...

In any case, SkipRead and SkipWrite do not offer any advantage over Seek IMO.

Report message to a moderator

Re: PROPOSAL: small / usefull Stream iface extension [message #28004 is a reply to message #28002]

Fri, 13 August 2010 10:07

kohait00
Messages: 939
Registered: July 2009
Location: Germany

Experienced Contributor

Quote:

Sorry, but this is poorly defined...

thats why i'm asking people who know better Smile

would Peek return the underlying buffer? at ptr position?
then i'm fine with it and can tweak around with Seek ofcorse. indeed, Peek is quite a usefull function, anyway. in fact i prefer it, it keeps the API understandable, not too implementation bound, but usage bound. good idea.

SkipRead / SkipWrite are only helpers, and are not that much nessesary actually, neither my second post. one could would workaround it sth like

Seek(GetPos() + count);

Peek would need to deferentiate wrlim and rdlim right?
why const btw?

const byte * Peek(int size);

my point is solely driven by avoiding unwanted memcpy, thats why accessing buffer directly to perform tweaked read /write (some special communication protocol)

[Updated on: Fri, 13 August 2010 10:16]

Report message to a moderator

Re: PROPOSAL: small / usefull Stream iface extension [message #28006 is a reply to message #28004]

Fri, 13 August 2010 10:23

kohait00
Messages: 939
Registered: July 2009
Location: Germany

Experienced Contributor

another proposal, maybe this one's better...

	const byte * Peek(int size = 1)  { if(size <= 0) return NULL; if(ptr + size <= rdlim) return ptr; return NULL; }
	byte *    Head(int size = 1)     { if(size <= 0) return NULL; if(ptr + size <= wrlim) return ptr; return NULL; }

Report message to a moderator

Re: PROPOSAL: small / usefull Stream iface extension [message #28008 is a reply to message #28004]

Fri, 13 August 2010 10:29

mirek
Messages: 14105
Registered: November 2005

Ultimate Member

kohait00 wrote on Fri, 13 August 2010 04:07

Quote:

Sorry, but this is poorly defined...

thats why i'm asking people who know better Smile

Seek(GetPos() + count);

Which is in the interface as SeekCur.

Quote:

Peek would need to deferentiate wrlim and rdlim right?
why const btw?

const byte * Peek(int size);

Sure, only read variant.

Report message to a moderator

Re: PROPOSAL: small / usefull Stream iface extension [message #28009 is a reply to message #28006]

Fri, 13 August 2010 10:30

mirek
Messages: 14105
Registered: November 2005

Ultimate Member

kohait00 wrote on Fri, 13 August 2010 04:23

another proposal, maybe this one's better...

	const byte * Peek(int size = 1)  { if(size <= 0) return NULL; if(ptr + size <= rdlim) return ptr; return NULL; }
	byte *    Head(int size = 1)     { if(size <= 0) return NULL; if(ptr + size <= wrlim) return ptr; return NULL; }

For some reasons, write variant sounds somewhat dangerous to me...

Have to think about this one...

Well, in any case, it should advance the ptr...

[Updated on: Fri, 13 August 2010 10:31]

Report message to a moderator

Re: PROPOSAL: small / usefull Stream iface extension [message #28011 is a reply to message #28009]

Fri, 13 August 2010 10:46

mirek
Messages: 14105
Registered: November 2005

Ultimate Member

OK, here is what we got:

	const byte *Peek(int size = 1)   { ASSERT(size > 0); return ptr + size <= rdlim ? ptr : NULL; }
	byte       *PutPtr(int size = 1) { ASSERT(size > 0); if(ptr + size <= wrlim) { byte *p = ptr; ptr += size; return p; }; return NULL; }

Report message to a moderator

Re: PROPOSAL: small / usefull Stream iface extension [message #28012 is a reply to message #28009]

Fri, 13 August 2010 10:47

kohait00
Messages: 939
Registered: July 2009
Location: Germany

Experienced Contributor

another one...

	const byte * Peek(int size = 1)  { if(size <= 0 || (ptr + size > rdlim)) return NULL; return ptr; }
	byte *    PutReserve(int size)   { if(size <= 0 || (ptr + size > wrlim)) return NULL; byte * p = ptr; ptr+= size; return p; }

concerning Head():
if anyone really wants to break things its even possible with Peek then (brutal cast away const). those who really use these options are aware of it and use it with caution.

Peek shouldnt advance ptr. Peek only peeks Smile

advancing if needed should be done with Seek(GetPos() + size) afterwards, if desired.

advancing ptr in Head is a protection though, that same buffer section is not returned twice.. but the name is maybe irritating.

[Updated on: Fri, 13 August 2010 10:49]

Report message to a moderator